惗惉础滨侦嵟揔侧僷僼仅乕儅儞僗骋笔鲍
NVIDIA HGX™ H200
搶杒揹椡偺骋笔鲍僋儔僂僪僒乕价僗偱偛棙梡偄偨偩偗傑偡丅
LLama2 70B 悇榑
1.9攞偺崅懍壔
GPT3-175B 悇榑
1.6攞偺崅懍壔
崅惈擞僐儞僺儏乕侥傿儞僌
110攞偺崅懍壔
NVIDIA Hopper™ 傾乕僉僥僋僠儍傪儀乕僗偲偡傞 NVIDIA H200 偼丄枅昩 4.8 僥儔僶僀僩 (TB/s) 偱
141 僊僈僶僀僩 (GB) 偺 HBM3e 儊儌儕傪採嫙偡傞弶偺 GPU 偱偡丅
偙傟偼丄NVIDIA H100 Tensor 僐傾 GPU 偺栺 2 攞偺梕検偱丄儊儌儕懷堟暆偼 1.4 攞偱偡丅
H200 偺戝梕検偐偮崅懍側儊儌儕偼丄惗惉 AI 偲 LLM 傪壛懍偟丄僄僱儖僊乕岠棪傪岦忋偝偣丄憤強桳僐僗僩傪掅尭偟丄HPC 儚乕僋
儘乕僪偺偨傔偺壢妛僐儞僺儏乕僥傿儞僌傪慜恑偝偣傑偡丅
| 崁栚 | NVIDIA H200 乮SXM仸1乯 |
|---|---|
| 颁笔鲍儊儌儕 | 141 GB |
| GPU 儊儌儕懷堟暆 | 4.8 TB/昩 |
| FP64 | 34 TFLOPS |
| FP64 Tensor僐傾 | 67 TFLOPS |
| FP32 | 67 TFLOPS |
| T32 Tensor僐傾仸2 | 989 TFLOPS |
| BFLOAT16 Tensor僐傾仸2 | 1,979 TFLOPS |
| FP16 Tensor僐傾仸2 | 1,979 TFLOPS |
| FP8 Tensor僐傾仸2 | 3,958 TFLOPS |
| INT8 Tensor僐傾仸2 | 3,958 TFLOPS |
仸1巇条偼曄峏偝傟傞応崌偑偁傝杰偡
仸2慳惈偁傝
| 崁栚 | 惈擞 |
|---|---|
| GPU | NVIDIA H200 4婎 |
| 骋笔鲍娫捠怣 | 900骋产辫蝉乮狈痴尝颈苍办丒狈痴厂飞颈迟肠丑乯 |
| CPU | 2乣3.8骋贬锄憡摉 |
| 僔僗侥俭儊儌儕 | 嵟戝1罢叠 |
| 僗僩儗乕僕 | 2TB |
| 奜晹僱僢僩儚乕僋 | 1骋产辫蝉乮100惭产辫蝉怀堟曐彻乯 |
仸巇条偼曄峏偝傟傞応崌偑偁傝杰偡
偍栤偄崌傢偣