{"id":97507,"date":"2023-11-13T23:58:00","date_gmt":"2023-11-13T16:58:00","guid":{"rendered":"https:\/\/www.xinwengao.com\/cn\/pr\/2023111323580097507"},"modified":"2023-11-14T22:04:14","modified_gmt":"2023-11-14T14:04:14","slug":"supermicro-%e6%8b%93%e5%b1%95%e4%ba%ba%e5%b7%a5%e6%99%ba%e8%83%bd%e8%a7%a3%e5%86%b3%e6%96%b9%e6%a1%88","status":"publish","type":"post","link":"https:\/\/www.xinwengao.com\/cn\/pr\/2023111323580097507","title":{"rendered":"Supermicro \u62d3\u5c55\u4eba\u5de5\u667a\u80fd\u89e3\u51b3\u65b9\u6848"},"content":{"rendered":"\n\n\n
\"\"<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u501f\u52a9\u5373\u5c06\u4e0a\u5e02\u7684<\/b> NVIDIA HGX H200 <\/b>\u548c\u914d\u5907<\/b> HBM3e <\/b>\u5185\u5b58\u7684<\/b> MGX Grace Hopper <\/b>\u5e73\u53f0<\/b><\/p>\n

Supermicro \u901a\u8fc7\u5bf9 NVIDIA HGX H200 \u548c Grace Hopper \u8d85\u7ea7\u82af\u7247\u7684\u652f\u6301\uff0c\u62d3\u5c55\u4e86 8-GPU\u30014-GPU \u548c MGX \u4ea7\u54c1\u7ebf\uff0c\u7528\u4e8e\u6ee1\u8db3 LLM \u5e94\u7528\u7684\u9700\u6c42\uff0c\u5e76\u642d\u8f7d\u4e86\u66f4\u5feb\u3001\u66f4\u5927\u5bb9\u91cf\u7684 HBM3e \u5185\u5b58\u3002\u521b\u65b0\u6b3e Supermicro \u6db2\u51b7 4U \u670d\u52a1\u5668\u642d\u8f7d\u4e86 NVIDIA HGX 8-GPU\uff0c\u4f7f\u6bcf\u4e2a\u673a\u67b6\u7684\u8ba1\u7b97\u5bc6\u5ea6\u7ffb\u500d\uff0c\u9ad8\u8fbe 80 \u5343\u74e6\/\u673a\u67b6\uff0c\u5e76\u964d\u4f4e\u4e86\u603b\u4f53\u62e5\u6709\u6210\u672c (TCO)<\/p>\n

