{"id":49408,"date":"2023-11-14T14:06:22","date_gmt":"2023-11-14T06:06:22","guid":{"rendered":"https:\/\/www.ioiotimes.com\/?p=49408"},"modified":"2024-07-16T16:38:38","modified_gmt":"2024-07-16T08:38:38","slug":"%e6%8a%80%e5%98%89super-computing%e8%b6%85%e9%80%b2%e5%8c%96%e5%b1%95%e5%87%ba%e5%85%88%e9%80%b2%e7%9a%84%e6%95%a3%e7%86%b1%e6%8a%80%e8%a1%93%e5%92%8c%e5%8f%af%e6%93%b4%e5%b1%95%e8%b3%87%e6%96%99","status":"publish","type":"post","link":"https:\/\/www.ioiotimes.com\/?p=49408","title":{"rendered":"\u6280\u5609Super Computing\u8d85\u9032\u5316\u5c55\u51fa\u5148\u9032\u7684\u6563\u71b1\u6280\u8853\u548c\u53ef\u64f4\u5c55\u8cc7\u6599\u4e2d\u5fc3\u57fa\u790e\u8a2d\u65bd\u7522\u54c1\u00a0"},"content":{"rendered":"\n

NVIDIA<\/strong>\u7cbe\u7cb9\u800c\u751f\uff0c\u6280\u5609\u4f3a\u670d\u5668\u5e73\u53f0\u642d\u8f09\u4e0b\u4e00\u4ee3AI\u8655\u7406\u5668\u6280\u8853\u00a0<\/strong><\/h3>\n\n\n\n
\"\"<\/figure>\n\n\n\n

<\/p>\n\n\n\n

\u9ad8\u6027\u80fd\u4f3a\u670d\u5668\u8207\u5de5\u4f5c\u7ad9\u9818\u5c0e\u54c1\u724c\u6280\u5609\u79d1\u6280\uff08TWSE\uff1a2376\uff09\u96c6\u5718\u65d7\u4e0b\u5b50\u516c\u53f8\u6280\u92fc\u79d1\u6280\uff0c\u6301\u7e8c\u958b\u767c Arm \u548c x86 \u8655\u7406\u5668\u67b6\u69cb\u4f3a\u670d\u5668\u3001\u64f4\u5927\u9ad8\u6548 IT \u786c\u9ad4\u8a2d\u5099\u7684\u5148\u9032\u51b7\u537b\u6280\u8853\uff0c\u4e26\u5f37\u529b\u652f\u63f4 AI \u52a0\u901f\u5668\u7684\u591a\u6a23\u5316\u4f3a\u670d\u5668\u5e73\u53f0\u4ee5\u4fdd\u6301\u696d\u754c\u9818\u5148\u7684\u5730\u4f4d\u3002\u5728 SC23\uff0c\u6280\u5609\uff08\u5c55\u4f4d\uff03355\uff09\u5c07\u5c55\u793a\u591a\u6b3e\u5275\u65b0\u7684\u4f3a\u670d\u5668\uff0c\u5305\u62ec\u652f\u63f4 NVIDIA GH200 Grace Hopper\u00a0\u8d85\u7d1a\u6676\u7247\u548c\u4e0b\u4e00\u4ee3 AMD Instinct\u2122 APU \u7684\u5e73\u53f0\uff0c\u6280\u5609\u8457\u91cd\u65bc\u904b\u7b97\u4e2d\u5fc3\u6700\u91cd\u8981\u7684\u9700\u6c42\uff0c\u63d0\u4f9b\u9ad8\u6548\u80fd IT \u786c\u9ad4\u8a2d\u5099\u7684\u7d55\u4f73\u6563\u71b1\u6280\u8853\uff0c\u4ee5\u53ca\u652f\u63f4\u5be6\u6642\u5206\u6790\u548c\u5feb\u901f\u5be6\u73fe\u6210\u6548\u7684 AI \u52d5\u529b\u7522\u54c1\u3002\u00a0<\/p>\n\n\n\n

<\/p>\n\n\n\n

\u5148\u9032\u51b7\u537b\u6563\u71b1\u6280\u8853<\/strong> <\/strong><\/h3>\n\n\n\n

