{"product_id":"nvidia-900-2g193-0100-000-v-l4-24gb-gddr6-72w-fhhl-pcie-gpu","title":"NVIDIA 900-2G193-0100-000-V L4 24GB GDDR6 72W FHHL PCIE GPU","description":"\u003cdiv class=\"c4-rich\"\u003e\n\u003cp\u003eThe NVIDIA 900-2G193-0100-000-V is an L4 Tensor Core GPU with 24GB of GDDR6 memory, built on the Ada Lovelace architecture for data center servers. At a 72W maximum power draw in a single-slot PCIe card, it brings AI inference, video transcoding, and graphics acceleration to mainstream rack servers, which can hold one to eight cards depending on the platform.\u003c\/p\u003e\n\u003ch3 style=\"margin:1.4em 0 0.35em\"\u003eAt a glance\u003c\/h3\u003e\n\u003cul\u003e\n\u003cli\u003eNVIDIA L4 with 24GB GDDR6, 300GB\/s\u003c\/li\u003e\n\u003cli\u003e72W max power, single-slot PCIe Gen4 x16\u003c\/li\u003e\n\u003cli\u003eAI inference, video, and graphics acceleration\u003c\/li\u003e\n\u003cli\u003eRefurbished, tested twice, 1-year advanced replacement warranty\u003c\/li\u003e\n\u003c\/ul\u003e\n\u003ch3 style=\"margin:1.4em 0 0.35em\"\u003eDeploying L4 GPUs across several servers?\u003c\/h3\u003e\n\u003cp\u003eTell a Core 4 sales engineer how many cards your inference or video project needs. If you need more than we have listed, we can source additional quantities. \u003ca href=\"\/pages\/contact?part_number=900-2G193-0100-000-V\u0026amp;additional_requirements=Product%3A%20NVIDIA%20900-2G193-0100-000-V%20L4%2024GB%20GDDR6%2072W%20FHHL%20PCIE%20GPU%20%28https%3A%2F%2Fwww.core4solutions.com%2Fproducts%2Fnvidia-900-2g193-0100-000-v-l4-24gb-gddr6-72w-fhhl-pcie-gpu%29\"\u003e\u003cstrong\u003eTalk to a sales engineer\u003c\/strong\u003e\u003c\/a\u003e or call 1-855-55-CORE4.\u003c\/p\u003e\n\u003ch3 style=\"margin:1.4em 0 0.35em\"\u003eTechnical specifications\u003c\/h3\u003e\n\u003ctable style=\"border-collapse:collapse;width:100%;margin:0 0 1em\"\u003e\u003ctbody\u003e\n\u003ctr\u003e\n\u003cth scope=\"row\" style=\"text-align:left;vertical-align:top;font-weight:600;padding:6px 16px 6px 0;border-bottom:1px solid rgba(128,128,128,.2)\"\u003eGPU\u003c\/th\u003e\n\u003ctd style=\"padding:6px 0;border-bottom:1px solid rgba(128,128,128,.2)\"\u003eNVIDIA L4 Tensor Core (Ada Lovelace architecture)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003cth scope=\"row\" style=\"text-align:left;vertical-align:top;font-weight:600;padding:6px 16px 6px 0;border-bottom:1px solid rgba(128,128,128,.2)\"\u003eGPU memory\u003c\/th\u003e\n\u003ctd style=\"padding:6px 0;border-bottom:1px solid rgba(128,128,128,.2)\"\u003e24GB GDDR6\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003cth scope=\"row\" style=\"text-align:left;vertical-align:top;font-weight:600;padding:6px 16px 6px 0;border-bottom:1px solid rgba(128,128,128,.2)\"\u003eMemory bandwidth\u003c\/th\u003e\n\u003ctd style=\"padding:6px 0;border-bottom:1px solid rgba(128,128,128,.2)\"\u003e300GB\/s\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003cth scope=\"row\" style=\"text-align:left;vertical-align:top;font-weight:600;padding:6px 16px 6px 0;border-bottom:1px solid rgba(128,128,128,.2)\"\u003eMax thermal design power\u003c\/th\u003e\n\u003ctd style=\"padding:6px 0;border-bottom:1px solid