{"product_id":"k-node-a100-8x-sxm4-640gb","title":"Kentino AI Node A100 — 8× NVIDIA A100 SXM4 80 GB, 640 GB HBM2e","description":"\u003cdiv style=\"font-family: -apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,sans-serif; line-height: 1.7; color: #1a1a1a;\"\u003e\n\u003c!-- IMAGE PLACEHOLDER 1: 4U node front, white background --\u003e \u003c!-- IMAGE PLACEHOLDER 2: HGX A100 baseboard, 8 SXM4 modules --\u003e \u003c!-- IMAGE PLACEHOLDER 3: rear PSU bank --\u003e \u003c!-- IMAGE PLACEHOLDER 4: rack-installed --\u003e\n\u003cdiv style=\"background: linear-gradient(135deg,#0d0d0d 0%,#1a1a2e 100%); color: #fff; padding: 32px; border-radius: 12px; margin-bottom: 32px;\"\u003e\n\u003cp style=\"font-size: 18px; margin: 0 0 20px 0; color: #ccc;\"\u003eKentino AI Node A100\u003c\/p\u003e\n\u003cp style=\"font-size: 28px; font-weight: bold; margin: 0 0 16px 0; line-height: 1.3;\"\u003e640 GB HBM2e, NVSwitch-Pooled\u003cbr\u003e8x A100 SXM4 80 GB | 4,992 INT8 TOPS\u003c\/p\u003e\n\u003cdiv style=\"display: flex; gap: 16px; flex-wrap: wrap; margin-top: 24px;\"\u003e\n\u003cdiv style=\"background: rgba(51,79,180,0.12); border: 1px solid #334fb4; border-radius: 8px; padding: 14px 16px; text-align: center; flex: 1; min-width: 110px;\"\u003e\n\u003cdiv style=\"font-size: 24px; font-weight: 800; color: #334fb4;\"\u003e640 GB\u003c\/div\u003e\n\u003cdiv style=\"font-size: 11px; color: #555; text-transform: uppercase; letter-spacing: 1px;\"\u003eHBM2e pool\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv style=\"background: rgba(51,79,180,0.12); border: 1px solid #334fb4; border-radius: 8px; padding: 14px 16px; text-align: center; flex: 1; min-width: 110px;\"\u003e\n\u003cdiv style=\"font-size: 24px; font-weight: 800; color: #334fb4;\"\u003e2.04 TB\/s\u003c\/div\u003e\n\u003cdiv style=\"font-size: 11px; color: #555; text-transform: uppercase; letter-spacing: 1px;\"\u003eper-GPU bandwidth\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv style=\"background: rgba(51,79,180,0.12); border: 1px solid #334fb4; border-radius: 8px; padding: 14px 16px; text-align: center; flex: 1; min-width: 110px;\"\u003e\n\u003cdiv style=\"font-size: 24px; font-weight: 800; color: #334fb4;\"\u003e600 GB\/s\u003c\/div\u003e\n\u003cdiv style=\"font-size: 11px; color: #555; text-transform: uppercase; letter-spacing: 1px;\"\u003eany-to-any NVLink\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cdiv style=\"background: rgba(51,79,180,0.12); border: 1px solid #334fb4; border-radius: 8px; padding: 14px 16px; text-align: center; flex: 1; min-width: 110px;\"\u003e\n\u003cdiv style=\"font-size: 24px; font-weight: 800; color: #334fb4;\"\u003e3.2 kW\u003c\/div\u003e\n\u003cdiv style=\"font-size: 11px; color: #555; text-transform: uppercase; letter-spacing: 1px;\"\u003eGPU draw\u003c\/div\u003e\n\u003c\/div\u003e\n\u003c\/div\u003e\n\u003cp style=\"margin-top: 20px; font-size: 15px; color: #aaa;\"\u003eThe only node tier with a sane cost per gigabyte of VRAM.