\u8ffd\u6c42\u66f4\u9ad8\u6548\u80fd\u7684\u9700\u6c42\u9020\u6210\u4e86\u66f4\u591a\u71b1\u91cf\u7684\u7522\u751f\uff0c\u7576\u4eca\u8a31\u591a\u8cc7\u6599\u4e2d\u5fc3\u9700\u8981\u6f88\u5e95\u8f49\u8b8a\u5176\u6563\u71b1\u57fa\u790e\u8a2d\u65bd\u4ee5\u8ddf\u4e0a\u6642\u4ee3\u7684\u6b65\u4f10\u3002\u6280\u5609\u5148\u9032\u51b7\u537b\u6563\u71b1\u6280\u8853\u56e0\u61c9\u800c\u751f\uff0c\u8b93\u904b\u7b97\u8a2d\u5099\u5f97\u4ee5\u5728\u7dad\u6301\u9ad8\u6548\u80fd\u7684\u60c5\u6cc1\u4e0b\u66f4\u52a0\u7bc0\u80fd\u3002SC23 \u5c55\u4f4d\u4e0a\uff0c\u6280\u5609\u5c07\u5c55\u51fa\u65b0\u6b3e\u7684\u55ae\u76f8\u6d78\u6c92\u5f0f\u51b7\u537b\u69fd A1P0-EA0<\/a>\uff0c\u6280\u5609\u5728\u6d78\u6c92\u5f0f\u51b7\u537b\u65b9\u6848\u5177\u6709\u76f8\u7576\u8c50\u5bcc\u7684\u7d93\u9a57\uff0c\u6574\u9ad4\u7684\u89e3\u6c7a\u65b9\u6848\u6db5\u84cb\u6d78\u6c92\u5f0f\u51b7\u537b\u4f3a\u670d\u5668\u3001\u51b7\u537b\u69fd\u3001\u51b7\u537b\u6db2\u3001\u76f8\u95dc\u7684\u7dad\u904b\u5de5\u5177\u4ee5\u53ca\u904d\u53ca\u5168\u7403\u7684\u670d\u52d9\uff0c\u80fd\u63d0\u4f9b\u5ba2\u6236\u4e00\u689d\u9f8d\u7684\u6d78\u6c92\u51b7\u537b\u89e3\u6c7a\u65b9\u6848\u3002\u4eca\u5e74\u6280\u5609\u9084\u5c07\u5728\u5c55\u4f4d\u4e0a\u5c55\u73fe\u53e6\u4e00\u7a2e\u51b7\u537b\u6280\u8853\uff0c\u9069\u7528\u65bc NVIDIA Grace CPU\u00a0\u8d85\u7d1a\u6676\u7247\u3001NVIDIA Grace Hopper \u8d85\u7d1a\u6676\u7247\u3001AMD EPYC\u2122 9004 \u8655\u7406\u5668\u548c\u7b2c 4 \u4ee3 Intel\u00ae<\/sup>\u00a0Xeon\u00ae<\/sup>\u8655\u7406\u5668\u7684\u76f4\u63a5\u6db2\u9ad4\u51b7\u537b\u89e3\u6c7a\u65b9\u6848\uff08Direct Liquid Cool, DLC\uff09\uff0c\u914d\u5099\u6280\u5609\u7814\u767c\u7684\u6c34\u51b7\u5faa\u74b0\u6a21\u7d44\uff0c\u51b7\u537b\u6db2\u5206\u914d\u5c90\u7ba1\u7b49\u51b7\u537b\u88dd\u7f6e\u3002<\/p>\n\n\n\n

 <\/p>\n\n\n\n

\u6a21\u7d44\u5316\u7684<\/strong>AI<\/strong>\u8207HPC<\/strong>\u7cfb\u7d71<\/strong> <\/strong><\/h3>\n\n\n\n