\u52a0\u5229\u798f\u5c3c\u4e9a\u5dde\u5723\u4f55\u585e\u548c\u4e39\u4f5b\u5e02<\/span>2023\u5e7411\u670814\u65e5<\/span> \/\u7f8e\u901a\u793e\/ — Supercomputing Conference (SC23) – Supermicro, Inc.<\/b>\uff08\u7eb3\u65af\u8fbe\u514b\u80a1\u7968\u4ee3\u7801\uff1a<\/b>SMCI<\/b>\uff09<\/b> \u4f5c\u4e3a\u4eba\u5de5\u667a\u80fd\u3001\u4e91\u6280\u672f\u3001\u5b58\u50a8\u548c 5G\/\u8fb9\u7f18\u6574\u4f53 IT \u89e3\u51b3\u65b9\u6848\u4f9b\u5e94\u5546\uff0c\u501f\u52a9\u5bf9\u65b0\u6b3e\u5185\u7f6e H200 Tensor Core GPU \u7684 NVIDIA HGX H200 \u63d0\u4f9b\u652f\u6301\uff0c\u8fdb\u4e00\u6b65\u62d3\u5c55\u5176\u4eba\u5de5\u667a\u80fd\u7684\u4ea7\u54c1\u7ebf\u3002 \u4f5c\u4e3a\u4e1a\u754c\u9886\u5148\u7684\u4eba\u5de5\u667a\u80fd\u5e73\u53f0\uff0cSupermicro \u6db5\u76d6\u4e86 8U \u548c 4U \u901a\u7528 GPU \u7cfb\u7edf\uff0c\u53ef\u968f\u65f6\u5e94\u7528 HGX H200 8-GPU \u548c 4-GPU\u3002\u4e0e NVIDIA H100 Tensor Core GPU \u76f8\u6bd4\uff0c\u5176\u5177\u6709\u8fd1 2 \u500d\u7684\u5bb9\u91cf\u548c 1.4 \u500d\u66f4\u9ad8\u5e26\u5bbd\u7684 HBM3e \u5185\u5b58\u3002\u53e6\u5916\uff0cSupermicro NVIDIA MGX™<\/sup> \u7cfb\u7edf\u62e5\u6709\u6700\u4e30\u5bcc\u7684\u4ea7\u54c1\u7ec4\u5408\uff0c\u5e76\u517c\u5bb9\u5373\u5c06\u63a8\u51fa\u7684\u642d\u8f7d HBM3e \u5185\u5b58\u7684 NVIDIA Grace Hopper \u8d85\u7ea7\u82af\u7247\u3002\u51ed\u501f\u5176\u7a7a\u524d\u7684\u6027\u80fd\u3001\u53ef\u6269\u5c55\u6027\u548c\u53ef\u9760\u6027\uff0cSupermicro \u7684\u673a\u67b6\u89c4\u6a21\u4eba\u5de5\u667a\u80fd\u89e3\u51b3\u65b9\u6848\u53ef\u5728\u6ee1\u8db3\u65e5\u76ca\u589e\u957f\u7684\u6a21\u578b\u5c3a\u5bf8\u9700\u6c42\u7684\u540c\u65f6\uff0c\u5feb\u901f\u63d0\u5347\u9ad8\u5f3a\u5ea6\u751f\u6210\u5f0f\u4eba\u5de5\u667a\u80fd\u3001\u5927\u8bed\u8a00\u6a21\u578b (LLM) \u8bad\u7ec3\u4ee5\u53ca\u9ad8\u6027\u80fd\u8ba1\u7b97\u5e94\u7528\u7684\u7b97\u529b\u3002 \u501f\u52a9\u79ef\u6728\u5f0f\u67b6\u6784\uff0cSupermicro \u80fd\u591f\u5feb\u901f\u5c06\u65b0\u6280\u672f\u63a8\u5411\u5e02\u573a\uff0c\u8ba9\u5ba2\u6237\u66f4\u5feb\u63d0\u5347\u751f\u4ea7\u6548\u7387\u3002<\/p>\n

\"Supermicro<\/a>
Supermicro 4U 8GPU H200<\/span> <\/div>\n

Supermicro \u5145\u5206\u501f\u52a9\u6700\u65b0\u7684 Supermicro \u6db2\u51b7\u89e3\u51b3\u65b9\u6848\uff0c\u91c7\u7528\u6db2\u51b7 4U \u7cfb\u7edf\uff0c\u5e76\u57fa\u4e8e NVIDIA HGX H100 8-GPU \u7cfb\u7edf\uff0c\u63a8\u51fa\u4e86\u4e1a\u754c\u6700\u9ad8\u5bc6\u5ea6\u7684\u670d\u52a1\u5668\u3002\u4f5c\u4e3a\u4e1a\u754c\u6700\u7d27\u51d1\u7684\u9ad8\u6027\u80fd GPU \u670d\u52a1\u5668\uff0c\u5176\u80fd\u591f\u5e2e\u52a9\u6570\u636e\u4e2d\u5fc3\u5de5\u4f5c\u4eba\u5458\u51cf\u964d\u4f4e\u5bf9\u5360\u5730\u9762\u79ef\u7684\u9700\u6c42\uff0c\u5e76\u8282\u7ea6\u80fd\u6e90\u6210\u672c\uff0c\u540c\u65f6\u63d0\u4f9b\u9876\u7ea7\u7684\u5355\u673a\u67b6\u9ad8\u6027\u80fd\u4eba\u5de5\u667a\u80fd\u8bad\u7ec3\u80fd\u529b\u3002\u5229\u7528\u6700\u9ad8\u5bc6\u5ea6\u7684 GPU \u7cfb\u7edf\uff0c\u7ec4\u7ec7\u53ef\u901a\u8fc7\u91c7\u7528\u5c16\u7aef\u7684\u6db2\u51b7\u89e3\u51b3\u65b9\u6848\u964d\u4f4e\u5176\u603b\u4f53\u62e5\u6709\u6210\u672c \u3002<\/p>\n

