{"title":"GPUs \u0026 accelerators","description":"\u003cp\u003eGPUs, connectors, cables, and hardware components for mining rigs, AI servers, and high-performance computing builds.\u003c\/p\u003e","products":[{"product_id":"pcie-pci-express-16x-riser-card-flexible-cable-extension-cable","title":"PCIE PCI Express 16X Riser Card Flexible Cable Extension Cable","description":"\u003cul\u003e \u003cli\u003e\u003cspan\u003e PCI-E 16X slot soldered with normal IDE ribbon cable\u003c\/span\u003e\u003c\/li\u003e \u003cli\u003e\u003cspan\u003e Brand new and high quality\u003c\/span\u003e\u003c\/li\u003e \u003cli\u003e\u003cspan\u003e No Driver necessary and not support hot swappable\u003c\/span\u003e\u003c\/li\u003e \u003cli\u003e\u003cspan\u003e Total length: 19cm (7.4 inch)\u003c\/span\u003e\u003c\/li\u003e \u003c\/ul\u003e","brand":"Pcpraha","offers":[{"title":"Default Title","offer_id":49002883711304,"sku":null,"price":6.0,"currency_code":"EUR","in_stock":true}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/0843\/5479\/3800\/files\/pci-e-express-16x-to-16x-riser-card-adapter-flex-extension-cable-15cm-1.gif?v=1725528277"},{"product_id":"cpu-6pin-to-graphics-video-card-double-pci-e-pcie-power-supply-splitter-cable","title":"CPU 6Pin to Graphics Video Card Double PCI-E PCIe Power Supply Splitter Cable","description":"\u003cdiv\u003eConnect any graphics card to a single 6-pin PCI-Express Power Cable. The 6+2-pin male\u003c\/div\u003e\n\u003cdiv\u003econnectors offer maximum versatility.\u003c\/div\u003e\n\u003cdiv\u003eWarning: Do not overload your power supply or the 6-pin power cable, running more than 300W over a single 6-pin 18AWG PCI-Express cable is a fire hazard.\u003c\/div\u003e\n\u003cdiv\u003eConnectors: 6-pin PCIe Female, 2x 6+2-pin PCIe Male\u003c\/div\u003e\n\u003cdiv\u003eLength: 8 inches \/ 20cm\u003c\/div\u003e\n\u003cdiv\u003eWire: 18 AWG copper\u003c\/div\u003e\n\u003cdiv\u003ePackage listing :\u003c\/div\u003e\n\u003cdiv\u003e\u003c\/div\u003e\n\u003cdiv\u003e1x PCI-e 8pin to Dual 8Pin Video Card Power Extension Cable\u003c\/div\u003e","brand":"Pcpraha","offers":[{"title":"Default Title","offer_id":49002883744072,"sku":null,"price":15.0,"currency_code":"EUR","in_stock":true}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/0843\/5479\/3800\/files\/cpu-6pin-to-graphics-video-card-double-pci-e-pcie-power-supply-splitter-cable-3.jpg?v=1725528280"},{"product_id":"power-supply-connector-relay-adapter-dual-triple-quad-multi-psu-link-concurrent","title":"Power Supply Connector Relay Adapter Dual Triple Quad Multi PSU Link Concurrent","description":"\u003cstrong\u003eSpecification:\u003c\/strong\u003e \u003cul\u003e \u003cli\u003eBrand new\u003c\/li\u003e \u003cli\u003eSize(approx.):7.00*3.00cm\/2.76\"*1.18\"(L*W)\u003c\/li\u003e \u003cli\u003eQuantity:1Set\u003c\/li\u003e \u003cli\u003eSmall size - Can be installed anywhere\u003c\/li\u003e \u003cli\u003eDual triple Quad or Multiple PSU Power supply connector Relay Adapter link Concurrent\u003c\/li\u003e \u003cli\u003eFor your Bitcoin, Litecoin Minning machines with multiple power supply(PSU)\u003c\/li\u003e \u003cli\u003eMulti Power supply system suppoort (can connect Almost limitless)\u003c\/li\u003e \u003c\/ul\u003e \u003cstrong\u003ePackage Included:\u003c\/strong\u003e \u003cul\u003e \u003cli\u003e1 x Power Supply Adapter\u003c\/li\u003e \u003c\/ul\u003e","brand":"Pcpraha","offers":[{"title":"Default Title","offer_id":49002885218632,"sku":null,"price":15.0,"currency_code":"EUR","in_stock":true}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/0843\/5479\/3800\/files\/power-supply-connector-relay-adapter-dual-triple-quad-multi-psu-link-concurrent-2.jpg?v=1725528283"},{"product_id":"usb-3-0-60cm-pci-e-express-powered-riser-card-extender-cable-1xto16x","title":"USB 3.0 60CM PCI-E Express Powered Riser Card Extender Cable 1xto16x","description":"100% brand new and high quality Safer, more efficient, more practical The graphics card power supply adopts three FP solid-state capacitors, make graphics more stable power supply security. Parallel transfer tapes or big 4 pin interface to strengthen the power supply and make the video card power independence from motherboard Solve the problem that welding point fragile burn, transmission signal attenuation, platoon line length is not enough and so on Slots with fixed card buckle, remove the graphics card and the fixed graphics, the graphics card will not fall off from the slot stable and secure Specifications: Power Connector: Sata 15 pin power cable Compatible use: with any graphics cards Board Size:(L*W) 13.00*4.40cm\/5.12\"*1.73\" (Approx.) USB Cable Length: 62.00cm\/24.41\" USB length: 60cm \u003cstrong\u003ePackage Included: \u003c\/strong\u003e 1Pc x PCIE adapter 1Pc x PCIE riser board 1Pc x USB 3.0 cable 1Pc x 1X to 16X extension cord \u003cstrong\u003eNotes: \u003c\/strong\u003e Due to the difference between different monitors, the pictures may not reflect the actual color of the item. Compare the detail sizes with yours, please allow 1-3cm error, due to manual measurement. Please leaving a message before you give the bad feedback, if the products have some problems. Thanks for your understandings.","brand":"Pcpraha","offers":[{"title":"Default Title","offer_id":49002885251400,"sku":null,"price":18.0,"currency_code":"EUR","in_stock":true}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/0843\/5479\/3800\/files\/usb-3-0-60cm-pci-e-express-powered-riser-card-extender-cable-1xto16x-1.jpg?v=1725528285"},{"product_id":"hp-vga-nvidia-rtx-a4000-16gb-gddr6-pcie-4-0x16-card","title":"HP VGA NVIDIA RTX A4000 16GB GDDR6, PCIe 4.0×16 Card","description":"\u003ch2\u003eTechnical specifications:\u003c\/h2\u003e \u003cstrong\u003eChipset\u003c\/strong\u003e: NVIDIA RTX A4000 \u003cstrong\u003eMemory\u003c\/strong\u003e: 16 GB GDDR6 \u003cstrong\u003eMemory bus width\u003c\/strong\u003e: 256-bit \u003cstrong\u003eInterface\u003c\/strong\u003e: PCI Express Gen 4 x 16 \u003cstrong\u003eDirectX\u003c\/strong\u003e: DirectX 12.07 \u003cstrong\u003eOpen GL\u003c\/strong\u003e: 4.6 \u003cstrong\u003eVirtual reality\u003c\/strong\u003e: Yes \u003cstrong\u003eCooling\u003c\/strong\u003e: Active \u003cstrong\u003eDisplayPort\u003c\/strong\u003e: 4x DisplayPort 1.4 \u003cstrong\u003eExternal power supply\u003c\/strong\u003e: 1x 6-pin \u003cstrong\u003eMaximum power supply\u003c\/strong\u003e: 140 W \u003cstrong\u003eHP VGA NVIDIA RTX A4000 16GB GDDR6, PCIe 4.0×16 Card \u003c\/strong\u003e NVIDIA's Quadro products are a range of \u003cstrong\u003eprofessional\u003c\/strong\u003e graphics cards designed for both rackmount servers and, above all, high-performance workstations. These are high-quality GPUs with a large memory, ECC and a special set of drivers that are adapted for work in CAD applications. \u003cstrong\u003eCUDA technology\u003c\/strong\u003e Users of professional applications can use graphical CUDA stream processors thanks to the CUDA architecture. As a result, the gross performance of the graphics card can be used for specific calculations, which can speed up the work many times over using a conventional processor, which is significantly limited by a lower number of cores. \u003cstrong\u003eSpecifications:\u003c\/strong\u003e \u003cdiv\u003e \u003cdiv\u003e \u003cdiv\u003e \u003ctable\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd\u003e\n\u003cstrong\u003eC\u003c\/strong\u003e\u003cstrong\u003eUDA Cores\u003c\/strong\u003e\n\u003c\/td\u003e \u003ctd\u003e6144\u003c\/td\u003e \u003c\/tr\u003e \u003ctr\u003e \u003ctd\u003e\u003cstrong\u003eTensor Cores\u003c\/strong\u003e\u003c\/td\u003e \u003ctd\u003e192\u003c\/td\u003e \u003c\/tr\u003e \u003ctr\u003e \u003ctd\u003e\u003cstrong\u003eRT Cores\u003c\/strong\u003e\u003c\/td\u003e \u003ctd\u003e48\u003c\/td\u003e \u003c\/tr\u003e \u003ctr\u003e \u003ctd\u003e\u003cstrong\u003eSingle Precision Performance\u003c\/strong\u003e\u003c\/td\u003e \u003ctd\u003e19.2 TFLOPS\u003c\/td\u003e \u003c\/tr\u003e \u003ctr\u003e \u003ctd\u003e\u003cstrong\u003eRT Core Performance\u003c\/strong\u003e\u003c\/td\u003e \u003ctd\u003e37.4 TFLOPS\u003c\/td\u003e \u003c\/tr\u003e \u003ctr\u003e \u003ctd\u003e\u003cstrong\u003eTensor Performance\u003c\/strong\u003e\u003c\/td\u003e \u003ctd\u003e153.4 TFLOPS\u003c\/td\u003e \u003c\/tr\u003e \u003ctr\u003e \u003ctd\u003e\u003cstrong\u003eGPU Memory\u003c\/strong\u003e\u003c\/td\u003e \u003ctd\u003e16 GB GDDR6 with ECC\u003c\/td\u003e \u003c\/tr\u003e \u003ctr\u003e \u003ctd\u003e\u003cstrong\u003eMemory Interface\u003c\/strong\u003e\u003c\/td\u003e \u003ctd\u003e256-bit\u003c\/td\u003e \u003c\/tr\u003e \u003ctr\u003e \u003ctd\u003e\u003cstrong\u003eMemory Bandwidth\u003c\/strong\u003e\u003c\/td\u003e \u003ctd\u003e448 GB\/sec\u003c\/td\u003e \u003c\/tr\u003e \u003ctr\u003e \u003ctd\u003e\u003cstrong\u003eSystem Interface\u003c\/strong\u003e\u003c\/td\u003e \u003ctd\u003ePCI Express 4.0 x16\u003c\/td\u003e \u003c\/tr\u003e \u003ctr\u003e \u003ctd\u003e\u003cstrong\u003eDisplay Connectors\u003c\/strong\u003e\u003c\/td\u003e \u003ctd\u003e4x DisplayPort 1.4a\u003c\/td\u003e \u003c\/tr\u003e \u003ctr\u003e \u003ctd\u003e\u003cstrong\u003eMaximum Power Consumption\u003c\/strong\u003e\u003c\/td\u003e \u003ctd\u003e140 W\u003c\/td\u003e \u003c\/tr\u003e \u003c\/tbody\u003e \u003c\/table\u003e \u003c\/div\u003e \u003cstrong\u003eSupported technologies:\u003c\/strong\u003e NVIDIA NVLink NVIDIA Ampere NVIDIA Mosaic NVIDIA RTX Experience OpenGL 4.68 Shader Model 5.17 DirectX 12.07 Vulkan 1.2 HDCP 2.2 Max. resolution – 7680 x 4320 px DisplayPort 1.4a \u003c\/div\u003e \u003c\/div\u003e","brand":"HP","offers":[{"title":"Default Title","offer_id":49002885284168,"sku":null,"price":1436.49,"currency_code":"EUR","in_stock":true}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/0843\/5479\/3800\/files\/pi8-1511988-1559809-0a_-1_-1_78956.jpg?v=1725528288"},{"product_id":"nvidia-cmp-hx90-mining-gpu-97mh-s","title":"NVIDIA CMP HX90 – mining GPU – 97MH\/s","description":"\u003ch2\u003eNvidia CMP 90 HX90 – Jedná se o profesionální grafickou kartu od společnosti NVIDIA\u003c\/h2\u003e \u003cstrong\u003eParametry:\u003c\/strong\u003e This is a professional graphics card ✔ Ethereum Hash Rate(1) : 97-100 MH\/s ✔ Power connectors(2) : 2x 8-pin ✔ Memory size : 10GB ✔ Consumption : 210-250w The NVIDIA CMP HX90, built on an 8nm process and powered by a GA102 GPU, supports DirectX 12 Ultimate. It is a GPU GA102 large chip with a matrix area of ​​628 mm and 28,300 million transistors. \u003cimg src=\"https:\/\/pcpraha.cz\/wp-content\/uploads\/2020\/05\/hx90-mining-300x162.jpeg\"\u003e \u003ch2\u003eNVIDIA CMP HX90 designed for professionals\u003c\/h2\u003e \u003ch3\u003eOptimized for Crypto Mining\u003c\/h3\u003e The NVIDIA GPU architecture allows you to benefit more efficiently and return your mining investment faster. The 90HX platform is a proven solution for any mining system. \u003cdiv\u003e\u003c\/div\u003e \u003cdiv\u003e \u003cdiv\u003e \u003cdiv\u003e\u003c\/div\u003e \u003cdiv\u003e\u003cimg src=\"https:\/\/s.alicdn.com\/@sc04\/kf\/H3e2c04a9a7c641b7930e5beaa95681e5f\/258479783\/H3e2c04a9a7c641b7930e5beaa95681e5f.jpg?quality=close\"\u003e\u003c\/div\u003e \u003cdiv\u003e\u003cimg src=\"https:\/\/s.alicdn.com\/@sc04\/kf\/Hffcc26fb5098404096039dae450c61acE\/258479783\/Hffcc26fb5098404096039dae450c61acE.jpg?quality=close\"\u003e\u003c\/div\u003e \u003cdiv\u003e\u003cimg src=\"https:\/\/s.alicdn.com\/@sc04\/kf\/Hf00ba744821441a0aa169a87be2eb37a6\/258479783\/Hf00ba744821441a0aa169a87be2eb37a6.jpg?quality=close\"\u003e\u003c\/div\u003e \u003cdiv\u003e\u003cimg src=\"https:\/\/s.alicdn.com\/@sc04\/kf\/H6d68ceee975543ae85d84c7800d3fef1d\/258479783\/H6d68ceee975543ae85d84c7800d3fef1d.jpg?quality=close\"\u003e\u003c\/div\u003e \u003cdiv\u003e\u003cimg src=\"https:\/\/s.alicdn.com\/@sc04\/kf\/Hf66d11076496477ab85a3364217af883X\/258479783\/Hf66d11076496477ab85a3364217af883X.jpg?quality=close\"\u003e\u003c\/div\u003e \u003cdiv\u003e\u003cimg src=\"https:\/\/s.alicdn.com\/@sc04\/kf\/H439d76520a2b418b896b41fa99618e17b\/258479783\/H439d76520a2b418b896b41fa99618e17b.jpg?quality=close\"\u003e\u003c\/div\u003e \u003cdiv\u003e\u003cimg src=\"https:\/\/s.alicdn.com\/@sc04\/kf\/Hfe7dafb507994336bdb019f6d8f1aaabb\/258479783\/Hfe7dafb507994336bdb019f6d8f1aaabb.jpg?quality=close\"\u003e\u003c\/div\u003e \u003cdiv\u003eThe card also has 50 acceleration cores for ray tracing.