\u6280\u5609\u5c07\u5728 SC23 \u5c55\u793a\u4e00\u5ea7 GIGA POD \u6a5f\u6ac3\u7d1a\u904b\u7b97\u7522\u54c1\u3002\u6280\u5609\u5728\u4eca\u5e74\u6210\u529f\u7684\u70ba\u6b50\u7f8e\u96f2\u7aef\u670d\u52d9\u4f9b\u61c9\u5546\u90e8\u7f72\u5176 GIGA POD \u7522\u54c1\uff0c\u5354\u52a9\u9019\u4e9b\u65b0\u8208 GPU \u96f2\u7aef\u670d\u52d9\u4f9b\u61c9\u5546\u898f\u5283\u5176\u8cc7\u6599\u4e2d\u5fc3\u57fa\u790e\u67b6\u69cb\u3002\u76ee\u524d\u7684\u90e8\u5c6c\u67b6\u69cb\u4ee5 8 \u500b GIGA POD \u7d44\u6210\u4e00\u500b\u57fa\u790e\u5de5\u4f5c\u7fa4\u7d44\uff0c\u6bcf\u500b\u7fa4\u7d44\u88dd\u914d\u4e86 32 \u53f0 G593 \u4f3a\u670d\u5668\uff0c\u7e3d\u5171\u642d\u8f09 256 \u500b NVIDIA H100 Tensor Core GPU<\/a> \u4f86\u5be6\u73fe 1 exaflop\uff08\u6bcf\u79d2\u4e00\u5343\u5104\u6d6e\u9ede\u904b\u7b97\uff09\u7684 FP8 \u6d6e\u9ede\u904b\u7b97\u6027\u80fd\uff1bG593-SD0\u00a0\u8fd1\u671f\u4e5f\u53c3\u8207\u4e86 MLPerf \u57fa\u6e96\u6e2c\u8a66\uff0c\u4e26\u7372\u5f97\u91ab\u7642\u5f71\u50cf\u8fa8\u8b58\u8207\u81ea\u7136\u8a9e\u8a00\u8655\u7406\u61c9\u7528\u7684\u6700\u4f73\u6548\u80fd\u8868\u73fe\u3002\u540c\u6642\u6280\u5609\u9084\u5c07\u5c55\u793a\u652f\u63f4 NVIDIA Grace CPU \u8d85\u7d1a\u6676\u7247\u7684 H263-V11\uff0c\u4ee5\u53ca\u652f\u63f4 NVIDIA Grace Hopper\u00a0\u8d85\u7d1a\u6676\u7247\u7684 H223-V10 \u4f3a\u670d\u5668\uff0c\u9019\u4e9b\u57fa\u65bc Arm \u67b6\u69cb\u8655\u7406\u5668\u7684\u4f3a\u670d\u5668\u4e0d\u4f46\u80fd\u652f\u63f4 NVMe \u786c\u789f\uff0c\u5176\u914d\u5099\u7684 NVIDIA BlueField-3 DPU<\/a> \u5c07\u80fd\u66f4\u5feb\u901f\u8655\u7406\u4f3a\u670d\u5668\u9593\u7684\u8cc7\u6599\u50b3\u8f38\uff0c\u8b93\u5ba2\u6236\u80fd\u6839\u64da\u9700\u6c42\u9032\u884c\u5feb\u901f\u7684\u6a6b\u5411\u64f4\u5145\uff0c\u4ee5\u56e0\u61c9\u5e02\u5834\u5c0d\u751f\u6210\u5f0f AI \u6216\u5927\u578b\u8a9e\u8a00\u6a21\u578b\u7684\u9700\u6c42\u3002\u7576\u7136\u9084\u6709\u9996\u6b21\u516c\u958b\u4eae\u76f8\u7684\u5091\u51fa\u7cfb\u7d71\uff0cG383-R80\uff0c\u5c08\u70ba\u65b0\u4e00\u4ee3 AMD Instinct\u2122 APU \u8a2d\u8a08\uff0c\u65b0\u4e00\u4ee3\u7684 AMD Instinct \u52a0\u901f\u5668\u5177\u5099\u7a81\u51fa\u7684\u898f\u683c\uff0c\u80fd\u5c0dAI\u5de5\u4f5c\u8ca0\u8f09\u63d0\u4f9b\u76f8\u7576\u5927\u7684\u52a9\u529b\u3002\u00a0<\/p>\n\n\n\n

<\/p>\n\n\n\n

\u53ef\u64f4\u5c55\u7684\u8cc7\u6599\u4e2d\u5fc3\u57fa\u790e\u8a2d\u65bd<\/strong> <\/strong><\/h3>\n\n\n\n

\u5305\u542b\u4e86\u7372\u5f97 NVIDIA\u8a8d\u8b49<\/a>\u7684 G493-SB0<\/a>\u00a0\u4f3a\u670d\u5668\uff0c\u80fd\u652f\u63f4\u516b\u5f35 PCIe Gen5 GPU\uff0c\u9069\u7528\u65bc NVIDIA L4<\/a>\u00a0Tensor Core \u548c L40 GPU<\/a>\uff0c\u540c\u6642\u5177\u5099\u80fd\u63d0\u4f9b NVIDIA BlueField-3 DPU \u8207 ConnectX-7\u7db2\u8def\u5361<\/a>\u4f7f\u7528\u7684\u64f4\u5145\u63d2\u69fd\uff0c\u80fd\u652f\u63f4\u4f01\u696d\u5728\u5be6\u5883\u8996\u89ba\u5316\u5354\u4f5c\u7684\u61c9\u7528\uff0c\u5c07\u6210\u70ba NVIDIA OVX<\/a> \u7cfb\u7d71\u7684\u57fa\u77f3\u3002\u6b64\u5916\uff0c\u9075\u5faa NVIDIA MGX \u6a21\u7d44\u5316\u8a2d\u8a08\u7684 XH23-VG0 \u642d\u914d\u4e86 NVIDIA Grace Hopper \u8d85\u7d1a\u6676\u7247\uff0c\u63d0\u4f9b\u984d\u5916 FHFL \u7684\u64f4\u5145\u63d2\u69fd\uff0c\u53ef\u7528\u65bc\u52a0\u901f\u5927\u898f\u6a21\u6578\u5b78\u904b\u7b97\uff0c\u662f AI \u548c HPC \u61c9\u7528\u7684\u7d55\u4f73\u9078\u64c7\u3002\u00a0<\/p>\n\n\n\n

