The Greenest Generation: NVIDIA, Intel and Partners Supercharge AI Computing Efficiency

Acce짯le짯ra짯ted NVIDIA Hop짯per sys짯tems with 4th Gen Intel Xeon Sca짯lable pro짯ces짯sors inclu짯ding NVIDIA DGX H100 and 60+ sys짯tems from NVIDIA part짯ners pro짯vi짯de 25x more effi짯ci짯en짯cy than tra짯di짯tio짯nal data cen짯ter ser짯vers to save big on ener짯gy costs.

AI is at the heart of humanity셲 most trans짯for짯ma짯ti짯ve inno짯va짯tions from deve짯lo짯ping COVID vac짯ci짯nes at unpre짯ce짯den짯ted speeds and dia짯gno짯sing can짯cer to powe짯ring auto짯no짯mous vehic짯les and under짯stan짯ding cli짯ma짯te change.

Vir짯tual짯ly every indus짯try will bene짯fit from adop짯ting AI, but the tech짯no짯lo짯gy has beco짯me more resour짯ce inten짯si짯ve as neu짯ral net짯works have increased in com짯ple짯xi짯ty. To avo짯id pla짯cing unsus짯tainable demands on elec짯tri짯ci짯ty gene짯ra짯ti짯on to run this com짯pu짯ting infra짯struc짯tu짯re, the under짯ly짯ing tech짯no짯lo짯gy must be as effi짯ci짯ent as possible.

Acce짯le짯ra짯ted com짯pu짯ting powered by NVIDIA GPUs and the NVIDIA AI plat짯form offer the effi짯ci짯en짯cy that enables data cen짯ters to sus짯tain짯ab짯ly dri짯ve the next gene짯ra짯ti짯on of breakthroughs.

And now, timed with the launch of 4th Gen Intel Xeon Sca짯lable pro짯ces짯sors, NVIDIA and its part짯ners have kicked off a new gene짯ra짯ti짯on of acce짯le짯ra짯ted com짯pu짯ting sys짯tems that are built for ener짯gy-effi짯ci짯ent AI. When com짯bi짯ned with NVIDIA H100 Ten짯sor Core GPUs, the짯se sys짯tems can deli짯ver dra짯ma짯ti짯cal짯ly hig짯her per짯for짯mance, grea짯ter sca짯le and hig짯her effi짯ci짯en짯cy than the pri짯or gene짯ra짯ti짯on, pro짯vi짯ding more com짯pu짯ta짯ti짯on and pro짯blem-sol짯ving per watt.

The new Intel CPUs will be used in NVIDIA DGX H100 sys짯tems, as well as in more than 60 ser짯vers fea짯turing H100 GPUs from NVIDIA part짯ners around the world.

Supercharging Speed, Efficiency and Savings for Enterprise AI

The coming NVIDIA and Intel-powered sys짯tems will help enter짯pri짯ses run workloads an avera짯ge of 25x more effi짯ci짯ent짯ly than tra짯di짯tio짯nal CPU-only data cen짯ter ser짯vers. This incre짯di짯ble per짯for짯mance per watt means less power is nee짯ded to get jobs done, which helps ensu짯re the power available to data cen짯ters is used as effi짯ci짯ent짯ly as pos짯si짯ble to super짯char짯ge the most important work.

Com짯pared to pri짯or-gene짯ra짯ti짯on acce짯le짯ra짯ted sys짯tems, this new gene짯ra짯ti짯on of NVI짯DIA-acce짯le짯ra짯ted ser짯vers speed trai짯ning and infe짯rence to boost ener짯gy effi짯ci짯en짯cy by 3.5x which trans짯la짯tes into real cost savings, with AI data cen짯ters deli짯ve짯ring over 3x lower total cost of ownership.

New 4th Gen Intel Xeon CPUs Move More Data to Accelerate NVIDIA AI

Among the fea짯tures of the new 4th Gen Intel Xeon CPU is sup짯port for PCIe Gen 5, which can dou짯ble the data trans짯fer rates from CPU to NVIDIA GPUs and net짯wor짯king. Increased PCIe lanes allow for a grea짯ter den짯si짯ty of GPUs and high-speed net짯wor짯king within each server.

Fas짯ter memo짯ry band짯width also impro짯ves the per짯for짯mance of data-inten짯si짯ve workloads such as AI, while net짯wor짯king speeds up to 400 giga짯bits per second (Gbps) per con짯nec짯tion sup짯port fas짯ter data trans짯fers bet짯ween ser짯vers and storage.

NVIDIA DGX H100 sys짯tems and ser짯vers from NVIDIA part짯ners with H100 PCIe GPUs come with a licen짯se for NVIDIA AI Enter짯pri짯se, an end-to-end, secu짯re, cloud-nati짯ve suite of AI deve짯lo짯p짯ment and deploy짯ment soft짯ware, pro짯vi짯ding a com짯ple짯te plat짯form for excel짯lence in effi짯ci짯ent enter짯pri짯se AI.

NVIDIA DGX H100 Systems Supercharge Efficiency for Supersize AI

As the fourth gene짯ra짯ti짯on of the world셲 pre짯mier pur짯po짯se-built AI infra짯struc짯tu짯re, NVIDIA DGX H100 sys짯tems pro짯vi짯de a ful짯ly opti짯mi짯zed plat짯form powered by the ope짯ra짯ting sys짯tem of the acce짯le짯ra짯ted data cen짯ter, NVIDIA Base Com짯mand software.