\u003c\/div\u003e \u003c\/div\u003e \u003c\/div\u003e \u003cdiv\u003e \u003cdiv\u003e \u003cdiv\u003e \u003cdiv\u003e \u003cdiv\u003e \u003cdiv\u003e Unlike the fully unlocked GeForce RTX 3090 Ti, which uses the same GPU but supports all 10752 shaders, NVIDIA has disabled some shading units on the CMP 90HX to reach the target number of shaders. It has 6400 shading units, 200 texture overlay units and 80 ROPs. It also includes 200 tensor cores to help speed up machine learning applications. \u003cimg src=\"https:\/\/www.viperatech.com\/wp-content\/uploads\/2021\/09\/nvidia_cmp_1613726898268.webp\"\u003e The card also has 50 acceleration cores for ray tracing. NVIDIA has connected 10 GB of GDDR6X memory to the CMP 90HX, which must use a 320-bit memory interface. The GPU runs at 1500 MHz, which can be boosted up to 1710 MHz, the memory runs at 1188 MHz (19 Gb \/ s efficiently). As a two-slot card, the NVIDIA CMP 90HX is powered by two 8-pin power connectors with a maximum consumption of 320W. \u003cimg src=\"https:\/\/www.viperatech.com\/wp-content\/uploads\/2021\/09\/Nvidia-Ampere-Feature2.jpeg\"\u003e This device does not have display connectivity because it is not designed to connect monitors. The CMP 90HX connects to the rest of the system via the PCI-Express 4.0 x16 interface. The card is 285 mm long, 112 mm wide and is equipped with a two-slot cooling solution. with a maximum consumption of 320 watts. This device does not have display connectivity because it is not designed to connect monitors. The CMP 90HX connects to the rest of the system via the PCI-Express 4.0 x16 interface. The card is 285 mm long, 112 mm wide and is equipped with a two-slot cooling solution. with a maximum consumption of 320 watts. \u003c\/div\u003e \u003c\/div\u003e \u003c\/div\u003e \u003c\/div\u003e \u003c\/div\u003e \u003c\/div\u003e","brand":"nVidia","offers":[{"title":"Default Title","offer_id":49002885513544,"sku":null,"price":1029.0,"currency_code":"EUR","in_stock":true}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/0843\/5479\/3800\/files\/hx90-100mhs.jpg?v=1725528291"},{"product_id":"nvidia-l40","title":"nVidia L40","description":"\u003ch1\u003eNVIDIA L40 GPU - Powering Next-Gen AI and Graphics Workloads\u003c\/h1\u003e \u003ch2\u003eOverview\u003c\/h2\u003e \u003cp\u003eThe NVIDIA L40 GPU is a powerhouse designed for the most demanding AI, graphics, and compute workloads. With its exceptional performance and efficiency, the L40 is the ideal choice for complex simulations, large language model training, and high-fidelity rendering.\u003c\/p\u003e \u003ch2\u003eKey Features\u003c\/h2\u003e \u003cul\u003e \u003cli\u003e\n\u003cstrong\u003eAda Lovelace Architecture\u003c\/strong\u003e: Cutting-edge GPU design for unparalleled performance\u003c\/li\u003e \u003cli\u003e\n\u003cstrong\u003e48GB HBM3 Memory\u003c\/strong\u003e: Ultra-fast, high-bandwidth memory for handling massive datasets\u003c\/li\u003e \u003cli\u003e\n\u003cstrong\u003eFourth-Generation Tensor Cores\u003c\/strong\u003e: Accelerated AI and machine learning capabilities\u003c\/li\u003e \u003cli\u003e\n\u003cstrong\u003eNVIDIA NVLink\u003c\/strong\u003e: High-speed GPU-to-GPU interconnect for multi-GPU configurations\u003c\/li\u003e \u003cli\u003e\n\u003cstrong\u003ePCIe Gen 4\u003c\/strong\u003e: Ensures rapid data transfer between CPU and GPU\u003c\/li\u003e \u003c\/ul\u003e \u003ch2\u003ePerformance Metrics\u003c\/h2\u003e \u003cp\u003eBased on our internal benchmarks, the L40 delivers impressive performance across various tasks:\u003c\/p\u003e \u003cul\u003e \u003cli\u003e\n\u003cstrong\u003eImage Generation\u003c\/strong\u003e: $41.49 per day\u003c\/li\u003e \u003cli\u003e\n\u003cstrong\u003eVideo Generation\u003c\/strong\u003e: $50.53 per day\u003c\/li\u003e \u003cli\u003e\n\u003cstrong\u003eLarge Language Models\u003c\/strong\u003e: $58.10 per day\u003c\/li\u003e \u003cli\u003e\n\u003cstrong\u003eInstruction Services\u003c\/strong\u003e: $81.02 per day\u003c\/li\u003e \u003c\/ul\u003e \u003cp\u003eThese metrics represent conservative estimates of earning potential per content generated, showcasing the L40's capabilities in AI-driven tasks.\u003c\/p\u003e \u003ch2\u003eFlexible Configurations\u003c\/h2\u003e \u003cp\u003eWe offer the L40 in various high-performance computing setups:\u003c\/p\u003e \u003col\u003e \u003cli\u003e\n\u003cstrong\u003eINSTRUCT12\u003c\/strong\u003e: Up to 12 GPUs, 288-960GB VRAM\u003c\/li\u003e \u003cli\u003e\n\u003cstrong\u003e70B\u003c\/strong\u003e: Up to 12 GPUs 288 GB VRAM\u003c\/li\u003e \u003cli\u003e\n\u003cstrong\u003e35b\u003c\/strong\u003e: 192 GB VRAM\u003c\/li\u003e \u003c\/ol\u003e \u003cp\u003eOur most popular configuration, the 6 GPU setup, includes:\u003c\/p\u003e \u003cul\u003e \u003cli\u003eASRock Rack ROMED8-2T Motherboard\u003c\/li\u003e \u003cli\u003eAMD EPYC 7542 CPU\u003c\/li\u003e \u003cli\u003e512GB (8 x 64GB) SK Hynix 2666MHz REG ECC RAM\u003c\/li\u003e \u003cli\u003eUp to 6 NVIDIA L40 GPUs\u003c\/li\u003e \u003cli\u003eThis is Gpu for serious AI and multimedia workload\u003c\/li\u003e \u003c\/ul\u003e \u003ch2\u003eIdeal Applications\u003c\/h2\u003e \u003cul\u003e \u003cli\u003eAI and Machine Learning\u003c\/li\u003e \u003cli\u003eData Analytics\u003c\/li\u003e \u003cli\u003eScientific Simulations\u003c\/li\u003e \u003cli\u003e3D Rendering and Animation\u003c\/li\u003e \u003cli\u003eVirtual Reality and Augmented Reality\u003c\/li\u003e \u003cli\u003eHigh-Performance Computing (HPC)\u003c\/li\u003e \u003c\/ul\u003e \u003ch2\u003eWhy Choose the NVIDIA L40?\u003c\/h2\u003e \u003cul\u003e \u003cli\u003e\n\u003cstrong\u003eUnmatched Performance\u003c\/strong\u003e: Significantly outperforms consumer-grade GPUs like the RTX 3090 and 4090 in AI and compute tasks\u003c\/li\u003e \u003cli\u003e\n\u003cstrong\u003eScalability\u003c\/strong\u003e: Supports up to 12 GPUs in a single system for massive parallel processing\u003c\/li\u003e \u003cli\u003e\n\u003cstrong\u003eEnergy Efficiency\u003c\/strong\u003e: Optimized power consumption for better TCO\u003c\/li\u003e \u003cli\u003e\n\u003cstrong\u003eVersatility\u003c\/strong\u003e: Excels in a wide range of applications from AI to graphics\u003c\/li\u003e \u003cli\u003e\n\u003cstrong\u003eFuture-Proof\u003c\/strong\u003e: Stay ahead with the latest GPU technology\u003c\/li\u003e \u003c\/ul\u003e \u003ch2\u003eComparison with Other GPUs\u003c\/h2\u003e \u003ctable\u003e \u003cthead\u003e \u003ctr\u003e \u003cth\u003eGPU Model\u003c\/th\u003e \u003cth\u003eMax GPUs per System\u003c\/th\u003e \u003cth\u003eImage Gen ($\/day)\u003c\/th\u003e \u003cth\u003eVideo Gen ($\/day)\u003c\/th\u003e \u003cth\u003eLLM ($\/day)\u003c\/th\u003e \u003cth\u003e\u003c\/th\u003e \u003c\/tr\u003e \u003c\/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd\u003eL40\u003c\/td\u003e \u003ctd\u003e6 (up to 12 custom)\u003c\/td\u003e \u003ctd\u003e$41.49\u003c\/td\u003e \u003ctd\u003e$50.53\u003c\/td\u003e \u003ctd\u003e$58.10\u003c\/td\u003e \u003ctd\u003e\u003c\/td\u003e \u003c\/tr\u003e \u003ctr\u003e \u003ctd\u003eA100 40GB\u003c\/td\u003e \u003ctd\u003e6 (up to 12 custom)\u003c\/td\u003e \u003ctd\u003e$32.36\u003c\/td\u003e \u003ctd\u003e$44.87\u003c\/td\u003e \u003ctd\u003e$43.30\u003c\/td\u003e \u003ctd\u003e\u003c\/td\u003e \u003c\/tr\u003e \u003ctr\u003e \u003ctd\u003eRTX 4090\u003c\/td\u003e \u003ctd\u003e4 (up to 6 custom)\u003c\/td\u003e \u003ctd\u003e$4.64\u003c\/td\u003e \u003ctd\u003e$6.12\u003c\/td\u003e \u003ctd\u003e$6.76\u003c\/td\u003e \u003ctd\u003e\u003c\/td\u003e \u003c\/tr\u003e \u003c\/tbody\u003e \u003c\/table\u003e \u003cp\u003eAs shown, the L40 outperforms other high-end GPUs across various AI and compute tasks, making it an excellent investment for demanding workloads.\u003c\/p\u003e \u003cp\u003eElevate your computing capabilities with the NVIDIA L40 GPU. Contact our sales team for custom configurations and detailed pricing information.\u003c\/p\u003e NVIDIA L40 GPU for AI and Graphics Workloads","brand":"nVidia","offers":[{"title":"Default Title","offer_id":49002914218312,"sku":null,"price":8406.53,"currency_code":"EUR","in_stock":true}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/0843\/5479\/3800\/files\/nvidia-l40-gpu-e1724748108387.png?v=1725528643"},{"product_id":"nvidia-geforce-rtx-4090-24-gb-gddr6x-refurbished","title":"NVIDIA GeForce RTX 4090 24 GB GDDR6X (Refurbished)","description":"\u003cstyle\u003e\n.rtx{--bl:#334fb4;--dk:#26408f;--ink:#17171a;--soft:#f3f7fd;--muted:#6c6660;--line:#dce6f5;font-family:-apple-system,BlinkMacSystemFont,\"Segoe UI\",Roboto,Helvetica,Arial,sans-serif;color:var(--ink);line-height:1.6;font-size:15px;}\n.rtx *{box-sizing:border-box;}\n.rtx-head{background:var(--ink);color:#fff;border-radius:14px;padding:30px 28px;position:relative;overflow:hidden;}\n.rtx-head:before{content:\"\";position:absolute;inset:0;background:radial-gradient(120% 100% at 100% 0,rgba(51,79,180,.32),transparent 55%);}\n.rtx-head\u003e*{position:relative;}\n.rtx-head h2{font-size:clamp(22px,3.2vw,30px);font-weight:800;margin:0 0 10px;color:#fff !important;}\n.rtx-head p{color:#d7d2cb;max-width:64ch;margin:0;}\n.rtx-badge{display:inline-block;background:var(--bl);color:#fff !important;font-size:12px;font-weight:700;letter-spacing:.08em;text-transform:uppercase;padding:5px 11px;border-radius:999px;margin-bottom:14px;}\n.rtx-hi{display:grid;grid-template-columns:repeat(4,1fr);gap:1px;background:var(--line);border:1px solid var(--line);border-radius:12px;overflow:hidden;margin:18px 0 6px;}\n.rtx-hi div{background:#fff;padding:18px 12px;text-align:center;}\n.rtx-hi .v{font-size:20px;font-weight:800;}\n.rtx-hi .v small{color:var(--bl);font-size:12px;}\n.rtx-hi .k{font-size:11.5px;color:var(--muted);margin-top:4px;}\n.rtx-sec{padding:24px 0 4px;}\n.rtx-ey{font-size:12px;font-weight:700;letter-spacing:.16em;text-transform:uppercase;color:var(--dk);}\n.rtx h2.t{font-size:clamp(19px,2.4vw,24px);font-weight:800;margin:6px 0 12px;}\n.rtx p.lead{font-size:16px;color:#3a3a36;}\n.rtx table{width:100%;border-collapse:collapse;font-size:14px;margin:6px 0;}\n.rtx table.spec td{padding:9px 12px;border-bottom:1px solid var(--line);}\n.rtx table.spec tr:nth-child(odd) td{background:#f6f9fc;}\n.rtx table.spec td:first-child{width:42%;color:var(--muted);font-weight:600;}\n.rtx ul.tick{list-style:none;margin:6px 0;padding:0;}\n.rtx ul.tick li{position:relative;padding:7px 0 7px 26px;border-bottom:1px solid var(--line);font-size:14.5px;}\n.rtx ul.tick li:last-child{border-bottom:none;}\n.rtx ul.tick li:before{content:\"\";position:absolute;left:4px;top:14px;width:9px;height:9px;border-radius:2px;background:var(--bl);}\n.rtx-note{background:var(--soft);border-left:3px solid var(--bl);padding:13px 16px;border-radius:0 8px 8px 0;font-size:13.5px;color:#4a4641;margin-top:14px;}\n.rtx-faq details{border-bottom:1px solid var(--line);padding:14px 0;}\n.rtx-faq summary{font-weight:700;cursor:pointer;list-style:none;display:flex;justify-content:space-between;font-size:15px;}\n.rtx-faq summary:after{content:\"+\";color:var(--bl);font-weight:700;}\n.rtx-faq details[open] summary:after{content:\"–\";}\n.rtx-faq p{margin:9px 0 0;color:var(--muted);font-size:14px;}\n@media(max-width:760px){.rtx-hi{grid-template-columns:repeat(2,1fr);}}\n\u003c\/style\u003e\n\u003cdiv class=\"rtx\"\u003e\n\u003cdiv class=\"rtx-head\"\u003e\n\u003cspan class=\"rtx-badge\"\u003eRefurbished · Bench-tested\u003c\/span\u003e\n\u003ch2\u003eNVIDIA GeForce RTX 4090 — 24 GB for on-prem AI\u003c\/h2\u003e\n\u003cp\u003eA professionally refurbished RTX 4090: 24 GB of GDDR6X and Ada Lovelace compute, tested and verified stable under sustained load. The workhorse card for multi-GPU LLM inference and workstation AI at a fraction of datacenter-GPU cost.