<\/p>\n\n\n\n

\u4f01\u696d\u7d1a\u904b\u7b97<\/strong> <\/strong><\/h3>\n\n\n\n

\u9664\u4e86\u4e0a\u8ff0\u706b\u71b1\u7684\u751f\u6210\u5f0f AI \u53ca HPC \u7522\u54c1\u4ee5\u5916\uff0c\u6280\u5609\u4e5f\u5c0d\u4f01\u696d\u5ba2\u6236\u63d0\u51fa\u5404\u6709\u7279\u8272\u7684\u7522\u54c1\u4f86\u6eff\u8db3\u5e02\u5834\u9700\u6c42\u3002\u4f8b\u5982 S183-SH0 \u9069\u7528\u65bc\u5168\u5feb\u9583\u5132\u5b58\u9663\u5217\u90e8\u5c6c\uff0c\u8f15\u8584\u7684 1U \u5916\u5f62\u642d\u8f09\u4e86\u5169\u9846 Intel Xeon \u53ef\u64f4\u5145\u8655\u7406\u5668\uff0c\u652f\u63f4\u9ad8\u9054 32 \u500b E1.S \u898f\u683c\u7684\u5168\u5feb\u9583\u786c\u789f\uff0c\u80fd\u5be6\u73fe\u5feb\u901f\u3001\u9ad8\u5bc6\u5ea6\u7684\u5132\u5b58\u914d\u7f6e\uff1b\u53e6\u4e00\u6b3e\u652f\u63f4 E1.S \u7684 H253-Z10 2U \u4f3a\u670d\u5668\uff0c\u5c07 I\/O \u914d\u7f6e\u8a2d\u8a08\u5728\u6a5f\u7bb1\u524d\u7de3\uff0c\u8b93\u7ba1\u7406\u8005\u80fd\u65b9\u4fbf\u7684\u5f9e\u524d\u7aef\u9032\u884c\u7db2\u8def\u7dad\u8b77\uff0c2U4 \u7bc0\u9ede\u7684\u904b\u7b97\u7bc0\u9ede\u652f\u63f4\u63a1\u7528 3D V-cache \u6280\u8853\u7684 AMD EPYC 9004 \u8655\u7406\u5668\uff0c\u80fd\u5b8c\u6210\u908a\u7de3\u7684 AI \u52a0\u901f\u904b\u7b97\uff0c\u63d0\u4f9b\u5373\u6642\u7684\u4eba\u5de5\u667a\u6167\u53cd\u994b\u3002\u540c\u6642\u9084\u6709\u5169\u6b3e G293 GPU \u4f3a\u670d\u5668\uff0c\u5c08\u70ba AI \u8a13\u7df4\u6216AI\u63a8\u8ad6\u5de5\u4f5c\u9032\u884c\u6700\u4f73\u5316\u8a2d\u8a08\u3002G293-Z43 \u5c08\u653b AI \u63a8\u8ad6\u9818\u57df\uff0c2U \u4f3a\u670d\u5668\u6a5f\u7bb1\u5167\u6709\u56db\u7d44\u53ef\u63d2\u62d4\u7684 GPU\u00a0\u64f4\u5145\u69fd\uff0c\u53ef\u652f\u63f4 16 \u5f35 Alveo\u2122 V70 \u52a0\u901f\u5668\u5b89\u88dd\uff0c\u800c\u6027\u50f9\u6bd4\u5408\u5b9c\u7684\u00a0G293-Z23 \u4f3a\u670d\u5668\u80fd\u652f\u63f4\u9ad8\u5de5\u8017\uff08TDP\uff09\u7684\u8655\u7406\u5668\u548c PCIe Gen4 \u548c Gen5 GPU\uff0c\u5982 NVIDIA L40S GPU<\/a>\uff0c\u80fd\u517c\u9867\u95dc\u9375\u578b\u751f\u7522\u529b\u5de5\u4f5c\u6216\u4f01\u696d\u5167 AI \u6f14\u7b97\u904b\u7528\u7684\u4efb\u52d9\u3002\u00a0<\/p>\n\n\n\n

