Acce­le­ra­ted NVIDIA Hop­per sys­tems with 4th Gen Intel Xeon Sca­lable pro­ces­sors — inclu­ding NVIDIA DGX H100 and 60+ sys­tems from NVIDIA part­ners — pro­vi­de 25x more effi­ci­en­cy than tra­di­tio­nal data cen­ter ser­vers to save big on ener­gy costs.

AI is at the heart of humanity’s most trans­for­ma­ti­ve inno­va­tions — from deve­lo­ping COVID vac­ci­nes at unpre­ce­den­ted speeds and dia­gno­sing can­cer to powe­ring auto­no­mous vehic­les and under­stan­ding cli­ma­te change.

Vir­tual­ly every indus­try will bene­fit from adop­ting AI, but the tech­no­lo­gy has beco­me more resour­ce inten­si­ve as neu­ral net­works have increased in com­ple­xi­ty. To avo­id pla­cing unsus­tainable demands on elec­tri­ci­ty gene­ra­ti­on to run this com­pu­ting infra­struc­tu­re, the under­ly­ing tech­no­lo­gy must be as effi­ci­ent as possible.

Acce­le­ra­ted com­pu­ting powered by NVIDIA GPUs and the NVIDIA AI plat­form offer the effi­ci­en­cy that enables data cen­ters to sus­tain­ab­ly dri­ve the next gene­ra­ti­on of breakthroughs.

And now, timed with the launch of 4th Gen Intel Xeon Sca­lable pro­ces­sors, NVIDIA and its part­ners have kicked off a new gene­ra­ti­on of acce­le­ra­ted com­pu­ting sys­tems that are built for ener­gy-effi­ci­ent AI. When com­bi­ned with NVIDIA H100 Ten­sor Core GPUs, the­se sys­tems can deli­ver dra­ma­ti­cal­ly hig­her per­for­mance, grea­ter sca­le and hig­her effi­ci­en­cy than the pri­or gene­ra­ti­on, pro­vi­ding more com­pu­ta­ti­on and pro­blem-sol­ving per watt.

The new Intel CPUs will be used in NVIDIA DGX H100 sys­tems, as well as in more than 60 ser­vers fea­turing H100 GPUs from NVIDIA part­ners around the world.

Supercharging Speed, Efficiency and Savings for Enterprise AI

The coming NVIDIA and Intel-powered sys­tems will help enter­pri­ses run workloads an avera­ge of 25x more effi­ci­ent­ly than tra­di­tio­nal CPU-only data cen­ter ser­vers. This incre­di­ble per­for­mance per watt means less power is nee­ded to get jobs done, which helps ensu­re the power available to data cen­ters is used as effi­ci­ent­ly as pos­si­ble to super­char­ge the most important work.

Com­pared to pri­or-gene­ra­ti­on acce­le­ra­ted sys­tems, this new gene­ra­ti­on of NVI­DIA-acce­le­ra­ted ser­vers speed trai­ning and infe­rence to boost ener­gy effi­ci­en­cy by 3.5x – which trans­la­tes into real cost savings, with AI data cen­ters deli­ve­ring over 3x lower total cost of ownership.

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

Among the fea­tures of the new 4th Gen Intel Xeon CPU is sup­port for PCIe Gen 5, which can dou­ble the data trans­fer rates from CPU to NVIDIA GPUs and net­wor­king. Increased PCIe lanes allow for a grea­ter den­si­ty of GPUs and high-speed net­wor­king within each server.

Fas­ter memo­ry band­width also impro­ves the per­for­mance of data-inten­si­ve workloads such as AI, while net­wor­king speeds — up to 400 giga­bits per second (Gbps) per con­nec­tion — sup­port fas­ter data trans­fers bet­ween ser­vers and storage.

NVIDIA DGX H100 sys­tems and ser­vers from NVIDIA part­ners with H100 PCIe GPUs come with a licen­se for NVIDIA AI Enter­pri­se, an end-to-end, secu­re, cloud-nati­ve suite of AI deve­lo­p­ment and deploy­ment soft­ware, pro­vi­ding a com­ple­te plat­form for excel­lence in effi­ci­ent enter­pri­se AI.