\u003c\/p\u003e\n\u003c\/div\u003e\n\u003cp style=\"font-size: 17px; color: #333; margin-bottom: 24px;\"\u003eAn 8-GPU node carrying eight NVIDIA A100 SXM4 80 GB modules on an NVLink\/NVSwitch baseboard: 640 GB of HBM2e at 2.04 TB\/s per GPU with 600 GB\/s any-to-any GPU bandwidth. Ampere is two generations back, and for a large class of work that is exactly why you want it - the same 640 GB pool as a Hopper H100 node, at substantially less money and 2.4 kW less heat. If your models run at bf16 or Q4 and you are not going to use fp8, you should not be paying for it.\u003c\/p\u003e\n\u003ch2 style=\"font-size: 22px; font-weight: bold; margin: 40px 0 16px 0; padding-bottom: 8px; border-bottom: 3px solid #334fb4;\"\u003eHardware\u003c\/h2\u003e\n\u003ctable style=\"width: 100%; border-collapse: collapse; margin-bottom: 24px; font-size: 15px;\"\u003e\n\u003cthead\u003e\n\u003ctr style=\"background: #0d0d0d; color: #fff;\"\u003e\n\u003cth style=\"padding: 12px 16px; text-align: left;\"\u003eComponent\u003c\/th\u003e\n\u003cth style=\"padding: 12px 16px; text-align: left;\"\u003eDetail\u003c\/th\u003e\n\u003c\/tr\u003e\n\u003c\/thead\u003e\n\u003ctbody\u003e\n\u003ctr style=\"background: #f8f8f8;\"\u003e\n\u003ctd style=\"padding: 10px 16px; font-weight: 600;\"\u003eGPUs\u003c\/td\u003e\n\u003ctd style=\"padding: 10px 16px;\"\u003e8x NVIDIA A100 SXM4 80 GB HBM2e (Ampere, 400 W per module)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd style=\"padding: 10px 16px; font-weight: 600;\"\u003eVRAM pool\u003c\/td\u003e\n\u003ctd style=\"padding: 10px 16px;\"\u003e640 GB HBM2e, NVSwitch-pooled\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr style=\"background: #f8f8f8;\"\u003e\n\u003ctd style=\"padding: 10px 16px; font-weight: 600;\"\u003eMemory bandwidth\u003c\/td\u003e\n\u003ctd style=\"padding: 10px 16px;\"\u003e2.04 TB\/s per GPU\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd style=\"padding: 10px 16px; font-weight: 600;\"\u003eGPU interconnect\u003c\/td\u003e\n\u003ctd style=\"padding: 10px 16px;\"\u003eNVLink 3 via NVSwitch, 600 GB\/s any-to-any\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr style=\"background: #f8f8f8;\"\u003e\n\u003ctd style=\"padding: 10px 16px; font-weight: 600;\"\u003eCompute\u003c\/td\u003e\n\u003ctd style=\"padding: 10px 16px;\"\u003e2,496 TFLOPS bf16 dense | 4,992 INT8 TOPS. No fp8 tensor path\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd style=\"padding: 10px 16px; font-weight: 600;\"\u003eCPUs\u003c\/td\u003e\n\u003ctd style=\"padding: 10px 16px;\"\u003e2x AMD EPYC 7643 (Milan, 48C\/96T each) on the Supermicro AS-4124GO-NART+ platform\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr style=\"background: #f8f8f8;\"\u003e\n\u003ctd style=\"padding: 10px 16px; font-weight: 600;\"\u003eSystem memory\u003c\/td\u003e\n\u003ctd style=\"padding: 10px 16px;\"\u003e1 TB DDR4-3200 ECC RDIMM (16x 64 GB)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd style=\"padding: 10px 16px; font-weight: 600;\"\u003eStorage\u003c\/td\u003e\n\u003ctd style=\"padding: 10px 16px;\"\u003e2x 1.92 TB U.2 NVMe boot + 2x 3.84 TB U.2 NVMe data\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr style=\"background: #f8f8f8;\"\u003e\n\u003ctd style=\"padding: 10px 16px; font-weight: 600;\"\u003eNetworking\u003c\/td\u003e\n\u003ctd style=\"padding: 10px 16px;\"\u003e2x NVIDIA ConnectX-7 400 Gb\/s\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd style=\"padding: 10px 16px; font-weight: 600;\"\u003eChassis\u003c\/td\u003e\n\u003ctd