\u003c\/p\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-hi\"\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003e24 \u003csmall\u003eGB\u003c\/small\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003eGDDR6X VRAM\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003e16,384\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003eCUDA cores\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003e~1,008 \u003csmall\u003eGB\/s\u003c\/small\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003eMemory bandwidth\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003e450 \u003csmall\u003eW\u003c\/small\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003eBoard power\u003c\/div\u003e\n\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eOverview\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eThe 24 GB workhorse for AI inference\u003c\/h2\u003e\n\u003cp class=\"lead\"\u003eThe RTX 4090 pairs 24 GB of GDDR6X with 16,384 Ada Lovelace CUDA cores and 4th-gen Tensor cores — enough VRAM to serve quantized 7B–13B models comfortably, or to run a 70B-class model across two to four cards. It is the price-performance backbone of Kentino's multi-GPU AI builds. Each refurbished unit is cleaned, inspected and stress-tested before it ships.\u003c\/p\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eSpecifications\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eTechnical data\u003c\/h2\u003e\n\u003ctable class=\"spec\"\u003e\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd\u003eGPU\u003c\/td\u003e\n\u003ctd\u003eNVIDIA AD102 — Ada Lovelace\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eMemory\u003c\/td\u003e\n\u003ctd\u003e24 GB GDDR6X, 384-bit\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eMemory bandwidth\u003c\/td\u003e\n\u003ctd\u003e~1,008 GB\/s\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eCUDA cores\u003c\/td\u003e\n\u003ctd\u003e16,384\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eTensor cores\u003c\/td\u003e\n\u003ctd\u003e512 (4th gen) · up to ~330 TFLOPS FP16\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eRT cores\u003c\/td\u003e\n\u003ctd\u003e128 (3rd gen)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eInterface\u003c\/td\u003e\n\u003ctd\u003ePCIe 4.0 ×16\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eBoard power\u003c\/td\u003e\n\u003ctd\u003e450 W (16-pin 12VHPWR)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eCondition\u003c\/td\u003e\n\u003ctd\u003eRefurbished — bench-tested, verified stable\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003c\/tbody\u003e\u003c\/table\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eBest for\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eWhere it fits\u003c\/h2\u003e\n\u003cul class=\"tick\"\u003e\n\u003cli\u003eOn-prem LLM inference — 24 GB hosts quantized 7B–13B models; scale to 70B across multiple cards.\u003c\/li\u003e\n\u003cli\u003eAI workstations and fine-tuning of small-to-mid models.\u003c\/li\u003e\n\u003cli\u003eRendering, simulation and content-creation workloads.\u003c\/li\u003e\n\u003cli\u003eCost-effective multi-GPU servers where datacenter GPUs are overkill.\u003c\/li\u003e\n\u003c\/ul\u003e\n\u003cdiv class=\"rtx-note\"\u003eBuilding a multi-GPU server around these? Kentino assembles, benchmarks and commissions complete RTX 4090 AI servers — ask us for a configured build.\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eQuestions\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eFAQ\u003c\/h2\u003e\n\u003cdiv class=\"rtx-faq\"\u003e\n\u003cdetails open\u003e\u003csummary\u003eWhat does \"refurbished\" mean here?\u003c\/summary\u003e\u003cp\u003eThe card has been professionally cleaned, visually inspected and stress-tested for stability and full VRAM function before dispatch. It is not a new retail unit.\u003c\/p\u003e\u003c\/details\u003e\n\u003cdetails\u003e\u003csummary\u003eCan it run a 70B model?\u003c\/summary\u003e\u003cp\u003eNot on a single card — a 70B model at 4-bit needs ~40 GB of VRAM. Two to four RTX 4090 (48–96 GB total) handle it via tensor-parallel inference.\u003c\/p\u003e\u003c\/details\u003e\n\u003cdetails\u003e\u003csummary\u003eDoes it fit a standard server?\u003c\/summary\u003e\u003cp\u003eThe RTX 4090 is a large triple-slot card. It fits tower and 4U rack builds; confirm slot clearance and 16-pin power availability in your chassis.\u003c\/p\u003e\u003c\/details\u003e\n\u003c\/div\u003e\n\u003c\/div\u003e\n\u003c\/div\u003e","brand":"Kentino","offers":[{"title":"Default Title","offer_id":53519050670408,"sku":null,"price":1830.0,"currency_code":"EUR","in_stock":true}]},{"product_id":"nvidia-rtx-pro-6000-blackwell-max-q-96-gb-gddr7-ecc","title":"NVIDIA RTX PRO 6000 Blackwell Max-Q 96 GB GDDR7 ECC","description":"\u003cstyle\u003e\n.rtx{--bl:#334fb4;--dk:#26408f;--ink:#17171a;--soft:#f3f7fd;--muted:#6c6660;--line:#dce6f5;font-family:-apple-system,BlinkMacSystemFont,\"Segoe UI\",Roboto,Helvetica,Arial,sans-serif;color:var(--ink);line-height:1.6;font-size:15px;}\n.rtx *{box-sizing:border-box;}\n.rtx-head{background:var(--ink);color:#fff;border-radius:14px;padding:30px 28px;position:relative;overflow:hidden;}\n.rtx-head:before{content:\"\";position:absolute;inset:0;background:radial-gradient(120% 100% at 100% 0,rgba(51,79,180,.32),transparent 55%);}\n.rtx-head\u003e*{position:relative;}\n.rtx-head h2{font-size:clamp(22px,3.2vw,30px);font-weight:800;margin:0 0 10px;color:#fff !important;}\n.rtx-head p{color:#d7d2cb;max-width:64ch;margin:0;}\n.rtx-badge{display:inline-block;background:var(--bl);color:#fff !important;font-size:12px;font-weight:700;letter-spacing:.08em;text-transform:uppercase;padding:5px 11px;border-radius:999px;margin-bottom:14px;}\n.rtx-hi{display:grid;grid-template-columns:repeat(4,1fr);gap:1px;background:var(--line);border:1px solid var(--line);border-radius:12px;overflow:hidden;margin:18px 0 6px;}\n.rtx-hi div{background:#fff;padding:18px 12px;text-align:center;}\n.rtx-hi .v{font-size:20px;font-weight:800;}\n.rtx-hi .v small{color:var(--bl);font-size:12px;}\n.rtx-hi .k{font-size:11.5px;color:var(--muted);margin-top:4px;}\n.rtx-sec{padding:24px 0 4px;}\n.rtx-ey{font-size:12px;font-weight:700;letter-spacing:.16em;text-transform:uppercase;color:var(--dk);}\n.rtx h2.t{font-size:clamp(19px,2.4vw,24px);font-weight:800;margin:6px 0 12px;}\n.rtx p.lead{font-size:16px;color:#3a3a36;}\n.rtx table{width:100%;border-collapse:collapse;font-size:14px;margin:6px 0;}\n.rtx table.spec td{padding:9px 12px;border-bottom:1px solid var(--line);}\n.rtx table.spec tr:nth-child(odd) td{background:#f6f9fc;}\n.rtx table.spec td:first-child{width:42%;color:var(--muted);font-weight:600;}\n.rtx ul.tick{list-style:none;margin:6px 0;padding:0;}\n.rtx ul.tick li{position:relative;padding:7px 0 7px 26px;border-bottom:1px solid var(--line);font-size:14.5px;}\n.rtx ul.tick li:last-child{border-bottom:none;}\n.rtx ul.tick li:before{content:\"\";position:absolute;left:4px;top:14px;width:9px;height:9px;border-radius:2px;background:var(--bl);}\n.rtx-note{background:var(--soft);border-left:3px solid var(--bl);padding:13px 16px;border-radius:0 8px 8px 0;font-size:13.5px;color:#4a4641;margin-top:14px;}\n.rtx-faq details{border-bottom:1px solid var(--line);padding:14px 0;}\n.rtx-faq summary{font-weight:700;cursor:pointer;list-style:none;display:flex;justify-content:space-between;font-size:15px;}\n.rtx-faq summary:after{content:\"+\";color:var(--bl);font-weight:700;}\n.rtx-faq details[open] summary:after{content:\"–\";}\n.rtx-faq p{margin:9px 0 0;color:var(--muted);font-size:14px;}\n@media(max-width:760px){.rtx-hi{grid-template-columns:repeat(2,1fr);}}\n\u003c\/style\u003e\u003cdiv class=\"rtx\"\u003e\n\u003cdiv class=\"rtx-head\"\u003e\n\u003cspan class=\"rtx-badge\"\u003eBlackwell · 96 GB ECC · 300 W\u003c\/span\u003e\n\u003ch2\u003eNVIDIA RTX PRO 6000 Blackwell Max-Q — 96 GB for on-prem AI\u003c\/h2\u003e\n\u003cp\u003eThe full 96 GB Blackwell GPU in a 300 W envelope. Same unified ECC VRAM pool and same compute silicon as the 600 W card, drawing half the power — the card to choose when thermals, noise or a shared workstation chassis set the limit.\u003c\/p\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-hi\"\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003e96 \u003csmall\u003eGB\u003c\/small\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003eGDDR7 ECC VRAM\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003e24,064\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003eCUDA cores\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003e~1.8 \u003csmall\u003eTB\/s\u003c\/small\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003eMemory bandwidth\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003e300 \u003csmall\u003eW\u003c\/small\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003eBoard power\u003c\/div\u003e\n\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eOverview\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003e96 GB unified VRAM at 300 W\u003c\/h2\u003e\n\u003cp class=\"lead\"\u003eThe Max-Q variant carries the same Blackwell die and the same 96 GB of ECC GDDR7 as the full-power RTX PRO 6000, but is power-limited to 300 W. That matters more than the headline suggests: a single card holds a 70B-class model at 4-bit entirely in one unified memory pool, so there is no tensor-parallel split, no inter-GPU PCIe hop, and no partitioning work. Compared with stacking consumer cards to reach the same VRAM, you get ECC memory, a fraction of the power draw, and one slot instead of several.