{"id":17289,"date":"2023-08-25T15:21:02","date_gmt":"2023-08-25T08:21:02","guid":{"rendered":"https:\/\/www.thegioimaychu.vn\/blog\/?p=17289"},"modified":"2023-09-15T15:49:52","modified_gmt":"2023-09-15T08:49:52","slug":"gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu","status":"publish","type":"post","link":"https:\/\/thegioimaychu.vn\/blog\/ai-hpc\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\/","title":{"rendered":"GPU NVIDIA ki\u1ebfn tr\u00fac Ampere: So s\u00e1nh Multi-Instance GPU vs Virtual GPU"},"content":{"rendered":"<style>.iblink{color:#975!important;}.iblink:hover{color:orange!important;}<\/style><p>NVIDIA Ampere l\u00e0 m\u1ed9t trong hai th\u1ebf h\u1ec7 GPU m\u1edbi nh\u1ea5t c\u1ee7a NVIDIA. NVIDIA Ampere GPU tr\u00ean VMware vSphere 7 Update 2 (ho\u1eb7c cao h\u01a1n) c\u00f3 th\u1ec3 \u0111\u01b0\u1ee3c chia s\u1ebb gi\u1eefa c\u00e1c m\u00e1y \u1ea3o (VMs) \u1edf m\u1ed9t trong hai mode: Virtual GPU\u00a0(vGPU) mode c\u1ee7a VMware ho\u1eb7c <a class=\"iblink\" href=\"\/blog\/tong-hop\/mig-multi-instance-gpu-la-gi-p5634\/\">Multi-instance GPU<\/a> (MIG) mode c\u1ee7a NVIDIA. NVIDIA vGPU \u0111\u00e3 \u0111\u01b0\u1ee3c k\u00e8m theo trong b\u1ed9 ph\u1ea7n m\u1ec1m <a class=\"iblink\" href=\"\/solution\/infra\/ai\/bo-giai-phap-trien-khai-ai-cap-doanh-nghiep-nvidia-ai-enterprise-p4126\/\">NVIDIA AI Enterprise<\/a>, ph\u1ea7n m\u1ec1m \u0111\u01b0\u1ee3c ch\u1ee9ng nh\u1eadn cho VMware vSphere.<\/p>\n<p>Trong vGPU mode, b\u1ed9 nh\u1edb tr\u00ean GPU \u0111\u01b0\u1ee3c ph\u00e2n chia t\u0129nh nh\u01b0ng t\u00e0i nguy\u00ean t\u00ednh to\u00e1n (compute) \u0111\u01b0\u1ee3c chia s\u1ebb theo th\u1eddi gian (time-shared) gi\u1eefa c\u00e1c m\u00e1y \u1ea3o d\u00f9ng chung GPU.\u00a0\u1ede mode n\u00e0y, khi m\u1ed9t m\u00e1y \u1ea3o \u0111ang ch\u1ea1y tr\u00ean GPU, n\u00f3 s\u1edf h\u1eefu t\u1ea5t c\u1ea3 kh\u1ea3 n\u0103ng t\u00ednh to\u00e1n c\u1ee7a GPU nh\u01b0ng ch\u1ec9 c\u00f3 quy\u1ec1n truy c\u1eadp v\u00e0o ph\u1ea7n b\u1ed9 nh\u1edb GPU c\u1ee7a ri\u00eang n\u00f3.<\/p>\n<p>Trong MIG mode, c\u1ea3 b\u1ed9 nh\u1edb v\u00e0 kh\u1ea3 n\u0103ng t\u00ednh to\u00e1n s\u1ebd \u0111\u01b0\u1ee3c ph\u00e2n chia t\u0129nh.\u00a0Khi m\u1ed9t m\u00e1y \u1ea3o s\u1eed d\u1ee5ng m\u1ed9t GPU trong MIG mode, n\u00f3 ch\u1ec9 c\u00f3 th\u1ec3 truy c\u1eadp b\u1ed9 nh\u1edb v\u00e0 s\u1eed d\u1ee5ng c\u00e1c core t\u00ednh to\u00e1n \u0111\u01b0\u1ee3c g\u00e1n cho n\u00f3.\u00a0V\u00ec v\u1eady, ngay c\u1ea3 khi c\u00e1c core t\u00ednh to\u00e1n c\u00f2n l\u1ea1i kh\u00f4ng ho\u1ea1t \u0111\u1ed9ng (l\u00e0 c\u00e1c core kh\u00f4ng \u0111\u01b0\u1ee3c g\u00e1n cho m\u00e1y \u1ea3o n\u00e0y) trong GPU, m\u00e1y \u1ea3o c\u0169ng kh\u00f4ng th\u1ec3 s\u1eed d\u1ee5ng c\u00e1c core kh\u00f4ng ho\u1ea1t \u0111\u1ed9ng \u0111\u00f3.<\/p>\n<p>B\u1ea5t k\u1ec3 m\u00e1y \u1ea3o s\u1eed d\u1ee5ng mode n\u00e0o \u0111\u1ec3 th\u1ef1c thi workload c\u1ee7a n\u00f3, nh\u1eefng k\u1ebft qu\u1ea3 t\u00ednh to\u00e1n l\u00e0 nh\u01b0 nhau.\u00a0S\u1ef1 kh\u00e1c bi\u1ec7t duy nh\u1ea5t s\u1ebd l\u00e0 \u1edf hi\u1ec7u su\u1ea5t \u0111\u1ea1t \u0111\u01b0\u1ee3c, \u0111\u01b0\u1ee3c \u0111o b\u1eb1ng th\u1eddi gian th\u1ef1c. C\u1ea3 hai mode vGPU v\u00e0 MIG \u0111\u1ec1u c\u00f3 nh\u1eefng \u01b0u \u0111i\u1ec3m v\u00e0 nh\u01b0\u1ee3c \u0111i\u1ec3m t\u01b0\u01a1ng \u1ee9ng: vGPU mode chia s\u1ebb th\u1eddi gian c\u00e1c core t\u00ednh to\u00e1n, trong khi MIG mode ph\u00e2n chia t\u0129nh c\u00e1c core.\u00a0V\u1edbi s\u1ef1 kh\u00e1c bi\u1ec7t v\u1ec1 c\u00e1ch c\u00e1c core \u0111\u01b0\u1ee3c chia s\u1ebb b\u1edfi hai mode n\u00e0y, c\u00e2u h\u1ecfi \u0111\u1eb7t ra l\u00e0 ch\u1ebf \u0111\u1ed9 n\u00e0o mang l\u1ea1i hi\u1ec7u su\u1ea5t t\u1ed1t nh\u1ea5t (ngh\u0129a l\u00e0 th\u1eddi gian ch\u1ea1y th\u1ea5p nh\u1ea5t) cho m\u1ed9t workload nh\u1ea5t \u0111\u1ecbnh.\u00a0H\u00e3y c\u00f9ng t\u00ecm hi\u1ec3u.<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/thegioimaychu.vn\/blog\/wp-content\/uploads\/2023\/08\/tgmc-blog-64e369b577ce3.png\" alt=\"\" \/><em>B\u1ea3ng 1. T\u00f3m t\u1eaft k\u1ebft qu\u1ea3 t\u1eeb vi\u1ec7c ch\u1ea1y th\u1eed nghi\u1ec7m \u0111\u01a1n v\u1ecb v\u1edbi c\u00e1c profile compute v\u00e0 data transfer kh\u00e1c nhau.