Supermicro \u603b\u88c1\u517c\u884c\u653f\u603b\u88c1 Charles Liang \u8868\u793a\uff1a"Supermicro \u4e0e NVIDIA \u643a\u624b\u8bbe\u8ba1\u4e86\u5148\u8fdb\u7684\u7cfb\u7edf\uff0c\u7528\u4e8e\u4eba\u5de5\u667a\u80fd\u8bad\u7ec3\u548c\u9ad8\u6027\u80fd\u8ba1\u7b97\u5e94\u7528" \u3002"\u501f\u52a9\u79ef\u6728\u5f0f\u67b6\u6784\uff0c\u6211\u4eec\u80fd\u591f\u5728\u5e02\u573a\u4e0a\u7387\u5148\u63a8\u51fa\u6700\u65b0\u79d1\u6280\uff0c\u8ba9\u5ba2\u6237\u66f4\u8fc5\u901f\u5730\u90e8\u7f72\u751f\u6210\u5f0f\u4eba\u5de5\u667a\u80fd\u3002\u6211\u4eec\u80fd\u591f\u5229\u7528\u5168\u7403\u5316\u751f\u4ea7\u8bbe\u65bd\uff0c\u66f4\u5feb\u5730\u5411\u5ba2\u6237\u4ea4\u4ed8\u8fd9\u4e9b\u65b0\u7cfb\u7edf\u3002\u65b0\u7cfb\u7edf\u91c7\u7528\u642d\u8f7d\u4e86 NVIDIA® NVLink™ \u548c NVSwitch™<\/sup> \u9ad8\u901f GPU-GPU \u4e92\u8054\u6280\u672f\uff08\u901f\u5ea6\u9ad8\u8fbe 900GB\/s\uff09\u7684 NVIDIA H200 GPU\uff0c\u5728\u6211\u4eec\u7684\u673a\u67b6\u89c4\u6a21\u4eba\u5de5\u667a\u80fd\u89e3\u51b3\u65b9\u6848\u4e2d\uff0c\u53ef\u4e3a\u6bcf\u4e2a\u8282\u70b9\u63d0\u4f9b\u9ad8\u8fbe 1.1TB \u7684\u9ad8\u5e26\u5bbd HBM3e \u5185\u5b58\uff0c\u4ee5\u5b9e\u73b0 LLM \u548c\u751f\u6210\u5f0f\u4eba\u5de5\u667a\u80fd\u7684\u6700\u4f73\u6a21\u578b\u5e76\u884c\u6027\u80fd\u3002\u540c\u65f6\uff0c\u5bf9\u4e8e\u80fd\u63d0\u4f9b\u5168\u7403\u6700\u4e3a\u7d27\u51d1\u7684 NVIDIA HGX 8-GPU \u6db2\u51b7\u670d\u52a1\u5668\uff0c\u76f4\u63a5\u5c06\u6211\u4eec\u7684\u673a\u67b6\u89c4\u6a21\u4eba\u5de5\u667a\u80fd\u89e3\u51b3\u65b9\u6848\u5bc6\u5ea6\u7ffb\u500d\uff0c\u5e76\u964d\u4f4e\u80fd\u6e90\u6210\u672c\uff0c\u4e3a\u5f53\u4eca\u901f\u5ea6\u66f4\u5feb\u7684\u6570\u636e\u4e2d\u5fc3\u5b9e\u73b0\u7eff\u8272\u8ba1\u7b97\uff0c\u6211\u4eec\u6df1\u611f\u5174\u594b\u3002"<\/p>\n

\u6df1\u5165\u4e86\u89e3\u642d\u8f7d NVIDIA GPU \u7684 Supermicro \u670d\u52a1\u5668<\/a><\/p>\n