\u003c\/p\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eSpecifications\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eTechnical data\u003c\/h2\u003e\n\u003ctable class=\"spec\"\u003e\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd\u003eGPU\u003c\/td\u003e\n\u003ctd\u003eNVIDIA GB202 — Blackwell\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003ePart number\u003c\/td\u003e\n\u003ctd\u003e900-5G153\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eMemory\u003c\/td\u003e\n\u003ctd\u003e96 GB GDDR7 ECC, 512-bit\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eMemory bandwidth\u003c\/td\u003e\n\u003ctd\u003e~1.8 TB\/s\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eCUDA cores\u003c\/td\u003e\n\u003ctd\u003e24,064\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eTensor cores\u003c\/td\u003e\n\u003ctd\u003e752 (5th gen) · ~2,000 TOPS INT8\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eRT cores\u003c\/td\u003e\n\u003ctd\u003e188 (4th gen)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eInterface\u003c\/td\u003e\n\u003ctd\u003ePCIe 5.0 ×16\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eBoard power\u003c\/td\u003e\n\u003ctd\u003e300 W (16-pin 12V-2×6)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eCooling\u003c\/td\u003e\n\u003ctd\u003eActive blower, dual-slot\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eDisplay outputs\u003c\/td\u003e\n\u003ctd\u003e4× DisplayPort 2.1b\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eWarranty\u003c\/td\u003e\n\u003ctd\u003e36 months\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003c\/tbody\u003e\u003c\/table\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eBest for\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eWhere it fits\u003c\/h2\u003e\n\u003cul class=\"tick\"\u003e\n\u003cli\u003eSingle-card 70B inference — 96 GB holds a 70B model at 4-bit in one unified pool, no model splitting.\u003c\/li\u003e\n\u003cli\u003eQuiet workstations and offices where a 600 W card is too much heat and noise.\u003c\/li\u003e\n\u003cli\u003eMulti-GPU builds on a constrained power budget — four Max-Q draw about the same as two full-power cards.\u003c\/li\u003e\n\u003cli\u003eFine-tuning mid-size models where ECC memory and long-run stability matter.\u003c\/li\u003e\n\u003cli\u003eImage and video generation with large models held resident in VRAM.\u003c\/li\u003e\n\u003c\/ul\u003e\n\u003cdiv class=\"rtx-note\"\u003eIndicative single-stream throughput: roughly 25–35 tok\/s on a 70B model at 4-bit, and well over 100 tok\/s on an 8B model. Figures are indicative estimates from published external references, not measured on Kentino hardware — tell us your model and context length and we will size it properly.\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eQuestions\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eFAQ\u003c\/h2\u003e\n\u003cdiv class=\"rtx-faq\"\u003e\n\u003cdetails open\u003e\u003csummary\u003eHow does Max-Q differ from the 600 W card?\u003c\/summary\u003e\u003cp\u003eSame GPU, same 96 GB of ECC GDDR7, same memory bandwidth. The Max-Q is capped at 300 W instead of 600 W, so sustained throughput is lower on compute-bound work — but any model that fits in 96 GB still fits, and VRAM capacity is what usually decides whether a model runs at all.\u003c\/p\u003e\u003c\/details\u003e\n\u003cdetails\u003e\u003csummary\u003eCan one card really run a 70B model?\u003c\/summary\u003e\u003cp\u003eYes. A 70B model at 4-bit needs roughly 40 GB, so it fits in 96 GB with substantial room left for KV cache and long context. That is the main reason to choose this card over several smaller ones.\u003c\/p\u003e\u003c\/details\u003e\n\u003cdetails\u003e\u003csummary\u003eDo I need ECC memory?\u003c\/summary\u003e\u003cp\u003eFor interactive work, no. For unattended training runs, long batch jobs or anything where a silent bit-flip would corrupt a result, ECC is the difference between a reliable machine and an unexplained failure.\u003c\/p\u003e\u003c\/details\u003e\n\u003cdetails\u003e\u003csummary\u003eWill it fit my chassis?\u003c\/summary\u003e\u003cp\u003eIt is a dual-slot active-cooled card, so it fits standard tower and 4U builds. Confirm you have a free PCIe 5.0 ×16 slot and a 16-pin 12V-2×6 power connector available.\u003c\/p\u003e\u003c\/details\u003e\n\u003cdetails\u003e\u003csummary\u003eWhat is the lead time?\u003c\/summary\u003e\u003cp\u003eThese are sourced to order — typically 10–21 days. Availability on this part moves quickly, so ask us to confirm before you plan around a date.\u003c\/p\u003e\u003c\/details\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-note\"\u003eWant this card in a finished machine? Kentino builds, benchmarks and commissions complete RTX PRO 6000 AI servers and workstations — ask us for a configured build.\u003c\/div\u003e\n\u003c\/div\u003e\n\u003c\/div\u003e\n","brand":"NVIDIA","offers":[{"title":"Default Title","offer_id":53617326391624,"sku":null,"price":22000.0,"currency_code":"EUR","in_stock":true}]},{"product_id":"nvidia-rtx-pro-6000-blackwell-workstation-edition-96-gb-gddr7-ecc","title":"NVIDIA RTX PRO 6000 Blackwell Workstation Edition 96 GB GDDR7 ECC","description":"\u003cstyle\u003e\n.rtx{--bl:#334fb4;--dk:#26408f;--ink:#17171a;--soft:#f3f7fd;--muted:#6c6660;--line:#dce6f5;font-family:-apple-system,BlinkMacSystemFont,\"Segoe UI\",Roboto,Helvetica,Arial,sans-serif;color:var(--ink);line-height:1.6;font-size:15px;}\n.rtx *{box-sizing:border-box;}\n.rtx-head{background:var(--ink);color:#fff;border-radius:14px;padding:30px 28px;position:relative;overflow:hidden;}\n.rtx-head:before{content:\"\";position:absolute;inset:0;background:radial-gradient(120% 100% at 100% 0,rgba(51,79,180,.32),transparent 55%);}\n.rtx-head\u003e*{position:relative;}\n.rtx-head h2{font-size:clamp(22px,3.2vw,30px);font-weight:800;margin:0 0 10px;color:#fff !important;}\n.rtx-head p{color:#d7d2cb;max-width:64ch;margin:0;}\n.rtx-badge{display:inline-block;background:var(--bl);color:#fff !important;font-size:12px;font-weight:700;letter-spacing:.08em;text-transform:uppercase;padding:5px 11px;border-radius:999px;margin-bottom:14px;}\n.rtx-hi{display:grid;grid-template-columns:repeat(4,1fr);gap:1px;background:var(--line);border:1px solid var(--line);border-radius:12px;overflow:hidden;margin:18px 0 6px;}\n.rtx-hi div{background:#fff;padding:18px 12px;text-align:center;}\n.rtx-hi .v{font-size:20px;font-weight:800;}\n.rtx-hi .v small{color:var(--bl);font-size:12px;}\n.rtx-hi .k{font-size:11.5px;color:var(--muted);margin-top:4px;}\n.rtx-sec{padding:24px 0 4px;}\n.rtx-ey{font-size:12px;font-weight:700;letter-spacing:.16em;text-transform:uppercase;color:var(--dk);}\n.rtx h2.t{font-size:clamp(19px,2.4vw,24px);font-weight:800;margin:6px 0 12px;}\n.rtx p.lead{font-size:16px;color:#3a3a36;}\n.rtx table{width:100%;border-collapse:collapse;font-size:14px;margin:6px 0;}\n.rtx table.spec td{padding:9px 12px;border-bottom:1px solid var(--line);}\n.rtx table.spec tr:nth-child(odd) td{background:#f6f9fc;}\n.rtx table.spec td:first-child{width:42%;color:var(--muted);font-weight:600;}\n.rtx ul.tick{list-style:none;margin:6px 0;padding:0;}\n.rtx ul.tick li{position:relative;padding:7px 0 7px 26px;border-bottom:1px solid var(--line);font-size:14.5px;}\n.rtx ul.tick li:last-child{border-bottom:none;}\n.rtx ul.tick li:before{content:\"\";position:absolute;left:4px;top:14px;width:9px;height:9px;border-radius:2px;background:var(--bl);}\n.rtx-note{background:var(--soft);border-left:3px solid var(--bl);padding:13px 16px;border-radius:0 8px 8px 0;font-size:13.5px;color:#4a4641;margin-top:14px;}\n.rtx-faq details{border-bottom:1px solid var(--line);padding:14px 0;}\n.rtx-faq summary{font-weight:700;cursor:pointer;list-style:none;display:flex;justify-content:space-between;font-size:15px;}\n.rtx-faq summary:after{content:\"+\";color:var(--bl);font-weight:700;}\n.rtx-faq details[open] summary:after{content:\"–\";}\n.rtx-faq p{margin:9px 0 0;color:var(--muted);font-size:14px;}\n@media(max-width:760px){.rtx-hi{grid-template-columns:repeat(2,1fr);}}\n\u003c\/style\u003e\n\u003cdiv class=\"rtx\"\u003e\n\u003cdiv class=\"rtx-head\"\u003e\n\u003cspan class=\"rtx-badge\"\u003eBlackwell · 96 GB ECC · 600 W\u003c\/span\u003e\n\u003ch2\u003eNVIDIA RTX PRO 6000 Blackwell Workstation Edition — 96 GB\u003c\/h2\u003e\n\u003cp\u003eThe full-power Blackwell professional card: 96 GB of unified ECC GDDR7 and 24,064 CUDA cores at a 600 W board limit. The fastest single-GPU option we ship for local model training, fine-tuning and heavy inference.\u003c\/p\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-hi\"\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003e96 \u003csmall\u003eGB\u003c\/small\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003eGDDR7 ECC VRAM\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003e24,064\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003eCUDA cores\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003e~1.8 \u003csmall\u003eTB\/s\u003c\/small\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003eMemory bandwidth\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003e600 \u003csmall\u003eW\u003c\/small\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003eBoard power\u003c\/div\u003e\n\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eOverview\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eMaximum single-GPU throughput\u003c\/h2\u003e\n\u003cp class=\"lead\"\u003eThis is the unrestricted version of the card: the same 96 GB unified ECC memory pool as the Max-Q, with the power ceiling lifted to 600 W so the GPU can hold high clocks under sustained load. On compute-bound work — fine-tuning, long training runs, batched inference, heavy rendering and simulation — that headroom is where the extra performance comes from. A single card holds a 70B-class model at 4-bit with room to spare for long context, so there is no tensor-parallel split to manage.\u003c\/p\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eSpecifications\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eTechnical data\u003c\/h2\u003e\n\u003ctable class=\"spec\"\u003e\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd\u003eGPU\u003c\/td\u003e\n\u003ctd\u003eNVIDIA GB202 — Blackwell\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003ePart number\u003c\/td\u003e\n\u003ctd\u003e900-5G144\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eMemory\u003c\/td\u003e\n\u003ctd\u003e96 GB GDDR7 ECC, 512-bit\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eMemory bandwidth\u003c\/td\u003e\n\u003ctd\u003e~1.8 TB\/s\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eCUDA cores\u003c\/td\u003e\n\u003ctd\u003e24,064\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eTensor cores\u003c\/td\u003e\n\u003ctd\u003e752 (5th gen) · ~2,000 TOPS INT8\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eRT cores\u003c\/td\u003e\n\u003ctd\u003e188 (4th gen)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eInterface\u003c\/td\u003e\n\u003ctd\u003ePCIe 5.0 ×16\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eBoard power\u003c\/td\u003e\n\u003ctd\u003e600 W (16-pin 12V-2×6)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eCooling\u003c\/td\u003e\n\u003ctd\u003eActive flow-through, dual-slot\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eDisplay outputs\u003c\/td\u003e\n\u003ctd\u003e4× DisplayPort 2.1b\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eWarranty\u003c\/td\u003e\n\u003ctd\u003e3-year manufacturer warranty\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003c\/tbody\u003e\u003c\/table\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eBest for\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eWhere it fits\u003c\/h2\u003e\n\u003cul class=\"tick\"\u003e\n\u003cli\u003eLocal fine-tuning and training where sustained compute, not just VRAM capacity, sets the pace.\u003c\/li\u003e\n\u003cli\u003eSingle-card 70B inference with long context held entirely in one memory pool.\u003c\/li\u003e\n\u003cli\u003eBatched or multi-user inference serving where throughput per card matters.\u003c\/li\u003e\n\u003cli\u003eProfessional visualisation, rendering and simulation alongside AI work.\u003c\/li\u003e\n\u003cli\u003eWorkstations where one full-power card is preferable to several smaller GPUs.