<\/em><\/p>\n<p>Ch\u00fang t\u00f4i \u0111\u00e3 ch\u1ea1y c\u00e1c th\u1eed nghi\u1ec7m \u0111\u01a1n v\u1ecb &#8211; l\u1ea5y c\u1ea3m h\u1ee9ng t\u1eeb v\u00ed d\u1ee5 nh\u00e2n ma tr\u1eadn (matrix multiplication) CUDA c\u1ed5 \u0111i\u1ec3n, v\u1edbi c\u00e1c t\u1ef7 l\u1ec7 t\u00ednh to\u00e1n (compute) cho \u0111\u1ebfn truy\u1ec1n d\u1eef li\u1ec7u (data transfer) kh\u00e1c nhau t\u1eeb host sang GPU v\u00e0 ng\u01b0\u1ee3c l\u1ea1i, \u0111\u1ec3 x\u00e1c \u0111\u1ecbnh ti\u00eau ch\u00ed cho c\u00e1c workload n\u00e0o s\u1ebd cho hi\u1ec7u su\u1ea5t t\u1ed1t h\u01a1n \u1edf MIG mode v\u00e0 ti\u00eau ch\u00ed n\u00e0o s\u1ebd ho\u1ea1t \u0111\u1ed9ng t\u1ed1t h\u01a1n \u1edf vGPU mode.\u00a0Th\u1eed nghi\u1ec7m c\u1ee7a ch\u00fang t\u00f4i \u0111\u00e3 s\u1eed d\u1ee5ng 10 ma tr\u1eadn, m\u1ed7i ma tr\u1eadn c\u00f3 k\u00edch th\u01b0\u1edbc 1000&#215;1000.<\/p>\n<p>C\u00e1c k\u1ebft qu\u1ea3 \u0111\u01b0\u1ee3c tr\u00ecnh b\u00e0y trong blog n\u00e0y cho th\u1ea5y r\u1eb1ng c\u00e1c workload th\u1ef1c thi CUDA kernels t\u00ednh to\u00e1n l\u1edbn, n\u1eb7ng m\u00e0 kh\u00f4ng b\u1ecb gi\u00e1n \u0111o\u1ea1n khi truy\u1ec1n d\u1eef li\u1ec7u ho\u1eb7c t\u00ednh to\u00e1n CPU, cho th\u1ea5y hi\u1ec7u su\u1ea5t \u1edf MIG mode t\u1ed1t h\u01a1n so v\u1edbi vGPU mode.\u00a0Nh\u01b0ng khi CUDA kernels \u0111\u01b0\u1ee3c xen k\u1ebd v\u1edbi truy\u1ec1n d\u1eef li\u1ec7u ho\u1eb7c gi\u00e1n \u0111o\u1ea1n \u0111\u1ec3 th\u1ef1c thi c\u00e1c t\u00ednh to\u00e1n c\u1ee7a CPU, vGPU mode s\u1ebd mang l\u1ea1i hi\u1ec7u su\u1ea5t t\u1ed1t h\u01a1n.<\/p>\n<p>V\u00ec v\u1eady, c\u00e1c \u0111\u1eb7c \u0111i\u1ec3m c\u1ee7a workload x\u00e1c \u0111\u1ecbnh GPU mode, vGPU hay MIG, s\u1ebd mang l\u1ea1i hi\u1ec7u su\u1ea5t t\u1ed1t nh\u1ea5t.<\/p>\n<h2>M\u00f4i tr\u01b0\u1eddng test, code v\u00e0 k\u1ebft qu\u1ea3<\/h2>\n<p><img decoding=\"async\" src=\"https:\/\/thegioimaychu.vn\/blog\/wp-content\/uploads\/2023\/08\/tgmc-blog-64e369b70b8ca.png\" alt=\"\" \/><em>H\u00ecnh 1. Pseudocode \u0111\u1ed1i v\u1edbi th\u1eed nghi\u1ec7m \u0111\u01a1n v\u1ecb &#8211; xen k\u1ebd truy\u1ec1n d\u1eef li\u1ec7u v\u1edbi t\u00ednh to\u00e1n CUDA.\u00a0B\u00e0i test c\u00f3 10 ma tr\u1eadn, m\u1ed7i ma tr\u1eadn c\u00f3 k\u00edch th\u01b0\u1edbc 1000&#215;1000.<\/em><\/p>\n<p>C\u00e1c th\u1eed nghi\u1ec7m \u0111\u00e3 \u0111\u01b0\u1ee3c ch\u1ea1y tr\u00ean Dell R740 (2 CPU Intel Xeon Gold 6140, 768GB RAM, SSD storage) v\u1edbi 2 GPU A100.\u00a0M\u1ed9t GPU \u0111\u01b0\u1ee3c c\u1ea5u h\u00ecnh \u1edf vGPU mode v\u00e0 GPU th\u1ee9 hai \u0111\u01b0\u1ee3c c\u1ea5u h\u00ecnh \u1edf MIG mode.\u00a0M\u00e3 gi\u1ea3 (pseudocode) cho b\u00e0i test \u0111\u1ea7u ti\u00ean \u0111\u01b0\u1ee3c hi\u1ec3n th\u1ecb trong h\u00ecnh 1.<\/p>\n<p>B\u00e0i test \u0111\u00e3 \u0111\u01b0\u1ee3c ch\u1ea1y v\u1edbi c\u00e1c t\u1ef7 l\u1ec7 truy\u1ec1n d\u1eef li\u1ec7u \u0111\u1ebfn t\u00ednh to\u00e1n kh\u00e1c nhau. C\u00e1c gi\u00e1 tr\u1ecb t\u1ef7 l\u1ec7 m\u00e0 ch\u00fang t\u00f4i \u0111\u00e3 test l\u00e0 0% data transfer, 10% data transfer, 20% data transfer v\u00e0 nh\u01b0 th\u1ebf v\u1edbi b\u01b0\u1edbc nh\u1ea3y l\u00e0 10 cho \u0111\u1ebfn 100% data transfer.\u00a0\u0110\u1ed1i v\u1edbi m\u1ed7i t\u1ef7 l\u1ec7 truy\u1ec1n d\u1eef li\u1ec7u \u0111\u1ebfn t\u00ednh to\u00e1n, ch\u00fang t\u00f4i \u0111\u00e3 thay \u0111\u1ed5i s\u1ed1 l\u01b0\u1ee3ng m\u00e1y \u1ea3o ch\u1ea1y \u0111\u1ed3ng th\u1eddi t\u1eeb m\u1ed9t cho \u0111\u1ebfn m\u1ee9c t\u1ed1i \u0111a \u0111\u01b0\u1ee3c ph\u00e9p \u0111\u1ed1i v\u1edbi profile vGPU ho\u1eb7c MIG \u0111\u00f3.<\/p>\n<p>V\u00ed d\u1ee5, v\u1edbi MIG profile a100-1-5c (bi\u1ec3u th\u1ecb profile v\u1edbi m\u1ed9t compute slice v\u00e0 b\u1ed9 nh\u1edb GPU 5 GB) n\u00ean t\u1ed1i \u0111a 7 m\u00e1y \u1ea3o c\u00f3 th\u1ec3 chia s\u1ebb GPU.\u00a0V\u00ec v\u1eady, v\u1edbi profile n\u00e0y, ch\u00fang t\u00f4i \u0111\u00e3 thay \u0111\u1ed5i s\u1ed1 l\u01b0\u1ee3ng m\u00e1y \u1ea3o t\u1eeb 1 \u0111\u1ebfn 7. MIG profile a100-2-10c c\u00f3 th\u1ec3 h\u1ed7 tr\u1ee3 t\u1ed1i \u0111a 3 m\u00e1y \u1ea3o.