Each DGX H100 sys짯tem fea짯tures eight NVIDIA H100 GPUs, 10 NVIDIA ConnectX7 net짯work adap짯ters and dual 4th Gen Intel Xeon Sca짯lable pro짯ces짯sors to deli짯ver the per짯for짯mance requi짯red to build lar짯ge gene짯ra짯ti짯ve AI models, lar짯ge lan짯guage modelsrecom짯men짯der sys짯tems and more.

Com짯bi짯ned with NVIDIA net짯wor짯king, this archi짯tec짯tu짯re super짯char짯ges effi짯ci짯ent com짯pu짯ting at sca짯le by deli짯ve짯ring up to 9x more per짯for짯mance than the pre짯vious gene짯ra짯ti짯on and 20x to 40x more per짯for짯mance than unac짯ce짯le짯ra짯ted X86 dual-socket ser짯vers for AI trai짯ning and HPC workloads. If a lan짯guage model pre짯vious짯ly requi짯red 40 days to train on a clus짯ter of X86-only ser짯vers, the NVIDIA DGX H100 using Intel Xeon CPUs and ConnectX7 powered net짯wor짯king could com짯ple짯te the same work in as litt짯le as 12 days.

NVIDIA DGX H100 sys짯tems are the buil짯ding blocks of an enter짯pri짯se-rea짯dy, turn짯key NVIDIA DGX Super짯POD, which deli짯vers up to one exa짯flop of AI per짯for짯mance, pro짯vi짯ding a leap in effi짯ci짯en짯cy for lar짯ge-sca짯le enter짯pri짯se AI deployment.

NVIDIA Partners Boost Data Center Efficiency 

For AI data cen짯ter workloads, NVIDIA H100 GPUs enable enter짯pri짯ses to build and deploy appli짯ca짯ti짯ons more efficiently.

Brin짯ging a new gene짯ra짯ti짯on of per짯for짯mance and ener짯gy effi짯ci짯en짯cy to enter짯pri짯ses world짯wi짯de, a broad port짯fo짯lio of sys짯tems with H100 GPUs and 4th Gen Intel Xeon Sca짯lable CPUs are coming soon from NVIDIA part짯ners, inclu짯ding ASUS, Atos, Cis짯co, Dell Tech짯no짯lo짯gies, Fuji짯tsu, GIGABYTE, Hew짯lett Packard Enter짯pri짯se, Leno짯vo, QCT and Supermicro.

As the bell짯we짯ther of the effi짯ci짯en짯cy gains to come, the Fla짯ti짯ron Institute셲 Leno짯vo Think짯Sys짯tem with NVIDIA H100 GPUs tops the latest Green500 list and NVIDIA tech짯no짯lo짯gies power 23 of the top 30 sys짯tems on the list. The Fla짯ti짯ron sys짯tem uses pri짯or-gene짯ra짯ti짯on Intel CPUs, so even more effi짯ci짯en짯cy is expec짯ted from the sys짯tems now coming to market.

Addi짯tio짯nal짯ly, con짯nec짯ting ser짯vers with NVIDIA ConnectX7 net짯wor짯king and Intel 4th Gen Xeon Sca짯lable pro짯ces짯sors will increase effi짯ci짯en짯cy and redu짯ce infra짯struc짯tu짯re and power consumption.

NVIDIA ConnectX7 adap짯ters sup짯port PCIe Gen 5 and 400 Gbps per con짯nec짯tion using Ether짯net or Infi짯ni짯Band, doubling net짯wor짯king through짯put bet짯ween ser짯vers and to sto짯rage. The adap짯ters sup짯port advan짯ced net짯wor짯king, sto짯rage and secu짯ri짯ty off짯loads. ConnectX7 redu짯ces the num짯ber of cables and switch ports nee짯ded, saving 17% or more on elec짯tri짯ci짯ty nee짯ded for the net짯wor짯king of lar짯ge GPU-acce짯le짯ra짯ted HPC and AI clus짯ters and con짯tri짯bu짯ting to the bet짯ter ener짯gy effi짯ci짯en짯cy of the짯se new servers.

NVIDIA AI Enterprise Software Delivers Full-Stack AI Solution

The짯se next-gene짯ra짯ti짯on sys짯tems also deli짯ver a leap for짯ward in ope짯ra짯tio짯nal effi짯ci짯en짯cy as they셱e opti짯mi짯zed for the NVIDIA AI Enter짯pri짯se soft짯ware suite.

Run짯ning on NVIDIA H100, NVIDIA AI Enter짯pri짯se acce짯le짯ra짯tes the data sci짯ence pipe짯line and stream짯li짯nes the deve짯lo짯p짯ment and deploy짯ment of pre짯dic짯ti짯ve AI models to auto짯ma짯te essen짯ti짯al pro짯ces짯ses and gain rapid insights from data.

With an exten짯si짯ve libra짯ry of full-stack soft짯ware, inclu짯ding AI work짯flows of refe짯rence appli짯ca짯ti짯ons, frame짯works, pre짯trai짯ned models and infra짯struc짯tu짯re opti짯miza짯ti짯on, the soft짯ware pro짯vi짯des an ide짯al foun짯da짯ti짯on for sca짯ling enter짯pri짯se AI success.

To try out NVIDIA H100 run짯ning AI work짯flows and frame짯works sup짯port짯ed in NVIDIA AI Enter짯pri짯se, sign up for NVIDIA Launch짯Pad free of charge.

Watch NVIDIA foun짯der and CEO Jen짯sen Huang speak at the 4th Gen Intel Xeon Sca짯lable pro짯ces짯sor launch event.