Supermicro \u8bbe\u8ba1\u5236\u9020\u4f17\u591a\u4e0d\u540c\u5916\u5f62\u5c3a\u5bf8\u7684\u4eba\u5de5\u667a\u80fd\u670d\u52a1\u5668\u3002\u5e7f\u53d7\u6b22\u8fce\u7684 8U \u548c 4U \u901a\u7528 GPU \u7cfb\u7edf\u642d\u8f7d\u4e86\u56db\u901a\u9053\u548c\u516b\u901a\u9053 NVIDIA HGX H100 GPU\uff0c\u73b0\u5df2\u652f\u6301\u65b0\u6b3e H200 GPU\uff0c\u53ef\u5728\u66f4\u77ed\u65f6\u95f4\u5185\u5b8c\u6210\u66f4\u5927\u89c4\u6a21\u8bed\u8a00\u6a21\u578b\u7684\u8bad\u7ec3\u3002\u6bcf\u4e2a NVIDIA H200 GPU \u90fd\u642d\u8f7d\u4e86 141 GB \u5185\u5b58\uff0c\u5e26\u5bbd\u9ad8\u8fbe 4.8 TB\/s\u3002<\/p>\n

"Supermicro \u5373\u5c06\u63a8\u51fa\u7684\u670d\u52a1\u5668\u8bbe\u8ba1\u91c7\u7528\u4e86 NVIDIA HGX H200\uff0c\u80fd\u591f\u5e2e\u52a9\u52a0\u901f\u751f\u6210\u5f0f\u4eba\u5de5\u667a\u80fd\u548c\u9ad8\u6027\u80fd\u8ba1\u7b97\u7684\u5de5\u4f5c\u8d1f\u8f7d\uff0c\u8ba9\u4f01\u4e1a\u548c\u673a\u6784\u80fd\u591f\u5145\u5206\u53d1\u6325\u5176\u4eba\u5de5\u667a\u80fd\u67b6\u6784\u7684\u80fd\u529b\uff0c" NVIDIA \u7684\u9ad8\u6027\u80fd\u8ba1\u7b97\u3001\u4eba\u5de5\u667a\u80fd\u548c\u91cf\u5b50\u8fd0\u7b97\u6570\u636e\u4e2d\u5fc3\u4ea7\u54c1\u89e3\u51b3\u65b9\u6848\u603b\u76d1 Dion Harris \u8868\u793a\uff0c"\u642d\u8f7d\u9ad8\u901f HBM3e \u5185\u5b58\u7684 NVIDIA H200 GPU \u5c06\u80fd\u591f\u5e94\u5bf9\u5404\u79cd\u6765\u81ea\u5de5\u4f5c\u8d1f\u8f7d\u7684\u6d77\u91cf\u6570\u636e\u3002"<\/p>\n

\u6b64\u5916\uff0c\u8fd1\u671f\u63a8\u51fa\u7684 Supermicro MGX<\/a> \u670d\u52a1\u5668\u642d\u8f7d\u4e86 NVIDIA GH200 Grace Hopper<\/span> \u8d85\u7ea7\u82af\u7247\uff0c\u4e13\u4e3a\u6574\u5408\u642d\u8f7d HBM3e \u5185\u5b58\u7684 NVIDIA H200 GPU \u89e3\u51b3\u65b9\u6848\u800c\u8bbe\u8ba1\u3002<\/p>\n