<\/p>\n\n\n\n

NVIDIA H200 Tensor Core GPU <\/strong><\/h3>\n\n\n\n

NVIDIA \u540c\u6642\u5728 SC23 \u5c55\u6703\u4e0a\u5ba3\u544a\u4e86\u4e0b\u4e00\u4ee3\u7684 NVIDIA H200 Tensor Core GPU \u898f\u5283\uff0c\u76f8\u8f03\u73fe\u884c\u7684 GPU \u63d0\u4f9b\u66f4\u597d\u7684\u8a18\u61b6\u9ad4\u6548\u80fd\uff0c\u505a\u70ba NVIDIA \u7684\u6280\u8853\u5408\u4f5c\u5925\u4f34\uff0c\u6280\u5609\u4e5f\u5c07\u5728 H200 Tensor Core GPU \u4e0a\u5e02\u540c\u6642\u63a8\u51fa\u4e00\u7cfb\u5217\u7684\u4f3a\u670d\u5668\u4f86\u652f\u6301\u8a72 GPU\u3002\u00a0<\/p>\n\n\n\n

NVIDIA H200 GPU \u7a81\u7834\u6027\u7684\u6548\u80fd\u548c\u8a18\u61b6\u9ad4\u5bb9\u91cf\u80fd\u52a0\u901f\u751f\u6210\u5f0f AI \u548c HPC \u61c9\u7528\u3002\u4f5c\u70ba\u7b2c\u4e00\u500b\u64c1\u6709 HBM3e \u7684 GPU\uff0cNVIDIA\u00a0HGX H200<\/a> \u662f\u9818\u5148\u5168\u7403\u7684 AI \u904b\u7b97\u5e73\u53f0\uff0c\u4e00\u7d44 HGX-H200 8-GPU \u63d0\u4f9b\u8d85\u904e 32 petaflops\u7684FP8\u6df1\u5ea6\u5b78\u7fd2\u904b\u7b97\u548c 1.1TB \u7684\u9ad8\u983b\u5bec\u8a18\u61b6\u9ad4\uff0c\u5c07\u5be6\u73fe\u66f4\u5feb\u7684\u751f\u6210\u5f0f AI \u548c HPC \u61c9\u7528\u6210\u679c\uff0c\u63a8\u52d5\u4eba\u985e\u793e\u6703\u671d\u5411\u6578\u4f4d\u5316\u751f\u6d3b\u9081\u9032\u3002\u00a0<\/p>\n\n\n\n

<\/p>\n\n\n\n

<\/p>\n\n\n\n

\ud83d\udfe6\u73fe\u5728\u5c31\u52a0\u5165 ioioTIMES \u81c9\u66f8\u7c89\u7d72\u5718<\/a> \u66f4\u591a\u4e92\u52d5\u3001\u66f4\u591a\u597d\u5eb7\u650f\u62b5\u52a0!!<\/strong>
\ud83d\udfe6\u6211\u5011\u6709LINE TODAY<\/a>\u983b\u9053\u4e86\uff0c\u5feb\u4f86\u8ffd\u8e2a\u6211\u5011\u5427!!–\u6700\u65b0\u79d1\u6280\u65b0\u805e \u76e1\u5728\u4f60\u624b<\/strong><\/h4>\n","protected":false},"excerpt":{"rendered":"

NVIDIA\u7cbe\u7cb9\u800c\u751f\uff0c\u6280\u5609\u4f3a\u670d\u5668\u5e73\u53f0<\/p>\n","protected":false},"author":3,"featured_media":49409,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"rank_math_lock_modified_date":false,"footnotes":""},"categories":[10287,13],"tags":[7794,1284],"class_list":["post-49408","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-gigabyte-zone","category-news","tag-super-computing","tag-1284"],"_links":{"self":[{"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts\/49408"}],"collection":[{"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=49408"}],"version-history":[{"count":3,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts\/49408\/revisions"}],"predecessor-version":[{"id":74937,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts\/49408\/revisions\/74937"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/media\/49409"}],"wp:attachment":[{"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=49408"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=49408"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=49408"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}