\u00a0V\u00ec v\u1eady, \u0111\u1ed1i v\u1edbi profile n\u00e0y, ch\u00fang t\u00f4i \u0111\u00e3 thay \u0111\u1ed5i s\u1ed1 l\u01b0\u1ee3ng m\u00e1y \u1ea3o t\u1eeb 1 \u0111\u1ebfn 3.<\/p>\n<p>Th\u1eddi gian ch\u1ea1y \u0111\u1ec3 th\u1ef1c hi\u1ec7n t\u00ednh to\u00e1n CUDA cho th\u1eed nghi\u1ec7m n\u00e0y v\u1edbi 10% truy\u1ec1n d\u1eef li\u1ec7u \u0111\u1ebfn 90% ho\u1ea1t \u0111\u1ed9ng t\u00ednh to\u00e1n \u0111\u01b0\u1ee3c hi\u1ec3n th\u1ecb trong H\u00ecnh 2 (b\u00ean d\u01b0\u1edbi).\u00a0T\u1eeb h\u00ecnh n\u00e0y, ch\u00fang ta c\u00f3 th\u1ec3 th\u1ea5y r\u1eb1ng khi truy\u1ec1n d\u1eef li\u1ec7u \u0111\u01b0\u1ee3c xen k\u1ebd v\u1edbi t\u00ednh to\u00e1n v\u00e0 t\u1ef7 l\u1ec7 truy\u1ec1n d\u1eef li\u1ec7u l\u00e0 10%, vGPU mode mang l\u1ea1i hi\u1ec7u su\u1ea5t t\u1ed1t h\u01a1n cho c\u00e1c t\u00ednh to\u00e1n CUDA.<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/thegioimaychu.vn\/blog\/wp-content\/uploads\/2023\/08\/tgmc-blog-64e369b822a34.png\" alt=\"\" \/>\u00a0<em>B\u1ea3ng 2. C\u00e1c c\u1ea5u h\u00ecnh ch\u00fang t\u00f4i \u0111\u00e3 s\u1eed d\u1ee5ng cho c\u1ea3 ba b\u1ed9 th\u1eed nghi\u1ec7m. vGPU mode c\u00f3 th\u1ec3 h\u1ed7 tr\u1ee3 h\u01a1n 7 m\u00e1y \u1ea3o ch\u1ea1y \u0111\u1ed3ng th\u1eddi trong profile A100-4c v\u00e0 A100-5c nh\u01b0ng v\u00ec s\u1ed1 l\u01b0\u1ee3ng m\u00e1y \u1ea3o t\u1ed1i \u0111a c\u00f3 th\u1ec3 ch\u1ea1y \u0111\u1ed3ng th\u1eddi v\u1edbi MIG mode l\u00e0 7 n\u00ean ch\u00fang t\u00f4i ch\u1ec9 test v\u1edbi t\u1ed1i \u0111a 7 m\u00e1y \u1ea3o ch\u1ea1y \u0111\u1ed3ng th\u1eddi.\u00a0Ch\u00fang t\u00f4i c\u0169ng kh\u00f4ng test v\u1edbi c\u00e1c mode MIG h\u1ed7n h\u1ee3p, ch\u1eb3ng h\u1ea1n nh\u01b0 trong m\u1ed9t VM v\u1edbi MIG mode A100-2-10c v\u00e0 VM th\u1ee9 hai v\u1edbi A100-1-5c.<\/em><img decoding=\"async\" src=\"https:\/\/thegioimaychu.vn\/blog\/wp-content\/uploads\/2023\/08\/tgmc-blog-64e369b9c24cc.png\" alt=\"\" \/><em>H\u00ecnh 2. Th\u1eddi gian ch\u1ea1y chu\u1ea9n h\u00f3a trung b\u00ecnh \u0111\u1ec3 th\u1ef1c thi c\u00e1c t\u00ednh to\u00e1n CUDA cho th\u1eed nghi\u1ec7m \u0111\u01b0\u1ee3c hi\u1ec3n th\u1ecb trong H\u00ecnh 1. Nh\u1eefng k\u1ebft qu\u1ea3 n\u00e0y thu \u0111\u01b0\u1ee3c v\u1edbi t\u1ef7 l\u1ec7 10% data transfer.\u00a0\u0110\u1ec3 chu\u1ea9n h\u00f3a th\u1eddi gian ch\u1ea1y, ch\u00fang t\u00f4i chia th\u1eddi gian ch\u1ea1y \u0111o \u0111\u01b0\u1ee3c cho b\u1ed9 d\u1eef li\u1ec7u n\u00e0y cho th\u1eddi gian t\u1ed1i \u0111a trong b\u1ed9 d\u1eef li\u1ec7u n\u00e0y.<\/em><\/p>\n<p>Trong h\u00ecnh 3 (b\u00ean d\u01b0\u1edbi), ch\u00fang t\u00f4i hi\u1ec3n th\u1ecb k\u1ebft qu\u1ea3 t\u1eeb th\u1eed nghi\u1ec7m n\u00e0y v\u1edbi t\u1ef7 l\u1ec7 50% data transfer v\u1edbi 50% compute.\u00a0C\u00e1c bi\u1ec3u \u0111\u1ed3 hi\u1ec3n th\u1ecb th\u1eddi gian ch\u1ea1y cho to\u00e0n b\u1ed9 b\u00e0i ki\u1ec3m tra v\u00e0 ch\u1ec9 cho c\u00e1c t\u00ednh to\u00e1n.\u00a0Trong h\u00ecnh 2 v\u00e0 3, ch\u00fang ta c\u00f3 th\u1ec3 th\u1ea5y r\u1eb1ng vGPU ho\u1ea1t \u0111\u1ed9ng t\u1ed1t h\u01a1n ch\u1ebf \u0111\u1ed9 MIG khi truy\u1ec1n d\u1eef li\u1ec7u \u0111\u01b0\u1ee3c xen k\u1ebd v\u1edbi t\u00ednh to\u00e1n v\u00e0 t\u1ef7 l\u1ec7 truy\u1ec1n d\u1eef li\u1ec7u l\u00e0 50%.\u00a0T\u1eeb d\u1eef li\u1ec7u \u1edf c\u00e1c gi\u00e1 tr\u1ecb kh\u00e1c c\u1ee7a t\u1ef7 l\u1ec7 truy\u1ec1n d\u1eef li\u1ec7u so v\u1edbi t\u00ednh to\u00e1n (kh\u00f4ng hi\u1ec3n th\u1ecb \u1edf \u0111\u00e2y), vGPU ho\u1ea1t \u0111\u1ed9ng t\u1ed1t h\u01a1n ch\u1ebf \u0111\u1ed9 MIG b\u1ea5t c\u1ee9 khi n\u00e0o t\u00ednh to\u00e1n CUDA \u0111\u01b0\u1ee3c xen k\u1ebd v\u1edbi truy\u1ec1n d\u1eef li\u1ec7u b\u1ea5t k\u1ec3 t\u1ef7 l\u1ec7 truy\u1ec1n d\u1eef li\u1ec7u so v\u1edbi t\u00ednh to\u00e1n CUDA.