\u6700\u65b0\u7684 NVIDIA GPU \u4f7f\u5f97\u52a0\u901f\u5f53\u4eca\u548c\u672a\u6765\u7684\u5927\u578b\u8bed\u8a00\u6a21\u578b (LLM) \u6210\u4e3a\u53ef\u80fd\uff0c\u8fd9\u4e9b\u6a21\u578b\u5177\u6709\u7684\u6570\u5343\u4ebf\u7684\u53c2\u6570\u4f7f\u5176\u80fd\u591f\u9002\u5e94\u66f4\u7d27\u51d1\u3001\u9ad8\u6548\u7684\u96c6\u7fa4\uff0c\u4ece\u800c\u5728\u66f4\u77ed\u65f6\u95f4\u5185\u8bad\u7ec3\u751f\u6210\u5f0f\u4eba\u5de5\u667a\u80fd\u3002\u6b64\u5916\uff0c\u5176\u8fd8\u652f\u6301\u5728\u5355\u4e2a\u7cfb\u7edf\u4e2d\u5bb9\u7eb3\u591a\u4e2a\u66f4\u5927\u7684\u6a21\u578b\uff0c\u7528\u4e8e\u5b9e\u65f6 LLM \u63a8\u7406\uff0c\u4ece\u800c\u670d\u52a1\u4e8e\u6570\u767e\u4e07\u7528\u6237\u7684\u751f\u6210\u5f0f\u4eba\u5de5\u667a\u80fd\u3002<\/p>\n

Supermicro \u5c06\u5728 SC23 \u4e0a\u5c55\u793a\u5176\u6700\u65b0\u4ea7\u54c1\u2014\u2014\u914d\u5907\u516b\u901a\u9053 NVIDIA HGX H100 \u53ca\u6700\u65b0\u7684\u6db2\u51b7\u521b\u65b0\u6280\u672f\u7684 4U \u901a\u7528 GPU \u7cfb\u7edf\uff0c\u7528\u4e8e\u8fdb\u4e00\u6b65\u63d0\u5347\u5bc6\u5ea6\u548c\u80fd\u6548\uff0c\u8fdb\u800c\u63a8\u52a8\u4eba\u5de5\u667a\u80fd\u7684\u6f14\u5316\u3002\u642d\u914d Supermicro \u4e1a\u754c\u9886\u5148\u7684 GPU \u548c CUP \u51b7\u5374\u677f\u3001\u4e13\u4e3a\u7eff\u8272\u8ba1\u7b97\u8bbe\u8ba1\u7684 CDU\uff08\u51b7\u91cf\u5206\u914d\u5355\u5143\uff09\u548c CDM\uff08\u51b7\u91cf\u5206\u914d\u6b67\u7ba1\uff09\uff0c\u5176\u65b0\u6b3e\u6db2\u51b7 4U \u901a\u7528 GPU \u7cfb\u7edf\u540c\u6837\u53ef\u642d\u8f7d\u516b\u901a\u9053 NVIDIA HGX H200\uff0c\u5e76\u4e14\uff0c\u901a\u8fc7 Supermicro\u5168\u96c6\u6210\u7684\u6db2\u51b7\u673a\u67b6\u89e3\u51b3\u65b9\u6848\u548c\u6211\u4eec\u7684 L10\u3001L11 \u548c L12 \u9a8c\u8bc1\u6d4b\u8bd5\uff0c\u5176\u80fd\u591f\u5927\u5e45\u964d\u4f4e\u6570\u636e\u4e2d\u5fc3\u7684\u4f53\u79ef\u3001\u524a\u51cf\u80fd\u6e90\u6210\u672c\u5e76\u51cf\u5c11\u90e8\u7f72\u4e2d\u7684\u5404\u79cd\u963b\u788d\u3002<\/p>\n

\u5982\u9700\u4e86\u89e3\u66f4\u591a\u4fe1\u606f\uff0c\u656c\u8bf7\u8385\u4e34 Supermicro \u5728 SC23 \u4e0a\u7684\u5c55\u4f4d<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"