style=\"padding: 10px 16px;\"\u003e4U rack-mount, front-to-back airflow, 24\/7 rated\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr style=\"background: #f8f8f8;\"\u003e\n\u003ctd style=\"padding: 10px 16px; font-weight: 600;\"\u003ePower\u003c\/td\u003e\n\u003ctd style=\"padding: 10px 16px;\"\u003e4x 3000 W redundant\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003c\/tbody\u003e\n\u003c\/table\u003e\n\u003ch2 style=\"font-size: 22px; font-weight: bold; margin: 40px 0 16px 0; padding-bottom: 8px; border-bottom: 3px solid #334fb4;\"\u003eWhat it runs\u003c\/h2\u003e\n\u003cp\u003e640 GB at bf16 and Q4 covers most production serving and fine-tuning. Estimates are for this 8-GPU configuration with tensor parallelism.\u003c\/p\u003e\n\u003ctable style=\"width: 100%; border-collapse: collapse; margin-bottom: 24px; font-size: 15px;\"\u003e\n\u003cthead\u003e\n\u003ctr style=\"background: #0d0d0d; color: #fff;\"\u003e\n\u003cth style=\"padding: 12px 16px; text-align: left;\"\u003eComponent\u003c\/th\u003e\n\u003cth style=\"padding: 12px 16px; text-align: left;\"\u003eDetail\u003c\/th\u003e\n\u003c\/tr\u003e\n\u003c\/thead\u003e\n\u003ctbody\u003e\n\u003ctr style=\"background: #f8f8f8;\"\u003e\n\u003ctd style=\"padding: 10px 16px; font-weight: 600;\"\u003eLlama 3.3 70B \/ Qwen3-72B\u003c\/td\u003e\n\u003ctd style=\"padding: 10px 16px;\"\u003ebf16 full precision. ~35-50 tok\/s single stream, 1,500+ tok\/s batched\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd style=\"padding: 10px 16px; font-weight: 600;\"\u003eQwen3-235B-A22B MoE\u003c\/td\u003e\n\u003ctd style=\"padding: 10px 16px;\"\u003eQ4 across the full 640 GB pool\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr style=\"background: #f8f8f8;\"\u003e\n\u003ctd style=\"padding: 10px 16px; font-weight: 600;\"\u003eMistral Large 2 123B\u003c\/td\u003e\n\u003ctd style=\"padding: 10px 16px;\"\u003eQ4-Q8 with context headroom\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd style=\"padding: 10px 16px; font-weight: 600;\"\u003eQLoRA \/ full fine-tune 70B\u003c\/td\u003e\n\u003ctd style=\"padding: 10px 16px;\"\u003e8-way tensor parallel, NVSwitch-backed\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr style=\"background: #f8f8f8;\"\u003e\n\u003ctd style=\"padding: 10px 16px; font-weight: 600;\"\u003eClassic HPC\u003c\/td\u003e\n\u003ctd style=\"padding: 10px 16px;\"\u003e9.7 TFLOPS FP64 per GPU - 77.6 TFLOPS FP64 per node\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd style=\"padding: 10px 16px; font-weight: 600;\"\u003eMulti-tenant serving\u003c\/td\u003e\n\u003ctd style=\"padding: 10px 16px;\"\u003eseveral 30B-70B models resident at once\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003c\/tbody\u003e\n\u003c\/table\u003e\n\u003cp style=\"font-size: 14px; color: #666;\"\u003e\u003cem\u003eThroughput figures are estimates derived from published vendor specifications and community benchmarks for this GPU generation, not measurements taken on this exact configuration. We will give you measured numbers for your model and context length before you commit.