<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/thegioimaychu.vn\/blog\/wp-content\/uploads\/2023\/08\/tgmc-blog-64e369ba58afd.png\" alt=\"\" \/><em>H\u00ecnh 3. Th\u1eddi gian ch\u1ea1y chu\u1ea9n h\u00f3a trung b\u00ecnh \u0111\u1ec3 th\u1ef1c hi\u1ec7n t\u00ednh to\u00e1n CUDA v\u00e0 to\u00e0n b\u1ed9 th\u1eed nghi\u1ec7m cho th\u1eed nghi\u1ec7m \u0111\u01b0\u1ee3c th\u1ec3 hi\u1ec7n trong h\u00ecnh 1. Nh\u1eefng k\u1ebft qu\u1ea3 n\u00e0y thu \u0111\u01b0\u1ee3c v\u1edbi t\u1ef7 l\u1ec7 truy\u1ec1n d\u1eef li\u1ec7u 50%.\u00a0\u0110\u1ec3 chu\u1ea9n h\u00f3a th\u1eddi gian ch\u1ea1y, ch\u00fang t\u00f4i chia th\u1eddi gian ch\u1ea1y \u0111o \u0111\u01b0\u1ee3c cho t\u1eadp d\u1eef li\u1ec7u n\u00e0y cho th\u1eddi gian t\u1ed1i \u0111a trong t\u1eadp d\u1eef li\u1ec7u n\u00e0y.<\/em><\/p>\n<p><img decoding=\"async\" src=\"https:\/\/thegioimaychu.vn\/blog\/wp-content\/uploads\/2023\/08\/tgmc-blog-64e369baea507.png\" alt=\"\" \/><\/p>\n<p><em>H\u00ecnh 4. M\u00e3 gi\u1ea3 cho th\u1eed nghi\u1ec7m th\u1ef1c hi\u1ec7n c\u1ea3 truy\u1ec1n d\u1eef li\u1ec7u v\u00e0 t\u00ednh to\u00e1n, nh\u01b0ng c\u00e1c ho\u1ea1t \u0111\u1ed9ng \u0111\u01b0\u1ee3c t\u1ed5ng h\u1ee3p;\u00a0t\u1ea5t c\u1ea3 qu\u00e1 tr\u00ecnh truy\u1ec1n d\u1eef li\u1ec7u \u0111\u01b0\u1ee3c ho\u00e0n th\u00e0nh tr\u01b0\u1edbc khi th\u1ef1c hi\u1ec7n t\u00ednh to\u00e1n (ngh\u0129a l\u00e0 c\u00e1c ho\u1ea1t \u0111\u1ed9ng\u00a0KH\u00d4NG\u00a0\u0111\u01b0\u1ee3c xen k\u1ebd).\u00a0B\u00e0i ki\u1ec3m tra c\u00f3 m\u01b0\u1eddi ma tr\u1eadn, m\u1ed7i ma tr\u1eadn c\u00f3 k\u00edch th\u01b0\u1edbc 1000X1000.<\/em><\/p>\n<p>Trong lo\u1ea1t th\u1eed nghi\u1ec7m th\u1ee9 hai, ch\u00fang t\u00f4i \u0111\u00e3 thay \u0111\u1ed5i m\u00e3 \u0111\u1ec3 t\u1ea5t c\u1ea3 c\u00e1c ho\u1ea1t \u0111\u1ed9ng truy\u1ec1n d\u1eef li\u1ec7u \u0111\u01b0\u1ee3c ho\u00e0n th\u00e0nh tr\u01b0\u1edbc khi ch\u00fang t\u00f4i ch\u1ea1y b\u1ea5t k\u1ef3 ph\u00e9p t\u00ednh CUDA n\u00e0o (xem m\u00e3 gi\u1ea3 trong h\u00ecnh 4 \u1edf tr\u00ean).\u00a0\u0110\u1ed1i v\u1edbi lo\u1ea1i kh\u1ed1i l\u01b0\u1ee3ng c\u00f4ng vi\u1ec7c n\u00e0y, ch\u1ebf \u0111\u1ed9 MIG ho\u1ea1t \u0111\u1ed9ng t\u1ed1t h\u01a1n vGPU v\u1edbi t\u1ed1i \u0111a kho\u1ea3ng ba m\u00e1y \u1ea3o ch\u1ea1y \u0111\u1ed3ng th\u1eddi.\u00a0V\u1edbi h\u01a1n ba m\u00e1y \u1ea3o ch\u1ea1y \u0111\u1ed3ng th\u1eddi, s\u1ebd c\u00f3 m\u1ed9t \u0111i\u1ec3m giao nhau v\u00e0 ch\u1ebf \u0111\u1ed9 vGPU mang l\u1ea1i hi\u1ec7u su\u1ea5t t\u1ed1t h\u01a1n \u1edf m\u1ee9c \u0111\u1ed9 h\u1ee3p nh\u1ea5t cao h\u01a1n (ngh\u0129a l\u00e0 v\u1edbi nhi\u1ec1u h\u01a1n ba m\u00e1y \u1ea3o).\u00a0H\u00ecnh 5 cho th\u1ea5y d\u1eef li\u1ec7u t\u1eeb th\u1eed nghi\u1ec7m n\u00e0y.\u00a0D\u1eef li\u1ec7u \u0111\u01b0\u1ee3c c\u1eaft b\u1edbt \u1edf b\u1ed1n m\u00e1y \u1ea3o \u0111\u1ec3 l\u00e0m n\u1ed5i b\u1eadt \u0111i\u1ec3m giao nhau.<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/thegioimaychu.vn\/blog\/wp-content\/uploads\/2023\/08\/tgmc-blog-64e369bb7ecc2.png\" alt=\"\" \/><em>H\u00ecnh 5. Th\u1eddi gian ch\u1ea1y chu\u1ea9n h\u00f3a trung b\u00ecnh \u0111\u1ec3 th\u1ef1c hi\u1ec7n t\u00ednh to\u00e1n CUDA v\u00e0 cho to\u00e0n b\u1ed9 b\u00e0i ki\u1ec3m tra cho b\u00e0i ki\u1ec3m tra \u0111\u01b0\u1ee3c hi\u1ec3n th\u1ecb trong h\u00ecnh 4. \u0110\u1ec3 chu\u1ea9n h\u00f3a th\u1eddi gian ch\u1ea1y, ch\u00fang t\u00f4i chia th\u1eddi gian ch\u1ea1y \u0111o \u0111\u01b0\u1ee3c cho t\u1eadp d\u1eef li\u1ec7u n\u00e0y cho th\u1eddi gian t\u1ed1i \u0111a trong t\u1eadp d\u1eef li\u1ec7u n\u00e0y.\u00a0T\u1ed5ng th\u1eddi gian bao g\u1ed3m c\u1ea3 th\u1eddi gian t\u00ednh to\u00e1n v\u00e0 truy\u1ec1n d\u1eef li\u1ec7u.<\/em><\/p>\n<p>Trong lo\u1ea1t th\u1eed nghi\u1ec7m cu\u1ed1i c\u00f9ng, ch\u00fang t\u00f4i \u0111\u00e3 thay \u0111\u1ed5i m\u00e3 \u0111\u1ec3 ch\u1ec9 c\u00e1c t\u00ednh to\u00e1n CUDA \u0111\u01b0\u1ee3c th\u1ef1c thi.\u00a0M\u00e3 gi\u1ea3 \u0111\u01b0\u1ee3c hi\u1ec3n th\u1ecb trong h\u00ecnh 6 b\u00ean d\u01b0\u1edbi.<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/thegioimaychu.vn\/blog\/wp-content\/uploads\/2023\/08\/tgmc-blog-64e369bc43d49.png\" alt=\"\" \/><em>H\u00ecnh 6. M\u00e3 gi\u1ea3 cho th\u1eed nghi\u1ec7m ch\u1ec9 th\u1ef1c hi\u1ec7n t\u00ednh to\u00e1n CUDA ch\u1ec9 v\u1edbi m\u1ed9t l\u1ea7n truy\u1ec1n d\u1eef li\u1ec7u \u0111\u1ec3 t\u1ea3i d\u1eef li\u1ec7u khi b\u1eaft \u0111\u1ea7u.