\u003c\/li\u003e\n\u003c\/ul\u003e\n\u003cdiv class=\"rtx-note\"\u003ePlan the power and cooling before you buy: at 600 W this card needs a quality PSU with real headroom and a chassis that can actually move the heat. We size PSU and airflow as part of any build we supply.\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eQuestions\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eFAQ\u003c\/h2\u003e\n\u003cdiv class=\"rtx-faq\"\u003e\n\u003cdetails open\u003e\u003csummary\u003eShould I take this or the Max-Q?\u003c\/summary\u003e\u003cp\u003eBoth hold the same 96 GB, so both run the same models. Choose this one if your work is compute-bound and you can supply 600 W of power and cooling. Choose the Max-Q if power, heat or noise is the binding constraint — it costs less and draws half the power for the same VRAM capacity.\u003c\/p\u003e\u003c\/details\u003e\n\u003cdetails\u003e\u003csummary\u003eWhat PSU do I need?\u003c\/summary\u003e\u003cp\u003eFor a single card, plan on a 1,200 W or larger quality ATX 3.0 unit with a native 16-pin 12V-2×6 connector, and more if you are pairing it with a high-core-count CPU. We specify this properly for machines we build.\u003c\/p\u003e\u003c\/details\u003e\n\u003cdetails\u003e\u003csummary\u003eCan I run more than one?\u003c\/summary\u003e\u003cp\u003eYes, but 600 W per card adds up fast — multi-card configurations need deliberate power and airflow design, and in most chassis the Max-Q is the better multi-GPU choice. Talk to us before committing to a multi-card layout.\u003c\/p\u003e\u003c\/details\u003e\n\u003cdetails\u003e\u003csummary\u003eIs this the same as the Server Edition?\u003c\/summary\u003e\u003cp\u003eNo. The Server Edition is passively cooled for front-to-back rack airflow and has no display outputs. This card is actively cooled with four DisplayPort outputs, intended for workstations and tower builds.\u003c\/p\u003e\u003c\/details\u003e\n\u003cdetails\u003e\u003csummary\u003eWhat is the lead time?\u003c\/summary\u003e\u003cp\u003eTypically 10–21 days, sourced to order. Availability moves quickly on this part — ask us to confirm before planning around a date.\u003c\/p\u003e\u003c\/details\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-note\"\u003eWant this card in a finished machine? Kentino builds, benchmarks and commissions complete RTX PRO 6000 AI servers and workstations — ask us for a configured build.\u003c\/div\u003e\n\u003c\/div\u003e\n\u003c\/div\u003e","brand":"NVIDIA","offers":[{"title":"Default Title","offer_id":53617381769544,"sku":null,"price":0.0,"currency_code":"EUR","in_stock":true}]},{"product_id":"nvidia-rtx-pro-6000-blackwell-server-edition-96-gb-gddr7-ecc","title":"NVIDIA RTX PRO 6000 Blackwell Server Edition 96 GB GDDR7 ECC","description":"\u003cstyle\u003e\n.rtx{--bl:#334fb4;--dk:#26408f;--ink:#17171a;--soft:#f3f7fd;--muted:#6c6660;--line:#dce6f5;font-family:-apple-system,BlinkMacSystemFont,\"Segoe UI\",Roboto,Helvetica,Arial,sans-serif;color:var(--ink);line-height:1.6;font-size:15px;}\n.rtx *{box-sizing:border-box;}\n.rtx-head{background:var(--ink);color:#fff;border-radius:14px;padding:30px 28px;position:relative;overflow:hidden;}\n.rtx-head:before{content:\"\";position:absolute;inset:0;background:radial-gradient(120% 100% at 100% 0,rgba(51,79,180,.32),transparent 55%);}\n.rtx-head\u003e*{position:relative;}\n.rtx-head h2{font-size:clamp(22px,3.2vw,30px);font-weight:800;margin:0 0 10px;color:#fff !important;}\n.rtx-head p{color:#d7d2cb;max-width:64ch;margin:0;}\n.rtx-badge{display:inline-block;background:var(--bl);color:#fff !important;font-size:12px;font-weight:700;letter-spacing:.08em;text-transform:uppercase;padding:5px 11px;border-radius:999px;margin-bottom:14px;}\n.rtx-hi{display:grid;grid-template-columns:repeat(4,1fr);gap:1px;background:var(--line);border:1px solid var(--line);border-radius:12px;overflow:hidden;margin:18px 0 6px;}\n.rtx-hi div{background:#fff;padding:18px 12px;text-align:center;}\n.rtx-hi .v{font-size:20px;font-weight:800;}\n.rtx-hi .v small{color:var(--bl);font-size:12px;}\n.rtx-hi .k{font-size:11.5px;color:var(--muted);margin-top:4px;}\n.rtx-sec{padding:24px 0 4px;}\n.rtx-ey{font-size:12px;font-weight:700;letter-spacing:.16em;text-transform:uppercase;color:var(--dk);}\n.rtx h2.t{font-size:clamp(19px,2.4vw,24px);font-weight:800;margin:6px 0 12px;}\n.rtx p.lead{font-size:16px;color:#3a3a36;}\n.rtx table{width:100%;border-collapse:collapse;font-size:14px;margin:6px 0;}\n.rtx table.spec td{padding:9px 12px;border-bottom:1px solid var(--line);}\n.rtx table.spec tr:nth-child(odd) td{background:#f6f9fc;}\n.rtx table.spec td:first-child{width:42%;color:var(--muted);font-weight:600;}\n.rtx ul.tick{list-style:none;margin:6px 0;padding:0;}\n.rtx ul.tick li{position:relative;padding:7px 0 7px 26px;border-bottom:1px solid var(--line);font-size:14.5px;}\n.rtx ul.tick li:last-child{border-bottom:none;}\n.rtx ul.tick li:before{content:\"\";position:absolute;left:4px;top:14px;width:9px;height:9px;border-radius:2px;background:var(--bl);}\n.rtx-note{background:var(--soft);border-left:3px solid var(--bl);padding:13px 16px;border-radius:0 8px 8px 0;font-size:13.5px;color:#4a4641;margin-top:14px;}\n.rtx-faq details{border-bottom:1px solid var(--line);padding:14px 0;}\n.rtx-faq summary{font-weight:700;cursor:pointer;list-style:none;display:flex;justify-content:space-between;font-size:15px;}\n.rtx-faq summary:after{content:\"+\";color:var(--bl);font-weight:700;}\n.rtx-faq details[open] summary:after{content:\"–\";}\n.rtx-faq p{margin:9px 0 0;color:var(--muted);font-size:14px;}\n@media(max-width:760px){.rtx-hi{grid-template-columns:repeat(2,1fr);}}\n\u003c\/style\u003e\n\u003cdiv class=\"rtx\"\u003e\n\u003cdiv class=\"rtx-head\"\u003e\n\u003cspan class=\"rtx-badge\"\u003eBlackwell · 96 GB ECC · Passive · Rack\u003c\/span\u003e\n\u003ch2\u003eNVIDIA RTX PRO 6000 Blackwell Server Edition — 96 GB\u003c\/h2\u003e\n\u003cp\u003eThe datacenter variant: 96 GB of unified ECC GDDR7, passively cooled for front-to-back rack airflow, with a configurable 400–600 W power envelope. Built to run continuously in a server chassis rather than on a desk.\u003c\/p\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-hi\"\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003e96 \u003csmall\u003eGB\u003c\/small\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003eGDDR7 ECC VRAM\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003e24,064\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003eCUDA cores\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003e~1.8 \u003csmall\u003eTB\/s\u003c\/small\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003eMemory bandwidth\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003e400–600 \u003csmall\u003eW\u003c\/small\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003eConfigurable TDP\u003c\/div\u003e\n\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eOverview\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eBuilt for 24\/7 rack operation\u003c\/h2\u003e\n\u003cp class=\"lead\"\u003eThe Server Edition carries the same Blackwell die and the same 96 GB ECC memory pool as the workstation cards, but is engineered for a different environment. It has no fan and no display outputs: cooling comes from the chassis, in a front-to-back path, which is what makes dense multi-GPU rack builds possible in the first place. The power envelope is configurable between 400 W and 600 W, so density and thermal budget can be traded against per-card throughput. This is the card behind Kentino's multi-GPU Kentino AI rack servers.\u003c\/p\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eSpecifications\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eTechnical data\u003c\/h2\u003e\n\u003ctable class=\"spec\"\u003e\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd\u003eGPU\u003c\/td\u003e\n\u003ctd\u003eNVIDIA GB202 — Blackwell\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eMemory\u003c\/td\u003e\n\u003ctd\u003e96 GB GDDR7 ECC, 512-bit\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eMemory bandwidth\u003c\/td\u003e\n\u003ctd\u003e~1.8 TB\/s\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eCUDA cores\u003c\/td\u003e\n\u003ctd\u003e24,064\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eTensor cores\u003c\/td\u003e\n\u003ctd\u003e752 (5th gen) · ~2,000 TOPS INT8\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eRT cores\u003c\/td\u003e\n\u003ctd\u003e188 (4th gen)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eInterface\u003c\/td\u003e\n\u003ctd\u003ePCIe 5.0 ×16\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eBoard power\u003c\/td\u003e\n\u003ctd\u003e400–600 W configurable\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eCooling\u003c\/td\u003e\n\u003ctd\u003ePassive — requires chassis front-to-back airflow\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eDisplay outputs\u003c\/td\u003e\n\u003ctd\u003eNone\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eForm factor\u003c\/td\u003e\n\u003ctd\u003eDual-slot, full-height\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eWarranty\u003c\/td\u003e\n\u003ctd\u003e3-year manufacturer warranty\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003c\/tbody\u003e\u003c\/table\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eBest for\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eWhere it fits\u003c\/h2\u003e\n\u003cul class=\"tick\"\u003e\n\u003cli\u003eDense multi-GPU rack servers — 2, 4, 6 or 8 cards in a single chassis.\u003c\/li\u003e\n\u003cli\u003eContinuous 24\/7 inference serving in a datacenter or server room.\u003c\/li\u003e\n\u003cli\u003eShared or multi-tenant AI infrastructure where ECC and stability are requirements.\u003c\/li\u003e\n\u003cli\u003eDeployments where per-card power must be capped to fit a rack thermal budget.\u003c\/li\u003e\n\u003cli\u003eLarge-model serving across several cards with 96 GB of unified VRAM each.\u003c\/li\u003e\n\u003c\/ul\u003e\n\u003cdiv class=\"rtx-note\"\u003eThis card cannot cool itself. It requires a server chassis with a proper front-to-back airflow path and adequate static pressure — it will overheat in a standard desktop case. If you are not already running a suitable chassis, ask us for a complete build instead.\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eQuestions\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eFAQ\u003c\/h2\u003e\n\u003cdiv class=\"rtx-faq\"\u003e\n\u003cdetails open\u003e\u003csummary\u003eCan I use this in a desktop PC?\u003c\/summary\u003e\u003cp\u003eNo. It has no fan and relies entirely on chassis airflow designed for passive GPUs. In a normal tower case it will throttle and then overheat. Use the Workstation or Max-Q edition for tower and desk builds.\u003c\/p\u003e\u003c\/details\u003e\n\u003cdetails\u003e\u003csummary\u003eHow does it compare with the workstation cards?\u003c\/summary\u003e\u003cp\u003eIdentical GPU and identical 96 GB of ECC memory. The differences are physical: passive cooling, no display outputs, and a configurable 400–600 W envelope for rack density. Choose it only if you have a server chassis to put it in.\u003c\/p\u003e\u003c\/details\u003e\n\u003cdetails\u003e\u003csummary\u003eWhy cap it at 400 W?\u003c\/summary\u003e\u003cp\u003eIn a dense build, total rack power and cooling — not the individual card — is usually the limit. Capping each card lets you fit more GPUs into the same thermal budget, which typically yields more total throughput than fewer cards running flat out.\u003c\/p\u003e\u003c\/details\u003e\n\u003cdetails\u003e\u003csummary\u003eDo you supply complete servers with these?