{"id":266158,"date":"2017-09-08T11:30:00","date_gmt":"2017-09-08T09:30:00","guid":{"rendered":"https:\/\/cms.galaxiemedia.fr\/tomshardware\/2017\/09\/08\/nvidia-a-vendu-son-premier-dgx-1v-960-tflops-pour-149-000-dollars\/"},"modified":"2023-06-22T18:16:31","modified_gmt":"2023-06-22T16:16:31","slug":"nvidia-a-vendu-son-premier-dgx-1v-960-tflops-pour-149-000-dollars","status":"publish","type":"post","link":"https:\/\/www.tomshardware.fr\/nvidia-a-vendu-son-premier-dgx-1v-960-tflops-pour-149-000-dollars\/","title":{"rendered":"NVIDIA a vendu son premier DGX-1V : 960 TFLOPS pour 149 000 dollars !"},"content":{"rendered":"
Un unboxing \u00e0 149 000 dollars !<\/span><\/span><\/span><\/span><\/span><\/p>\n L’intelligence artificielle et le Deep Learning sont des termes que l’on trouve actuellement sur toutes les l\u00e8vres, ou presque. NVIDIA semble bien d\u00e9cid\u00e9 \u00e0 s’imposer comme un des leaders du march\u00e9 avec ses syst\u00e8mes d\u00e9di\u00e9s DGX-1 – lanc\u00e9 en d\u00e9cembre dernier – et DGX-1V, tout juste sorti des cartons. Le MGH & BWH Center for Clinical Data Science (CCDS), un institut de recherche sur les applications d’intelligence artificielle et de Deep Learning en soins de sant\u00e9, est le premier \u00e0 recevoir \u00ab son \u00bb DGX-1V, achet\u00e9 pour la modique somme de 149 000 dollars (hors taxes).<\/p>\n \u00c9volution du DGX-1 dot\u00e9 de puces \u00e0 architecture Pascal, le DGX-1V embarque huit cartes Tesla V100 bas\u00e9es sur la r\u00e9cente architecture Volta du constructeur. C\u00f4t\u00e9 performances, on passe donc de 170 TFLOPS en FP16 \u00e0 240 TFLOPS, et m\u00eame jusqu’\u00e0 960 TFLOPS en Deep Learning gr\u00e2ce aux nouveaux Tensors Core de l’architecture Volta.<\/p>\n Le reste du syst\u00e8me reste similaire \u00e0 la version Broadwell-EP du DGX-1, avec deux CPU Xeon, 512 Go de m\u00e9moire vive pour le syst\u00e8me et quatre SSD de 2 To. L’ensemble a besoin de plusieurs alimentations pour fonctionner, d’une puissance totale maximale de 3200W !<\/p>\n <\/span><\/span><\/span><\/span><\/p>\n \t\t\tDonn\u00e9es: 3x 1.92 To SSD RAID 0<\/td>\n<\/tr>\n Et un Neural Net CPU, c'est combien de TFLOPS ?<\/p>","protected":false},"author":9,"featured_media":266159,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"ep_exclude_from_search":false,"footnotes":""},"categories":[487,4071],"tags":[538,1026],"hubs":[],"acf":{"post_show_excerpt":true,"post_source":{"title":"Anandtech","url":"http:\/\/www.anandtech.com\/show\/11824\/nvidia-ships-first-volta-dgx-systems","target":""}},"yoast_head":"\nJusqu’\u00e0 960 TFLOPS gr\u00e2ce aux Tensors Core<\/h3>\n
\n\n
\n <\/th>\n DGX-1
\t\t\t(Volta) \t\t\t<\/th>\nDGX-1
\t\t\t(Pascal) \t\t\t<\/th>\nDGX-1
\t\t\t(Pascal, premi\u00e8re version) \t\t\t<\/th>\nDGX Station \t\t<\/th>\n<\/tr>\n \n Configuration GPU
<\/strong><\/td>\n8x Tesla V100<\/td>\n 8x Tesla P100<\/td>\n 4x Tesla V100<\/td>\n<\/tr>\n \n GPU FP16 Compute<\/strong><\/th>\n G\u00e9n\u00e9ral<\/strong><\/td>\n 240 TFLOPS<\/td>\n 170 TFLOPS<\/td>\n 120 TFLOPS
\t\t\t\u00a0<\/td>\n<\/tr>\n\n Deep Learning<\/strong><\/td>\n 960 TFLOPS<\/td>\n 480 TFLOPS<\/td>\n<\/tr>\n \n Configuration CPU
<\/strong><\/td>\n2x Intel Xeon E5-2698 v4
\t\t\t(20-core, Broadwell-EP)<\/td>\n2x Intel Xeon E5-2698 v3
\t\t\t(16 core, Haswell-EP)<\/td>\n1x Intel Xeon E5-2698 v4
\t\t\t(20-core, Broadwell-EP)<\/td>\n<\/tr>\n\n M\u00e9moire syst\u00e8me
<\/strong><\/td>\n512 Go DDR4-2133
\t\t\t(LRDIMM)<\/td>\n256 Go DDR4
\t\t\t(LRDIMM)<\/td>\n<\/tr>\n\n M\u00e9moire GPU totale
<\/strong><\/td>\n128 Go HBM2
\t\t\t(8x 16Go)<\/td>\n64 Go HBM2
\t\t\t(4x 16Go)<\/td>\n<\/tr>\n\n Stockage<\/strong><\/td>\n 4x 1.92 To SSD RAID 0<\/td>\n OS: 1x 1.92 To SSD<\/p>\n \n R\u00e9seau<\/strong><\/td>\n Dual 10GbE
\t\t\t4 InfiniBand EDR<\/td>\nDual 10Gb LAN<\/td>\n<\/tr>\n \n Alimentation<\/strong><\/td>\n 3200W<\/td>\n 1500W<\/td>\n<\/tr>\n \n Dimensions<\/strong><\/td>\n 866mm x 444mm x 131mm
\t\t\t(3U)<\/td>\n518mm x 256mm x 639mm
\t\t\t(Tour)<\/td>\n<\/tr>\n\n Divers<\/strong><\/td>\n Ubuntu Linux Host OS
DGX Software Stack<\/td>\nUbuntu Desktop Linux OS
DGX Software Stack
3x DisplayPort<\/td>\n<\/tr>\n\n Prix<\/strong><\/td>\n 149 000$<\/td>\n Variable<\/td>\n 129 000$<\/td>\n 69 000 $<\/td>\n<\/tr>\n<\/tbody>\n<\/table><\/div>\n","protected":false},"excerpt":{"rendered":"