\u00a0B\u00e0i ki\u1ec3m tra c\u00f3 m\u01b0\u1eddi ma tr\u1eadn, m\u1ed7i ma tr\u1eadn c\u00f3 k\u00edch th\u01b0\u1edbc 1000X1000.<\/em><\/p>\n<p>Ch\u00fang t\u00f4i \u0111\u00e3 ch\u1ea1y th\u1eed nghi\u1ec7m n\u00e0y v\u1edbi t\u1ea5t c\u1ea3 c\u1ea5u h\u00ecnh vGPU v\u00e0 MIG.\u00a0K\u1ebft qu\u1ea3 th\u1ec3 hi\u1ec7n trong h\u00ecnh 7 (b\u00ean d\u01b0\u1edbi) cho th\u1ea5y r\u00f5 r\u00e0ng r\u1eb1ng ch\u1ebf \u0111\u1ed9 MIG mang l\u1ea1i hi\u1ec7u su\u1ea5t t\u1ed1t h\u01a1n \u1edf h\u1ea7u h\u1ebft m\u1ecdi c\u1ea5u h\u00ecnh v\u00e0 c\u1ea5u h\u00ecnh v\u1edbi kh\u1ed1i l\u01b0\u1ee3ng c\u00f4ng vi\u1ec7c n\u00e0y.<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/thegioimaychu.vn\/blog\/wp-content\/uploads\/2023\/08\/tgmc-blog-64e369bce21fb.png\" alt=\"\" \/><em>H\u00ecnh 7. T\u1ed5ng th\u1eddi gian ch\u1ea1y th\u1eed nghi\u1ec7m ch\u1ec9 v\u1edbi t\u00ednh to\u00e1n CUDA.\u00a0Ch\u1ebf \u0111\u1ed9 MIG mang l\u1ea1i hi\u1ec7u su\u1ea5t t\u1ed1t nh\u1ea5t \u1edf t\u1ea5t c\u1ea3 c\u00e1c c\u1ea5u h\u00ecnh v\u1edbi kh\u1ed1i l\u01b0\u1ee3ng c\u00f4ng vi\u1ec7c n\u00e0y.<\/em><\/p>\n<h2>K\u1ebft lu\u1eadn<\/h2>\n<p>NVIDIA A100 v\u00e0 A30 Tensor Core GPU tr\u00ean VMware vSphere h\u1ed7 tr\u1ee3 chia s\u1ebb m\u1ed9t GPU gi\u1eefa nhi\u1ec1u m\u00e1y \u1ea3o b\u1eb1ng hai mode: vGPU v\u00e0 MIG. S\u1ef1 kh\u00e1c bi\u1ec7t v\u1ec1 c\u00e1ch ph\u00e2n chia c\u00e1c core t\u00ednh to\u00e1n CUDA c\u00f3 th\u1ec3 g\u00e2y ra s\u1ef1 kh\u00e1c bi\u1ec7t v\u1ec1 hi\u1ec7u su\u1ea5t \u0111\u1ea1t \u0111\u01b0\u1ee3c khi s\u1eed d\u1ee5ng vGPU mode so v\u1edbi MIG mode cho c\u00f9ng m\u1ed9t workload.<\/p>\n<div>____<br \/><b>B\u00e0i vi\u1ebft li\u00ean quan<\/b><\/div>\n<ul>\n<li><a href=\"https:\/\/thegioimaychu.vn\/blog\/?post_type=post&#038;p=18236\">GPU NVIDIA L40S: S\u1ef1 \u0111\u1ed5i m\u1edbi v\u00e0 t\u00ednh hi\u1ec7u qu\u1ea3 cho Tr\u00ed Tu\u1ec7 Nh\u00e2n T\u1ea1o<\/a><\/li>\n<li><a href=\"https:\/\/thegioimaychu.vn\/blog\/?post_type=post&#038;p=15658\">V\u00ec sao ARM \u0111ang d\u1ea7n ph\u1ed5 bi\u1ebfn h\u01a1n trong c\u00e1c Trung t\u00e2m d\u1eef li\u1ec7u v\u00e0 HPC?<\/a><\/li>\n<li><a href=\"https:\/\/thegioimaychu.vn\/blog\/?post_type=post&#038;p=14610\">NVIDIA HGX 8x A100: N\u1ec1n t\u1ea3ng t\u0103ng t\u1ed1c m\u1ea1nh m\u1ebd cho AI v\u00e0 ChatGPT<\/a><\/li>\n<li><a href=\"https:\/\/thegioimaychu.vn\/blog\/?post_type=post&#038;p=14631\">Nhu c\u1ea7u GPU NVIDIA v\u01b0\u1ee3t qu\u00e1 ngu\u1ed3n cung khi ChatGPT v\u00e0 c\u00e1c h\u00e3ng \u0111\u1ea7u t\u01b0 m\u1ea1nh v\u00e0o AI<\/a><\/li>\n<li><a href=\"https:\/\/thegioimaychu.vn\/blog\/?post_type=post&#038;p=14414\">GPU NVIDIA A100 \u0111\u1ea1t k\u1ebft qu\u1ea3 t\u1ed1t nh\u1ea5t trong ch\u1ec9 s\u1ed1 Suy Lu\u1eadn AI cho ng\u00e0nh D\u1ecbch V\u1ee5 T\u00e0i Ch\u00ednh<\/a><\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>NVIDIA Ampere l&agrave; m&#7897;t trong hai th&#7871; h&#7879; GPU m&#7899;i nh&#7845;t c&#7911;a NVIDIA. NVIDIA Ampere GPU tr&ecirc;n VMware vSphere 7 Update 2 (ho&#7863;c cao h&#417;n) c&oacute; th&#7875; &#273;&#432;&#7907;c chia s&#7867; gi&#7919;a c&aacute;c m&aacute;y &#7843;o (VMs) &#7903; m&#7897;t trong hai mode: Virtual GPU&nbsp;(vGPU) mode c&#7911;a VMware ho&#7863;c Multi-instance GPU (MIG) mode c&#7911;a NVIDIA. NVIDIA vGPU &#273;&atilde;&#8230;<\/p>\n","protected":false},"author":29,"featured_media":17596,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"inline_featured_image":false,"footnotes":""},"categories":[3],"tags":[701,316,376,434],"class_list":["post-17289","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-hpc","tag-a100","tag-ampere","tag-mig","tag-vgpu"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v28.3 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>GPU NVIDIA ki\u1ebfn tr\u00fac Ampere: So s\u00e1nh Multi-Instance GPU vs Virtual GPU - Blog | TheGioiMayChu<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/thegioimaychu.vn\/blog\/ai-hpc\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\/\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:title\" content=\"GPU