\u003c\/em\u003e\u003c\/p\u003e\n\u003ch2 style=\"font-size: 22px; font-weight: bold; margin: 40px 0 16px 0; padding-bottom: 8px; border-bottom: 3px solid #334fb4;\"\u003ePower and cooling\u003c\/h2\u003e\n\u003cp\u003eRoughly 4.3 kW sustained: 3.2 kW of GPU plus CPU, fans and storage. That is 2.4 kW less than a Hopper node and it is the practical selling point - 4.3 kW fits a conventional datacentre rack with front-to-back airflow and no aisle containment retrofit. Over three years at EU industrial electricity rates the difference against an H100 node is a real line in the TCO, not a rounding error.\u003c\/p\u003e\n\u003ch2 style=\"font-size: 22px; font-weight: bold; margin: 40px 0 16px 0; padding-bottom: 8px; border-bottom: 3px solid #334fb4;\"\u003eWhat you give up, and export control\u003c\/h2\u003e\n\u003cp\u003e\u003cstrong\u003eNo fp8.\u003c\/strong\u003e Ampere predates the format. Models shipping fp8-native - the DeepSeek-V3 family, Hunyuan-Large - run here at bf16 or Q4 instead, which costs VRAM and throughput. If your roadmap is fp8 checkpoints, buy Hopper. This is the single real limitation of the tier and we would rather you hear it now.\u003c\/p\u003e\n\u003cp\u003e\u003cstrong\u003eExport control.\u003c\/strong\u003e A100 80 GB is above the Advanced Computing Item thresholds (ECCN 3A090.a), so it needs the same \u003cstrong\u003eKYC form\u003c\/strong\u003e and \u003cstrong\u003eEnd User Certificate\u003c\/strong\u003e as the Hopper nodes.\u003c\/p\u003e\n\u003cp\u003e\u003cstrong\u003eLate-life hardware.\u003c\/strong\u003e Ampere is near end of channel life. Warranty horizon and spares availability are shorter than on current generation, and we will tell you exactly what we can commit to in writing before you order.\u003c\/p\u003e\n\u003ch2 style=\"font-size: 22px; font-weight: bold; margin: 40px 0 16px 0; padding-bottom: 8px; border-bottom: 3px solid #334fb4;\"\u003eWhat you get from Kentino\u003c\/h2\u003e\n\u003cp\u003eKentino supplies, configures and supports this node as a complete deployment: rack and power design against your site's actual feed, front-to-back airflow sizing for 24\/7 operation, OS and driver stack (Ubuntu with pinned NVIDIA driver, CUDA, container toolkit), the inference or training stack of your choice, burn-in with a written report, EU-based logistics and warranty.\u003c\/p\u003e\n\u003cp\u003eWe also handle the export paperwork with you rather than mailing you a form and hoping. That part takes longer than the hardware and is where most first-time buyers of this class of machine get stuck.\u003c\/p\u003e\n\u003cp style=\"font-size: 14px; color: #666;\"\u003e\u003cem\u003ePrice shown is EUR excluding VAT, \u003cstrong\u003eEXW Prague\u003c\/strong\u003e (Incoterms 2020). Collection, freight, insurance and any destination import duties are for the buyer’s account. Lead time on request.\u003c\/em\u003e\u003c\/p\u003e\n\u003cp\u003e\u003cem\u003eTell us the models you need to serve and we will tell you honestly whether this node or a Hopper node is the right purchase - the answer is genuinely sometimes this one.\u003c\/em\u003e\u003c\/p\u003e\n\u003c\/div\u003e","brand":"Kentino","offers":[{"title":"Default Title","offer_id":54417265557832,"sku":null,"price":296026.0,"currency_code":"EUR","in_stock":true}],"url":"https:\/\/kentino.com\/es\/products\/k-node-a100-8x-sxm4-640gb","provider":"Kentino","version":"1.0","type":"link"}