\u003c\/summary\u003e\u003cp\u003eYes. The Kentino AI rack line is built around this card in 1, 2, 4, 6 and 8-GPU configurations, assembled, benchmarked and commissioned before delivery.\u003c\/p\u003e\u003c\/details\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-note\"\u003ePricing on this variant is quoted per enquiry — datacenter GPU cost moves week to week and we would rather give you a current number than a stale one. Tell us quantity and target chassis and we will come back with a firm quote and lead time.\u003c\/div\u003e\n\u003c\/div\u003e\n\u003c\/div\u003e","brand":"NVIDIA","offers":[{"title":"Default Title","offer_id":53617401037128,"sku":null,"price":23000.0,"currency_code":"EUR","in_stock":true}]},{"product_id":"nvidia-cmp-170hx-64-gb-hbm2e-modified-ex-mining","title":"NVIDIA CMP 170HX 64 GB HBM2e (Modified, Ex-Mining)","description":"\u003cstyle\u003e\n.rtx{--bl:#334fb4;--dk:#26408f;--ink:#17171a;--soft:#f3f7fd;--muted:#6c6660;--line:#dce6f5;font-family:-apple-system,BlinkMacSystemFont,\"Segoe UI\",Roboto,Helvetica,Arial,sans-serif;color:var(--ink);line-height:1.6;font-size:15px;}\n.rtx *{box-sizing:border-box;}\n.rtx-head{background:var(--ink);color:#fff;border-radius:14px;padding:30px 28px;position:relative;overflow:hidden;}\n.rtx-head:before{content:\"\";position:absolute;inset:0;background:radial-gradient(120% 100% at 100% 0,rgba(51,79,180,.32),transparent 55%);}\n.rtx-head\u003e*{position:relative;}\n.rtx-head h2{font-size:clamp(22px,3.2vw,30px);font-weight:800;margin:0 0 10px;color:#fff !important;}\n.rtx-head p{color:#d7d2cb;max-width:64ch;margin:0;}\n.rtx-badge{display:inline-block;background:var(--bl);color:#fff !important;font-size:12px;font-weight:700;letter-spacing:.08em;text-transform:uppercase;padding:5px 11px;border-radius:999px;margin-bottom:14px;}\n.rtx-hi{display:grid;grid-template-columns:repeat(4,1fr);gap:1px;background:var(--line);border:1px solid var(--line);border-radius:12px;overflow:hidden;margin:18px 0 6px;}\n.rtx-hi div{background:#fff;padding:18px 12px;text-align:center;}\n.rtx-hi .v{font-size:20px;font-weight:800;}\n.rtx-hi .v small{color:var(--bl);font-size:12px;}\n.rtx-hi .k{font-size:11.5px;color:var(--muted);margin-top:4px;}\n.rtx-sec{padding:24px 0 4px;}\n.rtx-ey{font-size:12px;font-weight:700;letter-spacing:.16em;text-transform:uppercase;color:var(--dk);}\n.rtx h2.t{font-size:clamp(19px,2.4vw,24px);font-weight:800;margin:6px 0 12px;}\n.rtx p.lead{font-size:16px;color:#3a3a36;}\n.rtx table{width:100%;border-collapse:collapse;font-size:14px;margin:6px 0;}\n.rtx table.spec td{padding:9px 12px;border-bottom:1px solid var(--line);}\n.rtx table.spec tr:nth-child(odd) td{background:#f6f9fc;}\n.rtx table.spec td:first-child{width:42%;color:var(--muted);font-weight:600;}\n.rtx ul.tick{list-style:none;margin:6px 0;padding:0;}\n.rtx ul.tick li{position:relative;padding:7px 0 7px 26px;border-bottom:1px solid var(--line);font-size:14.5px;}\n.rtx ul.tick li:last-child{border-bottom:none;}\n.rtx ul.tick li:before{content:\"\";position:absolute;left:4px;top:14px;width:9px;height:9px;border-radius:2px;background:var(--bl);}\n.rtx-note{background:var(--soft);border-left:3px solid var(--bl);padding:13px 16px;border-radius:0 8px 8px 0;font-size:13.5px;color:#4a4641;margin-top:14px;}\n.rtx-faq details{border-bottom:1px solid var(--line);padding:14px 0;}\n.rtx-faq summary{font-weight:700;cursor:pointer;list-style:none;display:flex;justify-content:space-between;font-size:15px;}\n.rtx-faq summary:after{content:\"+\";color:var(--bl);font-weight:700;}\n.rtx-faq details[open] summary:after{content:\"–\";}\n.rtx-faq p{margin:9px 0 0;color:var(--muted);font-size:14px;}\n@media(max-width:760px){.rtx-hi{grid-template-columns:repeat(2,1fr);}}\n\u003c\/style\u003e\n\u003cdiv class=\"rtx\"\u003e\n\u003cdiv class=\"rtx-head\"\u003e\n\u003cspan class=\"rtx-badge\"\u003eModified · Ex-Mining · 64 GB HBM2e\u003c\/span\u003e\n\u003ch2\u003eNVIDIA CMP 170HX 64 GB HBM2e — cheap VRAM for resident inference \u003c\/h2\u003e\n\u003cp\u003eA GA100-based CMP 170HX with its memory expanded to 64 GB of HBM2e. At €1,600 that is roughly €25 per gigabyte of high-bandwidth VRAM — around a fifth of what large-VRAM professional cards cost. It is modified hardware with real limits, and we spell them out below before you buy.\u003c\/p\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-hi\"\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003e64 \u003csmall\u003eGB\u003c\/small\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003eHBM2e VRAM\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003eGA100\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003eAmpere GPU\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003e~25 \u003csmall\u003e€\/GB\u003c\/small\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003eVRAM cost\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003ePCIe 1.0\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003e×4 or ×16 link\u003c\/div\u003e\n\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eRead this first\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eWhat this card is — and is not\u003c\/h2\u003e\n\u003cp class=\"lead\"\u003eThe CMP 170HX was sold by NVIDIA as a dedicated mining card: a cut-down GA100 with its PCIe link deliberately locked to PCIe 1.0 and most display and compute functions restricted. These units have been modified after the fact to carry 64 GB of HBM2e. That makes them an unusually cheap way to hold a large model entirely in high-bandwidth memory — and a poor choice for anything that depends on fast host transfers, multi-GPU scaling or training.\u003c\/p\u003e\n\u003cdiv class=\"rtx-note\"\u003e\n\u003cstrong\u003eThis is modified hardware operating outside its intended configuration.\u003c\/strong\u003e Compute is partially fused off compared with a full A100, so treat VRAM capacity — not throughput — as the reason to buy. Individual cards vary: some are more fully unlocked than others, and we cannot promise which you will receive. If you need guaranteed, predictable compute, buy an RTX PRO 6000 or a proper datacenter card instead and we will happily quote one.\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eThe PCIe limit\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eWhy the ×16 option costs €150 more\u003c\/h2\u003e\n\u003cp class=\"lead\"\u003eBoth versions are locked to PCIe \u003cstrong\u003e1.0\u003c\/strong\u003e signalling — that part cannot be undone. What differs is lane count, and it changes host-to-card bandwidth by 4×:\u003c\/p\u003e\n\u003ctable class=\"spec\"\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd\u003ePCIe 1.0 ×4 — €1,600\u003c\/td\u003e\n\u003ctd\u003e~1 GB\/s · filling 64 GB takes roughly a minute\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003ePCIe 1.0 ×16 — €1,750\u003c\/td\u003e\n\u003ctd\u003e~4 GB\/s · filling 64 GB takes roughly 15 seconds\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003c\/tbody\u003e\n\u003c\/table\u003e\n\u003cdiv class=\"rtx-note\"\u003eOnce weights are resident in VRAM, inference runs from HBM2e and the PCIe link barely matters. The link speed dominates \u003cem\u003emodel load time\u003c\/em\u003e, host↔device streaming, and any multi-GPU communication. If you load a model once and serve it for hours, ×4 is fine. If you swap models often, stream data continuously, or plan to split a model across cards, pay the €150.\u003c\/div\u003e\n\u003cdiv class=\"rtx-note\"\u003eBoth options come from the same batch of cards — the ×16 conversion is work we carry out in-house before dispatch, which is what the €150 covers. Allow a little extra time on ×16 orders for the conversion and re-testing.\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eSpecifications\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eTechnical data\u003c\/h2\u003e\n\u003ctable class=\"spec\"\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd\u003eGPU\u003c\/td\u003e\n\u003ctd\u003eNVIDIA GA100 — Ampere\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eMemory\u003c\/td\u003e\n\u003ctd\u003e64 GB HBM2e (modified — not a stock configuration)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eCompute\u003c\/td\u003e\n\u003ctd\u003ePartially fused off vs A100 — capacity-oriented, not throughput-oriented\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eInterface\u003c\/td\u003e\n\u003ctd\u003ePCIe 1.0 ×4 or PCIe 1.0 ×16 (select above)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eHost bandwidth\u003c\/td\u003e\n\u003ctd\u003e~1 GB\/s (×4) · ~4 GB\/s (×16)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eDisplay outputs\u003c\/td\u003e\n\u003ctd\u003eNone — compute only\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eCooling\u003c\/td\u003e\n\u003ctd\u003ePassive — requires chassis airflow\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eCondition\u003c\/td\u003e\n\u003ctd\u003eUsed, ex-mining — tested before dispatch\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003c\/tbody\u003e\n\u003c\/table\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eBest for\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eWhere it fits\u003c\/h2\u003e\n\u003cul class=\"tick\"\u003e\n\u003cli\u003eServing one large model that stays resident in VRAM — load once, run for hours.\u003c\/li\u003e\n\u003cli\u003eExperimenting with big models on a budget, where 64 GB for €1,600 is the whole point.\u003c\/li\u003e\n\u003cli\u003eBatch inference jobs that are VRAM-bound rather than transfer-bound.\u003c\/li\u003e\n\u003cli\u003eLearning and development on large-model workflows without datacenter-GPU spend.\u003c\/li\u003e\n\u003c\/ul\u003e\n\u003ch2 class=\"t\" style=\"margin-top: 18px;\"\u003eWhere it does not fit\u003c\/h2\u003e\n\u003cul class=\"tick\"\u003e\n\u003cli\u003eTraining or fine-tuning — PCIe 1.0 and reduced compute both work against you.\u003c\/li\u003e\n\u003cli\u003eMulti-GPU tensor-parallel setups — the interconnect is far too slow.\u003c\/li\u003e\n\u003cli\u003eWorkloads that stream data continuously from host memory or disk.\u003c\/li\u003e\n\u003cli\u003eAnything needing display output, or a card you can rely on for years of production duty.\u003c\/li\u003e\n\u003c\/ul\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eQuestions\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eFAQ\u003c\/h2\u003e\n\u003cdiv class=\"rtx-faq\"\u003e\n\u003cdetails\u003e\n\u003csummary\u003e\u003cbr\u003e\u003c\/summary\u003e\n\u003cp\u003e\u003cbr\u003e\u003c\/p\u003e\n\u003c\/details\u003e\n\u003cdetails\u003e\n\u003csummary\u003eWill my framework see all 64 GB?\u003c\/summary\u003e\n\u003cp\u003eIn our testing the full capacity is addressable, which is the entire reason to buy this card. Compute capability is a different matter — it is partially restricted versus a real A100, and cards vary between units. Tell us your intended workload before ordering and we will give you a straight answer about whether this is the right purchase.\u003c\/p\u003e\n\u003c\/details\u003e\n\u003cdetails\u003e\n\u003csummary\u003eCan I get a fully unlocked card?\u003c\/summary\u003e\n\u003cp\u003eSome units are less restricted than others, but it is genuinely a lottery and we will not sell you a promise we cannot keep. Order on the basis of 64 GB of VRAM at a low price; treat anything beyond that as a bonus.\u003c\/p\u003e\n\u003c\/details\u003e\n\u003cdetails open=\"\"\u003e\n\u003csummary\u003eCan I run several in one machine?