NVIDIA ki\u1ebfn tr\u00fac Ampere: So s\u00e1nh Multi-Instance GPU vs Virtual GPU - Blog | TheGioiMayChu\" \/>\n<meta name=\"twitter:description\" content=\"NVIDIA Ampere l&agrave; m&#7897;t trong hai th&#7871; h&#7879; GPU m&#7899;i nh&#7845;t c&#7911;a NVIDIA. NVIDIA Ampere GPU tr&ecirc;n VMware vSphere 7 Update 2 (ho&#7863;c cao h&#417;n) c&oacute; th&#7875; &#273;&#432;&#7907;c chia s&#7867; gi&#7919;a c&aacute;c m&aacute;y &#7843;o (VMs) &#7903; m&#7897;t trong hai mode: Virtual GPU&nbsp;(vGPU) mode c&#7911;a VMware ho&#7863;c Multi-instance GPU (MIG) mode c&#7911;a NVIDIA. NVIDIA vGPU &#273;&atilde;...\" \/>\n<meta name=\"twitter:image\" content=\"https:\/\/thegioimaychu.vn\/blog\/wp-content\/uploads\/2023\/08\/TheGioiMayChu-Blog-GPU-board.jpg\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"TGMC Blog\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"12 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/ai-hpc\\\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/ai-hpc\\\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\\\/\"},\"author\":{\"name\":\"TGMC Blog\",\"@id\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/#\\\/schema\\\/person\\\/777f41403962330933a4fbcc80bec570\"},\"headline\":\"GPU NVIDIA ki\u1ebfn tr\u00fac Ampere: So s\u00e1nh Multi-Instance GPU vs Virtual GPU\",\"datePublished\":\"2023-08-25T08:21:02+00:00\",\"dateModified\":\"2023-09-15T08:49:52+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/ai-hpc\\\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\\\/\"},\"wordCount\":2500,\"publisher\":{\"@id\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/ai-hpc\\\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/wp-content\\\/uploads\\\/2023\\\/08\\\/TheGioiMayChu-Blog-GPU-board.jpg\",\"keywords\":[\"a100\",\"ampere\",\"mig\",\"vgpu\"],\"articleSection\":[\"AI - HPC\"],\"inLanguage\":\"en-US\"},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/ai-hpc\\\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\\\/\",\"url\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/ai-hpc\\\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\\\/\",\"name\":\"GPU NVIDIA ki\u1ebfn tr\u00fac Ampere: So s\u00e1nh Multi-Instance GPU vs Virtual GPU - Blog | TheGioiMayChu\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/ai-hpc\\\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/ai-hpc\\\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/wp-content\\\/uploads\\\/2023\\\/08\\\/TheGioiMayChu-Blog-GPU-board.jpg\",\"datePublished\":\"2023-08-25T08:21:02+00:00\",\"dateModified\":\"2023-09-15T08:49:52+00:00\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/ai-hpc\\\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\\\/#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/ai-hpc\\\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/ai-hpc\\\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\\\/#primaryimage\",\"url\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/wp-content\\\/uploads\\\/2023\\\/08\\\/TheGioiMayChu-Blog-GPU-board.jpg\",\"contentUrl\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/wp-content\\\/uploads\\\/2023\\\/08\\\/TheGioiMayChu-Blog-GPU-board.jpg\",\"width\":740,\"height\":374},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/ai-hpc\\\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Trang ch\u1ee7\",\"item\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"AI - HPC\",\"item\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/.