rgba(128,128,128,.2)\"\u003e72W\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003cth scope=\"row\" style=\"text-align:left;vertical-align:top;font-weight:600;padding:6px 16px 6px 0;border-bottom:1px solid rgba(128,128,128,.2)\"\u003eInterface\u003c\/th\u003e\n\u003ctd style=\"padding:6px 0;border-bottom:1px solid rgba(128,128,128,.2)\"\u003ePCIe Gen4 x16, 64GB\/s\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003cth scope=\"row\" style=\"text-align:left;vertical-align:top;font-weight:600;padding:6px 16px 6px 0;border-bottom:1px solid rgba(128,128,128,.2)\"\u003eSlot width\u003c\/th\u003e\n\u003ctd style=\"padding:6px 0;border-bottom:1px solid rgba(128,128,128,.2)\"\u003eSingle slot\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003cth scope=\"row\" style=\"text-align:left;vertical-align:top;font-weight:600;padding:6px 16px 6px 0;border-bottom:1px solid rgba(128,128,128,.2)\"\u003eFP32\u003c\/th\u003e\n\u003ctd style=\"padding:6px 0;border-bottom:1px solid rgba(128,128,128,.2)\"\u003e30.3 teraFLOPS\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003cth scope=\"row\" style=\"text-align:left;vertical-align:top;font-weight:600;padding:6px 16px 6px 0;border-bottom:1px solid rgba(128,128,128,.2)\"\u003eFP8 Tensor Core\u003c\/th\u003e\n\u003ctd style=\"padding:6px 0;border-bottom:1px solid rgba(128,128,128,.2)\"\u003e485 teraFLOPS (with sparsity)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003cth scope=\"row\" style=\"text-align:left;vertical-align:top;font-weight:600;padding:6px 16px 6px 0;border-bottom:1px solid rgba(128,128,128,.2)\"\u003eINT8 Tensor Core\u003c\/th\u003e\n\u003ctd style=\"padding:6px 0;border-bottom:1px solid rgba(128,128,128,.2)\"\u003e485 TOPS (with sparsity)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003cth scope=\"row\" style=\"text-align:left;vertical-align:top;font-weight:600;padding:6px 16px 6px 0;border-bottom:1px solid rgba(128,128,128,.2)\"\u003eMedia engines\u003c\/th\u003e\n\u003ctd style=\"padding:6px 0;border-bottom:1px solid rgba(128,128,128,.2)\"\u003e2 NVENC, 4 NVDEC, 4 JPEG decoders\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003c\/tbody\u003e\u003c\/table\u003e\n\u003ch3 style=\"margin:1.4em 0 0.35em\"\u003eCompatibility\u003c\/h3\u003e\n\u003cp\u003eThe L4 installs in a PCIe Gen4 x16 slot and is offered in partner and NVIDIA-Certified Systems configured with one to eight GPUs. Its single-slot width and 72W power envelope let it fit servers that cannot take larger, higher-wattage accelerators.\u003c\/p\u003e\n\u003cp\u003eGPU support, slot placement, and cooling requirements vary by server model, so check that your platform lists the L4 as a supported option.\u003c\/p\u003e\n\u003ch3 style=\"margin:1.4em 0 0.35em\"\u003eCondition\u003c\/h3\u003e\n\u003cp\u003eThis NVIDIA L4 is refurbished: a fully functioning card that came out of a working production environment or from overstock, returns, or off-lease equipment. The vast majority of refurbished parts were never broken or repaired.\u003c\/p\u003e\n\u003cp\u003eCore 4 tests every part when it arrives and again before it ships, and anything that does not pass is repaired before it is sold. Because these cards have already been installed and burned in, early failures are weeded out, and Core 4 sees a lower dead-on-arrival rate than with new parts straight from the factory.