\u003c\/summary\u003e\n\u003cp\u003eYou can physically, and each card keeps its own 64 GB, so independent jobs per card work. What does not work well is splitting a single model across cards — PCIe 1.0 makes tensor-parallel communication the bottleneck by a wide margin.\u003c\/p\u003e\n\u003c\/details\u003e\n\u003cdetails open=\"\"\u003e\n\u003csummary\u003eDoes it need special cooling?\u003c\/summary\u003e\n\u003cp\u003eYes. It is a passive card with no fan of its own and expects a server chassis with a proper front-to-back airflow path. It will overheat in a normal desktop case.\u003c\/p\u003e\n\u003c\/details\u003e\n\u003cdetails\u003e\n\u003csummary\u003eHow does this compare with buying a professional card?\u003c\/summary\u003e\n\u003cp\u003eOn VRAM price nothing comes close — roughly €25\/GB here against about €141\/GB for a 96 GB RTX PRO 6000. On reliability, compute, warranty, PCIe bandwidth and resale, the professional card wins on every count. Pick based on which of those matters to you.\u003c\/p\u003e\n\u003c\/details\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-note\"\u003eNot sure this is the right card for your workload? Tell us the model and how you intend to serve it and we will say honestly whether this or a Kentino AI build fits better.\u003c\/div\u003e\n\u003c\/div\u003e\n\u003c\/div\u003e","brand":"NVIDIA","offers":[{"title":"PCIe 1.0 x4","offer_id":53617506648392,"sku":null,"price":2600.0,"currency_code":"EUR","in_stock":true},{"title":"PCIe 1.0 x16","offer_id":53617506681160,"sku":null,"price":2750.0,"currency_code":"EUR","in_stock":true}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/0843\/5479\/3800\/files\/kentino-ai-4-gpu-06_00001.png?v=1786356657"},{"product_id":"nvidia-l4-24-gb-gddr6-low-profile-passive","title":"NVIDIA L4 24 GB GDDR6 — Low Profile Passive","description":"\u003cstyle\u003e\n.rtx{--bl:#334fb4;--dk:#26408f;--ink:#17171a;--soft:#f3f7fd;--muted:#6c6660;--line:#dce6f5;font-family:-apple-system,BlinkMacSystemFont,\"Segoe UI\",Roboto,Helvetica,Arial,sans-serif;color:var(--ink);line-height:1.6;font-size:15px;}\n.rtx *{box-sizing:border-box;}\n.rtx-head{background:var(--ink);color:#fff;border-radius:14px;padding:30px 28px;position:relative;overflow:hidden;}\n.rtx-head:before{content:\"\";position:absolute;inset:0;background:radial-gradient(120% 100% at 100% 0,rgba(51,79,180,.32),transparent 55%);}\n.rtx-head\u003e*{position:relative;}\n.rtx-head h2{font-size:clamp(22px,3.2vw,30px);font-weight:800;margin:0 0 10px;color:#fff !important;}\n.rtx-head p{color:#d7d2cb;max-width:64ch;margin:0;}\n.rtx-badge{display:inline-block;background:var(--bl);color:#fff !important;font-size:12px;font-weight:700;letter-spacing:.08em;text-transform:uppercase;padding:5px 11px;border-radius:999px;margin-bottom:14px;}\n.rtx-hi{display:grid;grid-template-columns:repeat(4,1fr);gap:1px;background:var(--line);border:1px solid var(--line);border-radius:12px;overflow:hidden;margin:18px 0 6px;}\n.rtx-hi div{background:#fff;padding:18px 12px;text-align:center;}\n.rtx-hi .v{font-size:20px;font-weight:800;}\n.rtx-hi .v small{color:var(--bl);font-size:12px;}\n.rtx-hi .k{font-size:11.5px;color:var(--muted);margin-top:4px;}\n.rtx-sec{padding:24px 0 4px;}\n.rtx-ey{font-size:12px;font-weight:700;letter-spacing:.16em;text-transform:uppercase;color:var(--dk);}\n.rtx h2.t{font-size:clamp(19px,2.4vw,24px);font-weight:800;margin:6px 0 12px;}\n.rtx p.lead{font-size:16px;color:#3a3a36;}\n.rtx table{width:100%;border-collapse:collapse;font-size:14px;margin:6px 0;}\n.rtx table.spec td{padding:9px 12px;border-bottom:1px solid var(--line);}\n.rtx table.spec tr:nth-child(odd) td{background:#f6f9fc;}\n.rtx table.spec td:first-child{width:42%;color:var(--muted);font-weight:600;}\n.rtx ul.tick{list-style:none;margin:6px 0;padding:0;}\n.rtx ul.tick li{position:relative;padding:7px 0 7px 26px;border-bottom:1px solid var(--line);font-size:14.5px;}\n.rtx ul.tick li:last-child{border-bottom:none;}\n.rtx ul.tick li:before{content:\"\";position:absolute;left:4px;top:14px;width:9px;height:9px;border-radius:2px;background:var(--bl);}\n.rtx-note{background:var(--soft);border-left:3px solid var(--bl);padding:13px 16px;border-radius:0 8px 8px 0;font-size:13.5px;color:#4a4641;margin-top:14px;}\n.rtx-faq details{border-bottom:1px solid var(--line);padding:14px 0;}\n.rtx-faq summary{font-weight:700;cursor:pointer;list-style:none;display:flex;justify-content:space-between;font-size:15px;}\n.rtx-faq summary:after{content:\"+\";color:var(--bl);font-weight:700;}\n.rtx-faq details[open] summary:after{content:\"–\";}\n.rtx-faq p{margin:9px 0 0;color:var(--muted);font-size:14px;}\n@media(max-width:760px){.rtx-hi{grid-template-columns:repeat(2,1fr);}}\n\u003c\/style\u003e\n\u003cdiv class=\"rtx\"\u003e\n\u003cdiv class=\"rtx-head\"\u003e\n\u003cspan class=\"rtx-badge\"\u003eAda Lovelace · 24 GB · 72 W · Low Profile\u003c\/span\u003e\n\u003ch2\u003eNVIDIA L4 24 GB — the 72 W inference card\u003c\/h2\u003e\n\u003cp\u003eA single-slot, low-profile, passively cooled datacenter GPU that draws just 72 W straight from the PCIe slot — no power cable, no extra airflow design. The densest way to add 24 GB of inference capacity to a server.\u003c\/p\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-hi\"\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003e24 \u003csmall\u003eGB\u003c\/small\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003eGDDR6 ECC VRAM\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003e7,680\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003eCUDA cores\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003e72 \u003csmall\u003eW\u003c\/small\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003eSlot-powered\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003e1 \u003csmall\u003eslot\u003c\/small\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003eLow profile, passive\u003c\/div\u003e\n\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eOverview\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eDensity and efficiency, not peak throughput\u003c\/h2\u003e\n\u003cp class=\"lead\"\u003eThe L4 is built around a different priority to the big accelerator cards: performance per watt and per slot. At 72 W it needs no supplementary power connector at all, and being single-slot low-profile it fits chassis that physically cannot accept a full-height card. That combination is what makes it the practical choice for packing many independent inference workloads into one server, or for adding capable AI acceleration to compact and edge systems where power and space are the binding constraints. It is also a strong video card — hardware encode and decode including AV1 — which makes it a common pick for transcoding alongside inference.\u003c\/p\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eSpecifications\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eTechnical data\u003c\/h2\u003e\n\u003ctable class=\"spec\"\u003e\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd\u003eGPU\u003c\/td\u003e\n\u003ctd\u003eNVIDIA AD104 — Ada Lovelace\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003ePart number\u003c\/td\u003e\n\u003ctd\u003eTCSL4PCIE-PB\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eMemory\u003c\/td\u003e\n\u003ctd\u003e24 GB GDDR6 with ECC\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eMemory bandwidth\u003c\/td\u003e\n\u003ctd\u003e~300 GB\/s\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eCUDA cores\u003c\/td\u003e\n\u003ctd\u003e7,680\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eTensor cores\u003c\/td\u003e\n\u003ctd\u003e240 (4th gen) · ~242 TOPS INT8\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eRT cores\u003c\/td\u003e\n\u003ctd\u003e60 (3rd gen)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eInterface\u003c\/td\u003e\n\u003ctd\u003ePCIe 4.0 ×16\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eBoard power\u003c\/td\u003e\n\u003ctd\u003e72 W — drawn from the slot, no power connector\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eForm factor\u003c\/td\u003e\n\u003ctd\u003eSingle-slot, low profile, full-length\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eCooling\u003c\/td\u003e\n\u003ctd\u003ePassive — requires chassis airflow\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eDisplay outputs\u003c\/td\u003e\n\u003ctd\u003eNone — compute only\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eVideo engines\u003c\/td\u003e\n\u003ctd\u003eHardware encode \/ decode incl. AV1\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eWarranty\u003c\/td\u003e\n\u003ctd\u003e36 months\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003c\/tbody\u003e\u003c\/table\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eBest for\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eWhere it fits\u003c\/h2\u003e\n\u003cul class=\"tick\"\u003e\n\u003cli\u003eServing small and mid-size models — 24 GB comfortably holds a quantised 7B–13B model.\u003c\/li\u003e\n\u003cli\u003eMany-GPU inference servers: at 72 W a card, you can fit several without redesigning power or cooling.\u003c\/li\u003e\n\u003cli\u003eLow-profile and compact chassis that cannot take a full-height dual-slot card.\u003c\/li\u003e\n\u003cli\u003eVideo transcoding pipelines, including AV1, alongside inference on the same card.\u003c\/li\u003e\n\u003cli\u003eEdge and on-premise deployments where total power draw is capped.\u003c\/li\u003e\n\u003c\/ul\u003e\n\u003cdiv class=\"rtx-note\"\u003eThe L4 is a capacity-and-efficiency card, not a throughput champion. For heavy fine-tuning, large-model work or maximum tokens per second, an RTX PRO 6000 or a multi-GPU build is the better spend — tell us the workload and we will point you at the right one.\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eQuestions\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eFAQ\u003c\/h2\u003e\n\u003cdiv class=\"rtx-faq\"\u003e\n\u003cdetails open\u003e\u003csummary\u003eDoes it need a power cable?\u003c\/summary\u003e\u003cp\u003eNo. The whole card runs inside the PCIe slot's 75 W budget, so there is no 8-pin or 12-pin connector to route. That is a large part of its appeal in dense builds — it removes PSU cabling as a constraint entirely.\u003c\/p\u003e\u003c\/details\u003e\n\u003cdetails\u003e\u003csummary\u003eWill it cool itself?\u003c\/summary\u003e\u003cp\u003eNo. Like other datacenter cards it is passive and relies on chassis airflow. In a server with a proper front-to-back path this is ideal; in a quiet desktop case it will overheat. Ask us if you are unsure about your chassis.\u003c\/p\u003e\u003c\/details\u003e\n\u003cdetails\u003e\u003csummary\u003eWhat model sizes fit in 24 GB?\u003c\/summary\u003e\u003cp\u003eA quantised 7B–13B model fits with room for context. A 70B model does not — that needs roughly 40 GB at 4-bit, so look at a 96 GB card or a multi-GPU configuration instead.\u003c\/p\u003e\u003c\/details\u003e\n\u003cdetails\u003e\u003csummary\u003eHow does it compare with the L40?\u003c\/summary\u003e\u003cp\u003eThe L40 has 48 GB and far more compute, at around 300 W and a full-height dual-slot form factor. The L4 trades that throughput for 72 W, one slot and low profile. Choose the L4 when power and density decide the build, the L40 when you need the performance.\u003c\/p\u003e\u003c\/details\u003e\n\u003cdetails\u003e\u003csummary\u003eCan I put several in one server?\u003c\/summary\u003e\u003cp\u003eYes, and this is where the L4 is at its best — several cards at 72 W each stay within budgets that would be impossible with high-power GPUs. Kentino builds multi-L4 inference servers if you would rather buy the finished machine.