\u501f\u52a9\u5373\u5c06\u4e0a\u5e02\u7684 NVIDIA HGX H200 \u548c\u914d\u5907 HBM3e \u5185\u5b58\u7684 MGX Grace Hopper \u5e73\u53f0 Supermicro \u901a\u8fc7\u5bf9 NVIDIA HGX H200 \u548c Grace Hopper \u8d85\u7ea7\u82af\u7247\u7684\u652f\u6301\uff0c\u62d3\u5c55\u4e86 8-GPU\u30014-GPU \u548c MGX \u4ea7\u54c1\u7ebf\uff0c\u7528\u4e8e\u6ee1\u8db3 LLM \u5e94\u7528\u7684\u9700\u6c42\uff0c\u5e76\u642d\u8f7d\u4e86\u66f4\u5feb\u3001\u66f4\u5927\u5bb9\u91cf\u7684 HBM3e \u5185\u5b58\u3002\u521b\u65b0\u6b3e Supermicro \u6db2\u51b7 4U \u670d\u52a1\u5668\u642d\u8f7d\u4e86 NVIDIA HGX 8-GPU\uff0c\u4f7f\u6bcf\u4e2a\u673a\u67b6\u7684\u8ba1\u7b97\u5bc6\u5ea6\u7ffb\u500d\uff0c\u9ad8\u8fbe 80 \u5343\u74e6\/\u673a\u67b6\uff0c\u5e76\u964d\u4f4e\u4e86\u603b\u4f53\u62e5\u6709\u6210\u672c (TCO) \u52a0\u5229\u798f\u5c3c\u4e9a\u5dde\u5723\u4f55\u585e\u548c\u4e39\u4f5b\u5e022023\u5e7411\u670814\u65e5 \/\u7f8e\u901a\u793e\/ — Supercomputing Conference (SC23) – Supermicro, Inc.\uff08\u7eb3\u65af\u8fbe\u514b\u80a1\u7968\u4ee3\u7801\uff1aSMCI\uff09 \u4f5c\u4e3a\u4eba\u5de5\u667a\u80fd\u3001\u4e91\u6280\u672f\u3001\u5b58\u50a8\u548c 5G\/\u8fb9\u7f18\u6574\u4f53 IT \u89e3\u51b3\u65b9\u6848\u4f9b\u5e94\u5546\uff0c\u501f\u52a9\u5bf9\u65b0\u6b3e\u5185\u7f6e H200 Tensor Core GPU \u7684 NVIDIA HGX H200 \u63d0\u4f9b\u652f\u6301\uff0c\u8fdb\u4e00\u6b65\u62d3\u5c55\u5176\u4eba\u5de5\u667a\u80fd\u7684\u4ea7\u54c1\u7ebf\u3002 \u4f5c\u4e3a\u4e1a\u754c\u9886\u5148\u7684\u4eba\u5de5\u667a\u80fd\u5e73\u53f0\uff0cSupermicro \u6db5\u76d6\u4e86 8U \u548c 4U \u901a\u7528 GPU \u7cfb\u7edf\uff0c\u53ef\u968f\u65f6\u5e94\u7528 HGX H200 8-GPU \u548c 4-GPU\u3002\u4e0e NVIDIA H100 Tensor Core GPU \u76f8\u6bd4\uff0c\u5176\u5177\u6709\u8fd1 2 \u500d\u7684\u5bb9\u91cf\u548c 1.4… Read More<\/a><\/p>\n","protected":false},"author":556,"featured_media":97511,"comment_status":"close","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1965],"tags":[],"class_list":["post-97507","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-partnerships"],"amp_enabled":true,"_links":{"self":[{"href":"https:\/\/www.xinwengao.com\/cn\/wp-json\/wp\/v2\/posts\/97507","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.xinwengao.com\/cn\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.xinwengao.com\/cn\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.xinwengao.com\/cn\/wp-json\/wp\/v2\/users\/556"}],"replies":[{"embeddable":true,"href":"https:\/\/www.xinwengao.com\/cn\/wp-json\/wp\/v2\/comments?post=97507"}],"version-history":[{"count":1,"href":"https:\/\/www.xinwengao.com\/cn\/wp-json\/wp\/v2\/posts\/97507\/revisions"}],"predecessor-version":[{"id":97510,"href":"https:\/\/www.xinwengao.com\/cn\/wp-json\/wp\/v2\/posts\/97507\/revisions\/97510"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.xinwengao.com\/cn\/wp-json\/wp\/v2\/media\/97511"}],"wp:attachment":[{"href":"https:\/\/www.xinwengao.com\/cn\/wp-json\/wp\/v2\/media?parent=97507"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.xinwengao.com\/cn\/wp-json\/wp\/v2\/categories?post=97507"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.xinwengao.com\/cn\/wp-json\/wp\/v2\/tags?post=97507"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}