\u003c\/p\u003e\n\u003cp\u003eThe card is covered by Core 4's 1-year advanced replacement warranty: if it fails within a year, a replacement ships right away, before the failed unit comes back.\u003c\/p\u003e\n\u003ch3 style=\"margin:1.4em 0 0.35em\"\u003eDatasheets and documentation\u003c\/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003ca href=\"https:\/\/resources.nvidia.com\/en-us-gpu-resources\/l4-tensor-datasheet\" target=\"_blank\" rel=\"noopener\"\u003eNVIDIA L4 GPU datasheet\u003c\/a\u003e\u003c\/li\u003e\n\u003cli\u003e\u003ca href=\"https:\/\/www.nvidia.com\/en-us\/data-center\/l4\" target=\"_blank\" rel=\"noopener\"\u003eNVIDIA L4 Tensor Core GPU product page\u003c\/a\u003e\u003c\/li\u003e\n\u003cli\u003e\u003ca href=\"https:\/\/www.nvidia.com\/content\/dam\/en-zz\/Solutions\/Data-Center\/l4\/PB-11316-001_v01.pdf\" target=\"_blank\" rel=\"noopener\"\u003eNVIDIA L4 product brief (PB-11316-001)\u003c\/a\u003e\u003c\/li\u003e\n\u003cli\u003e\u003ca href=\"https:\/\/www.cisco.com\/c\/dam\/en\/us\/products\/collateral\/servers-unified-computing\/ucs-c-series-rack-servers\/nvidia-l4-gpu.pdf\" target=\"_blank\" rel=\"noopener\"\u003eNVIDIA L4 datasheet (Cisco UCS C-Series collateral)\u003c\/a\u003e\u003c\/li\u003e\n\u003c\/ul\u003e\n\u003ch3 style=\"margin:1.4em 0 0.35em\"\u003eCommon questions\u003c\/h3\u003e\n\u003cp\u003e\u003cstrong\u003eDoes the L4 need an auxiliary power cable?\u003c\/strong\u003e No. Its 72W maximum draw falls within what a PCIe x16 slot supplies, so the card runs on slot power alone with no separate GPU power connector.\u003c\/p\u003e\n\u003cp\u003e\u003cstrong\u003eWhat workloads is the L4 suited for?\u003c\/strong\u003e It targets AI inference, video processing, and graphics in standard servers. Fourth-generation Tensor Cores support FP8, FP16, BFLOAT16, TF32, and INT8, and third-generation RT Cores handle ray-traced graphics. The 24GB of memory holds mid-sized inference models.\u003c\/p\u003e\n\u003cp\u003e\u003cstrong\u003eHow much video work can one card handle?\u003c\/strong\u003e Each L4 has 2 NVENC encoders, 4 NVDEC decoders, and 4 JPEG decoders, which suits it to transcoding, streaming, and video analytics pipelines running many streams in parallel.\u003c\/p\u003e\n\u003cp\u003e\u003cstrong\u003eAre the Tensor Core figures with or without sparsity?\u003c\/strong\u003e The FP8 and INT8 figures of 485 teraFLOPS and 485 TOPS, along with the FP16 and BFLOAT16 figure of 242 teraFLOPS and the TF32 figure of 120 teraFLOPS, are with sparsity. Dense throughput is roughly half. FP32 is 30.3 teraFLOPS.\u003c\/p\u003e\n\u003c\/div\u003e","brand":"NVIDIA","offers":[{"title":"Default Title","offer_id":67927246700779,"sku":"nvidia 900-2g193-0100-000-v refurbished","price":4399.0,"currency_code":"USD","in_stock":true}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/0843\/8769\/2779\/files\/1_080a703d-1d24-4c35-8b6f-5d9ce2d9dcb3.jpg?v=1793191065","url":"https:\/\/www.core4solutions.com\/products\/nvidia-900-2g193-0100-000-v-l4-24gb-gddr6-72w-fhhl-pcie-gpu","provider":"Core 4 Solutions","version":"1.0","type":"link"}