\u003c\/p\u003e\u003c\/details\u003e\n\u003cdetails\u003e\u003csummary\u003eWhat is the lead time?\u003c\/summary\u003e\u003cp\u003eSourced to order, typically 10–21 days. Ask us to confirm current availability before planning around a date.\u003c\/p\u003e\u003c\/details\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-note\"\u003eWant it in a finished machine? Kentino builds, benchmarks and commissions multi-L4 inference servers — ask us for a configured build.\u003c\/div\u003e\n\u003c\/div\u003e\n\u003c\/div\u003e","brand":"NVIDIA","offers":[{"title":"Default Title","offer_id":53618834899272,"sku":"TCSL4PCIE-PB","price":2657.42,"currency_code":"EUR","in_stock":true}]},{"product_id":"pcie-4-0-16-riser-cable-40-cm-90-angled-right-out-left-in","title":"PCIe 4.0 ×16 Riser Cable 40 cm — 90° Angled, Right-Out \/ Left-In","description":"\u003cstyle\u003e\n.rtx{--bl:#334fb4;--dk:#26408f;--ink:#17171a;--soft:#f3f7fd;--muted:#6c6660;--line:#dce6f5;font-family:-apple-system,BlinkMacSystemFont,\"Segoe UI\",Roboto,Helvetica,Arial,sans-serif;color:var(--ink);line-height:1.6;font-size:15px;}\n.rtx *{box-sizing:border-box;}\n.rtx-head{background:var(--ink);color:#fff;border-radius:14px;padding:30px 28px;position:relative;overflow:hidden;}\n.rtx-head:before{content:\"\";position:absolute;inset:0;background:radial-gradient(120% 100% at 100% 0,rgba(51,79,180,.32),transparent 55%);}\n.rtx-head\u003e*{position:relative;}\n.rtx-head h2{font-size:clamp(22px,3.2vw,30px);font-weight:800;margin:0 0 10px;color:#fff !important;}\n.rtx-head p{color:#d7d2cb;max-width:64ch;margin:0;}\n.rtx-badge{display:inline-block;background:var(--bl);color:#fff !important;font-size:12px;font-weight:700;letter-spacing:.08em;text-transform:uppercase;padding:5px 11px;border-radius:999px;margin-bottom:14px;}\n.rtx-hi{display:grid;grid-template-columns:repeat(4,1fr);gap:1px;background:var(--line);border:1px solid var(--line);border-radius:12px;overflow:hidden;margin:18px 0 6px;}\n.rtx-hi div{background:#fff;padding:18px 12px;text-align:center;}\n.rtx-hi .v{font-size:20px;font-weight:800;}\n.rtx-hi .v small{color:var(--bl);font-size:12px;}\n.rtx-hi .k{font-size:11.5px;color:var(--muted);margin-top:4px;}\n.rtx-sec{padding:24px 0 4px;}\n.rtx-ey{font-size:12px;font-weight:700;letter-spacing:.16em;text-transform:uppercase;color:var(--dk);}\n.rtx h2.t{font-size:clamp(19px,2.4vw,24px);font-weight:800;margin:6px 0 12px;}\n.rtx p.lead{font-size:16px;color:#3a3a36;}\n.rtx table{width:100%;border-collapse:collapse;font-size:14px;margin:6px 0;}\n.rtx table.spec td{padding:9px 12px;border-bottom:1px solid var(--line);}\n.rtx table.spec tr:nth-child(odd) td{background:#f6f9fc;}\n.rtx table.spec td:first-child{width:42%;color:var(--muted);font-weight:600;}\n.rtx ul.tick{list-style:none;margin:6px 0;padding:0;}\n.rtx ul.tick li{position:relative;padding:7px 0 7px 26px;border-bottom:1px solid var(--line);font-size:14.5px;}\n.rtx ul.tick li:last-child{border-bottom:none;}\n.rtx ul.tick li:before{content:\"\";position:absolute;left:4px;top:14px;width:9px;height:9px;border-radius:2px;background:var(--bl);}\n.rtx-note{background:var(--soft);border-left:3px solid var(--bl);padding:13px 16px;border-radius:0 8px 8px 0;font-size:13.5px;color:#4a4641;margin-top:14px;}\n.rtx-faq details{border-bottom:1px solid var(--line);padding:14px 0;}\n.rtx-faq summary{font-weight:700;cursor:pointer;list-style:none;display:flex;justify-content:space-between;font-size:15px;}\n.rtx-faq summary:after{content:\"+\";color:var(--bl);font-weight:700;}\n.rtx-faq details[open] summary:after{content:\"–\";}\n.rtx-faq p{margin:9px 0 0;color:var(--muted);font-size:14px;}\n@media(max-width:760px){.rtx-hi{grid-template-columns:repeat(2,1fr);}}\n\u003c\/style\u003e\n\u003cdiv class=\"rtx\"\u003e\n\u003cdiv class=\"rtx-head\"\u003e\n\u003cspan class=\"rtx-badge\"\u003ePCIe 4.0 ×16 · 40 cm · Passive · Shielded\u003c\/span\u003e\n\u003ch2\u003ePCIe 4.0 ×16 Riser Cable — 40 cm, 90° Angled\u003c\/h2\u003e\n\u003cp\u003eA full-bandwidth Gen4 ×16 riser with aluminium-foil-shielded differential pairs and a braided sleeve. Passive, so it needs no power and stays invisible to the system. This is the riser we use in our own multi-GPU builds.\u003c\/p\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-hi\"\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003e4.0 \u003csmall\u003eGen\u003c\/small\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003ePCIe generation\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003e×16\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003eFull lane width\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003e40 \u003csmall\u003ecm\u003c\/small\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003eCable length\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv\u003e\n\u003cdiv class=\"v\"\u003e90\u003csmall\u003e°\u003c\/small\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"k\"\u003eAngled connectors\u003c\/div\u003e\n\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eOrientation\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eRight-out \/ left-in — check this before you order\u003c\/h2\u003e\n\u003cp class=\"lead\"\u003eThis is the \u003cstrong\u003eright-out \/ left-in\u003c\/strong\u003e version. The angled connectors are handed, so orientation is fixed and it is the one detail people get wrong. Work out which way the cable needs to leave the slot and enter the card in your chassis before ordering — if you need the mirrored version, ask us and we will supply that instead.\u003c\/p\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eWhy the shielding matters\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eSignal integrity at Gen4 speeds\u003c\/h2\u003e\n\u003cp class=\"lead\"\u003ePCIe 4.0 runs at 16 GT\/s per lane — roughly 32 GB\/s per direction across sixteen lanes. At that rate a riser stops being a cable and becomes a transmission line: crosstalk and loss decide whether the link trains at Gen4 or quietly falls back. This one uses aluminium-foil-shielded differential pairs to keep the pairs isolated, which is what makes a run of this length viable at full Gen4 rather than Gen3.\u003c\/p\u003e\n\u003cdiv class=\"rtx-note\"\u003ePractical note: 40 cm is a long run for a passive Gen4 riser. On most boards it trains at Gen4 without complaint. If yours is marginal — a heavily loaded platform, or a slot fed through a switch — you can pin the slot to Gen3 in BIOS and keep full stability with no other change. Tell us your board and GPU count and we will tell you what to expect.\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eSpecifications\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eTechnical data\u003c\/h2\u003e\n\u003ctable class=\"spec\"\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd\u003eInterface\u003c\/td\u003e\n\u003ctd\u003ePCIe 4.0 ×16 (backward compatible with 3.0 \/ 2.0)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eBandwidth\u003c\/td\u003e\n\u003ctd\u003eFull ×16 — up to ~32 GB\/s per direction at Gen4\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eLength\u003c\/td\u003e\n\u003ctd\u003e40 cm\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eCable type\u003c\/td\u003e\n\u003ctd\u003eLinear, braided mesh sleeve\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eShielding\u003c\/td\u003e\n\u003ctd\u003eAluminium foil over differential pairs\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eConnectors\u003c\/td\u003e\n\u003ctd\u003e90° angled — right-out \/ left-in\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eType\u003c\/td\u003e\n\u003ctd\u003ePassive — no redriver, no external power\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eCompliance\u003c\/td\u003e\n\u003ctd\u003eRoHS · UL\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003c\/tbody\u003e\n\u003c\/table\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eBest for\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eWhere it fits\u003c\/h2\u003e\n\u003cul class=\"tick\"\u003e\n\u003cli\u003eMulti-GPU AI servers where cards sit off-board on a riser plate.\u003c\/li\u003e\n\u003cli\u003eTight 4U and rack enclosures — the 90° exit removes the cable loop that blocks airflow.\u003c\/li\u003e\n\u003cli\u003eVertical GPU mounts in tower and workstation builds.\u003c\/li\u003e\n\u003cli\u003eRelocating a card away from a hot neighbour or a blocked slot.\u003c\/li\u003e\n\u003cli\u003eAny Gen4 GPU where you do not want to give up half your bandwidth to a mining-grade extender.\u003c\/li\u003e\n\u003c\/ul\u003e\n\u003cdiv class=\"rtx-note\"\u003eThis is not a mining riser. The ×1 powered USB extenders sold for GPU rigs throttle a modern GPU to a fraction of its bandwidth. This carries all sixteen lanes at Gen4 — which is the whole point if the card is doing AI work.\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-sec\"\u003e\n\u003cdiv class=\"rtx-ey\"\u003eQuestions\u003c\/div\u003e\n\u003ch2 class=\"t\"\u003eFAQ\u003c\/h2\u003e\n\u003cdiv class=\"rtx-faq\"\u003e\n\u003cdetails open=\"\"\u003e\n\u003csummary\u003eDoes it need power?\u003c\/summary\u003e\n\u003cp\u003eNo. It is a passive riser — it carries the slot's own signals and power straight through, with no connector to plug in and nothing for the system to configure.\u003c\/p\u003e\n\u003c\/details\u003e\n\u003cdetails open=\"\"\u003e\n\u003csummary\u003eWill I lose performance versus a direct slot?\u003c\/summary\u003e\n\u003cp\u003eWhen the link trains at Gen4 ×16, no — you have the same lane count and the same rate as the slot itself. The only failure mode worth knowing about is a marginal link dropping to Gen3, which halves per-lane rate. The shielding is there specifically to avoid that.\u003c\/p\u003e\n\u003c\/details\u003e\n\u003cdetails\u003e\n\u003csummary\u003eCan I use it with a PCIe 5.0 board or card?\u003c\/summary\u003e\n\u003cp\u003ePhysically yes, and it will work — but the link negotiates down to Gen4, which is the cable's rating. If you need full Gen5 on a riser, that is different hardware; ask us.\u003c\/p\u003e\n\u003c\/details\u003e\n\u003cdetails\u003e\n\u003csummary\u003eWhich orientation do I need?\u003c\/summary\u003e\n\u003cp\u003eThis one is right-out \/ left-in. Look at where the cable has to run in your case: if it needs to leave the slot the other way, you want the mirrored part. We stock both — just ask before ordering if you are unsure.\u003c\/p\u003e\n\u003c\/details\u003e\n\u003cdetails\u003e\n\u003csummary\u003eDo you have other lengths?\u003c\/summary\u003e\n\u003cp\u003eYes, 30 cm and 50 cm as well. Shorter is always better for signal integrity, so take the shortest run that physically reaches.\u003c\/p\u003e\n\u003c\/details\u003e\n\u003c\/div\u003e\n\u003cdiv class=\"rtx-note\"\u003eBuilding a multi-GPU machine? Kentino assembles, benchmarks and commissions complete AI servers and workstations — risers, plates, power and airflow all specified together. Browse the \u003ca href=\"\/collections\/ai-servers\"\u003eAI Servers collection\u003c\/a\u003e.\u003c\/div\u003e\n\u003c\/div\u003e\n\u003c\/div\u003e","brand":"Kentino","offers":[{"title":"Default Title","offer_id":53860489003336,"sku":null,"price":55.2,"currency_code":"EUR","in_stock":true}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/0843\/5479\/3800\/files\/Screenshot2026-08-19at18-33-34PCIe4.0x16ExternalGPURiserCable30-50cm.png?v=1787219195"}],"url":"https:\/\/kentino.se\/collections\/components.oembed","provider":"Kentino","version":"1.0","type":"link"}