\\\/ai-hpc\\\/\"},{\"@type\":\"ListItem\",\"position\":3,\"name\":\"GPU NVIDIA ki\u1ebfn tr\u00fac Ampere: So s\u00e1nh Multi-Instance GPU vs Virtual GPU\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/#website\",\"url\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/\",\"name\":\"Blog | TheGioiMayChu\",\"description\":\"C\u1eadp nh\u1eadt xu h\u01b0\u1edbng c\u00f4ng ngh\u1ec7 m\u00e1y ch\u1ee7\",\"publisher\":{\"@id\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/#organization\",\"name\":\"Blog | TheGioiMayChu\",\"url\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/wp-content\\\/uploads\\\/2018\\\/08\\\/cropped-ntcnlog-logo-2.png\",\"contentUrl\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/wp-content\\\/uploads\\\/2018\\\/08\\\/cropped-ntcnlog-logo-2.png\",\"width\":500,\"height\":73,\"caption\":\"Blog | TheGioiMayChu\"},\"image\":{\"@id\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/#\\\/schema\\\/logo\\\/image\\\/\"},\"sameAs\":[\"https:\\\/\\\/www.facebook.com\\\/thegioimaychu\\\/\"]},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/#\\\/schema\\\/person\\\/777f41403962330933a4fbcc80bec570\",\"name\":\"TGMC Blog\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/ff179df3aeac078da5e07e40f2e014e99ab4120e2276b687bbe7e6964d99655e?s=96&d=mm&r=g\",\"url\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/ff179df3aeac078da5e07e40f2e014e99ab4120e2276b687bbe7e6964d99655e?s=96&d=mm&r=g\",\"contentUrl\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/ff179df3aeac078da5e07e40f2e014e99ab4120e2276b687bbe7e6964d99655e?s=96&d=mm&r=g\",\"caption\":\"TGMC Blog\"},\"url\":\"https:\\\/\\\/thegioimaychu.vn\\\/blog\\\/author\\\/tgmc\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"GPU NVIDIA ki\u1ebfn tr\u00fac Ampere: So s\u00e1nh Multi-Instance GPU vs Virtual GPU - Blog | TheGioiMayChu","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/thegioimaychu.vn\/blog\/ai-hpc\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\/","twitter_card":"summary_large_image","twitter_title":"GPU NVIDIA ki\u1ebfn tr\u00fac Ampere: So s\u00e1nh Multi-Instance GPU vs Virtual GPU - Blog | TheGioiMayChu","twitter_description":"NVIDIA Ampere l&agrave; m&#7897;t trong hai th&#7871; h&#7879; GPU m&#7899;i nh&#7845;t c&#7911;a NVIDIA. NVIDIA Ampere GPU tr&ecirc;n VMware vSphere 7 Update 2 (ho&#7863;c cao h&#417;n) c&oacute; th&#7875; &#273;&#432;&#7907;c chia s&#7867; gi&#7919;a c&aacute;c m&aacute;y &#7843;o (VMs) &#7903; m&#7897;t trong hai mode: Virtual GPU&nbsp;(vGPU) mode c&#7911;a VMware ho&#7863;c Multi-instance GPU (MIG) mode c&#7911;a NVIDIA. NVIDIA vGPU &#273;&atilde;...","twitter_image":"https:\/\/thegioimaychu.vn\/blog\/wp-content\/uploads\/2023\/08\/TheGioiMayChu-Blog-GPU-board.jpg","twitter_misc":{"Written by":"TGMC Blog","Est. reading time":"12 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/thegioimaychu.vn\/blog\/ai-hpc\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\/#article","isPartOf":{"@id":"https:\/\/thegioimaychu.vn\/blog\/ai-hpc\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\/"},"author":{"name":"TGMC Blog","@id":"https:\/\/thegioimaychu.vn\/blog\/#\/schema\/person\/777f41403962330933a4fbcc80bec570"},"headline":"GPU NVIDIA ki\u1ebfn tr\u00fac Ampere: So s\u00e1nh Multi-Instance GPU vs Virtual GPU","datePublished":"2023-08-25T08:21:02+00:00","dateModified":"2023-09-15T08:49:52+00:00","mainEntityOfPage":{"@id":"https:\/\/thegioimaychu.vn\/blog\/ai-hpc\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\/"},"wordCount":2500,"publisher":{"@id":"https:\/\/thegioimaychu.vn\/blog\/#organization"},"image":{"@id":"https:\/\/thegioimaychu.vn\/blog\/ai-hpc\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\/#primaryimage"},"thumbnailUrl":"https:\/\/thegioimaychu.vn\/blog\/wp-content\/uploads\/2023\/08\/TheGioiMayChu-Blog-GPU-board.jpg","keywords":["a100","ampere","mig","vgpu"],"articleSection":["AI - HPC"],"inLanguage":"en-US"},{"@type":"WebPage","@id":"https:\/\/thegioimaychu.vn\/blog\/ai-hpc\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\/","url":"https:\/\/thegioimaychu.vn\/blog\/ai-hpc\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\/","name":"GPU NVIDIA ki\u1ebfn tr\u00fac Ampere: So s\u00e1nh Multi-Instance GPU vs Virtual GPU - Blog | TheGioiMayChu","isPartOf":{"@id":"https:\/\/thegioimaychu.vn\/blog\/#website"},"primaryImageOfPage":{"@id":"https:\/\/thegioimaychu.vn\/blog\/ai-hpc\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\/#primaryimage"},"image":{"@id":"https:\/\/thegioimaychu.vn\/blog\/ai-hpc\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\/#primaryimage"},"thumbnailUrl":"https:\/\/thegioimaychu.vn\/blog\/wp-content\/uploads\/2023\/08\/TheGioiMayChu-Blog-GPU-board.jpg","datePublished":"2023-08-25T08:21:02+00:00","dateModified":"2023-09-15T08:49:52+00:00","breadcrumb":{"@id":"https:\/\/thegioimaychu.vn\/blog\/ai-hpc\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\/#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/thegioimaychu.vn\/blog\/ai-hpc\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\/"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/thegioimaychu.vn\/blog\/ai-hpc\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\/#primaryimage","url":"https:\/\/thegioimaychu.vn\/blog\/wp-content\/uploads\/2023\/08\/TheGioiMayChu-Blog-GPU-board.jpg","contentUrl":"https:\/\/thegioimaychu.vn\/blog\/wp-content\/uploads\/2023\/08\/TheGioiMayChu-Blog-GPU-board.jpg","width":740,"height":374},{"@type":"BreadcrumbList","@id":"https:\/\/thegioimaychu.vn\/blog\/ai-hpc\/gpu-nvidia-kien-truc-ampere-so-sanh-multi-instance-gpu-vs-virtual-gpu-p17289\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Trang ch\u1ee7","item":"https:\/\/thegioimaychu.vn\/blog\/"},{"@type":"ListItem","position":2,"name":"AI - HPC","item":"https:\/\/thegioimaychu.vn\/blog\/.\/ai-hpc\/"},{"@type":"ListItem","position":3,"name":"GPU NVIDIA ki\u1ebfn tr\u00fac Ampere: So s\u00e1nh Multi-Instance GPU vs Virtual GPU"}]},{"@type":"WebSite","@id":"https:\/\/thegioimaychu.vn\/blog\/#website","url":"https:\/\/thegioimaychu.vn\/blog\/","name":"Blog | TheGioiMayChu","description":"C\u1eadp nh\u1eadt xu h\u01b0\u1edbng c\u00f4ng ngh\u1ec7 m\u00e1y ch\u1ee7","publisher":{"@id":"https:\/\/thegioimaychu.vn\/blog\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/thegioimaychu.vn\/blog\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/thegioimaychu.vn\/blog\/#organization","name":"Blog | TheGioiMayChu","url":"https:\/\/thegioimaychu.vn\/blog\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/thegioimaychu.vn\/blog\/#\/schema\/logo\/image\/","url":"https:\/\/thegioimaychu.vn\/blog\/wp-content\/uploads\/2018\/08\/cropped-ntcnlog-logo-2.png","contentUrl":"https:\/\/thegioimaychu.vn\/blog\/wp-content\/uploads\/2018\/08\/cropped-ntcnlog-logo-2.png","width":500,"height":73,"caption":"Blog | TheGioiMayChu"},"image":{"@id":"https:\/\/thegioimaychu.vn\/blog\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/www.facebook.com\/thegioimaychu\/"]},{"@type":"Person","@id":"https:\/\/thegioimaychu.vn\/blog\/#\/schema\/person\/777f41403962330933a4fbcc80bec570","name":"TGMC Blog","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/secure.gravatar.com\/avatar\/ff179df3aeac078da5e07e40f2e014e99ab4120e2276b687bbe7e6964d99655e?s=96&d=mm&r=g","url":"https:\/\/secure.gravatar.com\/avatar\/ff179df3aeac078da5e07e40f2e014e99ab4120e2276b687bbe7e6964d99655e?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/ff179df3aeac078da5e07e40f2e014e99ab4120e2276b687bbe7e6964d99655e?s=96&d=mm&r=g","caption":"TGMC Blog"},"url":"https:\/\/thegioimaychu.vn\/blog\/author\/tgmc\/"}]}},"_links":{"self":[{"href":"https:\/\/thegioimaychu.vn\/blog\/wp-json\/wp\/v2\/posts\/17289","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/thegioimaychu.vn\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/thegioimaychu.vn\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/thegioimaychu.vn\/blog\/wp-json\/wp\/v2\/users\/29"}],"replies":[{"embeddable":true,"href":"https:\/\/thegioimaychu.vn\/blog\/wp-json\/wp\/v2\/comments?post=17289"}],"version-history":[{"count":9,"href":"https:\/\/thegioimaychu.vn\/blog\/wp-json\/wp\/v2\/posts\/17289\/revisions"}],"predecessor-version":[{"id":17368,"href":"https:\/\/thegioimaychu.vn\/blog\/wp-json\/wp\/v2\/posts\/17289\/revisions\/17368"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/thegioimaychu.vn\/blog\/wp-json\/wp\/v2\/media\/17596"}],"wp:attachment":[{"href":"https:\/\/thegioimaychu.vn\/blog\/wp-json\/wp\/v2\/media?parent=17289"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/thegioimaychu.vn\/blog\/wp-json\/wp\/v2\/categories?post=17289"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/thegioimaychu.vn\/blog\/wp-json\/wp\/v2\/tags?post=17289"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}