NVIDIA DGX H100 Systems Supercharge Efficiency for Supersize AI

As the fourth gene­ra­ti­on of the world’s pre­mier pur­po­se-built AI infra­struc­tu­re, NVIDIA DGX H100 sys­tems pro­vi­de a ful­ly opti­mi­zed plat­form powered by the ope­ra­ting sys­tem of the acce­le­ra­ted data cen­ter, NVIDIA Base Com­mand software.

Each DGX H100 sys­tem fea­tures eight NVIDIA H100 GPUs, 10 NVIDIA ConnectX‑7 net­work adap­ters and dual 4th Gen Intel Xeon Sca­lable pro­ces­sors to deli­ver the per­for­mance requi­red to build lar­ge gene­ra­ti­ve AI models, lar­ge lan­guage models, recom­men­der sys­tems and more.

Com­bi­ned with NVIDIA net­wor­king, this archi­tec­tu­re super­char­ges effi­ci­ent com­pu­ting at sca­le by deli­ve­ring up to 9x more per­for­mance than the pre­vious gene­ra­ti­on and 20x to 40x more per­for­mance than unac­ce­le­ra­ted X86 dual-socket ser­vers for AI trai­ning and HPC workloads. If a lan­guage model pre­vious­ly requi­red 40 days to train on a clus­ter of X86-only ser­vers, the NVIDIA DGX H100 using Intel Xeon CPUs and ConnectX‑7 powered net­wor­king could com­ple­te the same work in as litt­le as 1–2 days.

NVIDIA DGX H100 sys­tems are the buil­ding blocks of an enter­pri­se-rea­dy, turn­key NVIDIA DGX Super­POD, which deli­vers up to one exa­flop of AI per­for­mance, pro­vi­ding a leap in effi­ci­en­cy for lar­ge-sca­le enter­pri­se AI deployment.

NVIDIA Partners Boost Data Center Efficiency 

For AI data cen­ter workloads, NVIDIA H100 GPUs enable enter­pri­ses to build and deploy appli­ca­ti­ons more efficiently.

Brin­ging a new gene­ra­ti­on of per­for­mance and ener­gy effi­ci­en­cy to enter­pri­ses world­wi­de, a broad port­fo­lio of sys­tems with H100 GPUs and 4th Gen Intel Xeon Sca­lable CPUs are coming soon from NVIDIA part­ners, inclu­ding ASUS, Atos, Cis­co, Dell Tech­no­lo­gies, Fuji­tsu, GIGABYTE, Hew­lett Packard Enter­pri­se, Leno­vo, QCT and Supermicro.

As the bell­we­ther of the effi­ci­en­cy gains to come, the Fla­ti­ron Institute’s Leno­vo Think­Sys­tem with NVIDIA H100 GPUs tops the latest Green500 list — and NVIDIA tech­no­lo­gies power 23 of the top 30 sys­tems on the list. The Fla­ti­ron sys­tem uses pri­or-gene­ra­ti­on Intel CPUs, so even more effi­ci­en­cy is expec­ted from the sys­tems now coming to market.

Addi­tio­nal­ly, con­nec­ting ser­vers with NVIDIA ConnectX‑7 net­wor­king and Intel 4th Gen Xeon Sca­lable pro­ces­sors will increase effi­ci­en­cy and redu­ce infra­struc­tu­re and power consumption.

NVIDIA ConnectX‑7 adap­ters sup­port PCIe Gen 5 and 400 Gbps per con­nec­tion using Ether­net or Infi­ni­Band, doubling net­wor­king through­put bet­ween ser­vers and to sto­rage. The adap­ters sup­port advan­ced net­wor­king, sto­rage and secu­ri­ty off­loads. ConnectX‑7 redu­ces the num­ber of cables and switch ports nee­ded, saving 17% or more on elec­tri­ci­ty nee­ded for the net­wor­king of lar­ge GPU-acce­le­ra­ted HPC and AI clus­ters and con­tri­bu­ting to the bet­ter ener­gy effi­ci­en­cy of the­se new servers.

NVIDIA AI Enterprise Software Delivers Full-Stack AI Solution

The­se next-gene­ra­ti­on sys­tems also deli­ver a leap for­ward in ope­ra­tio­nal effi­ci­en­cy as they’re opti­mi­zed for the NVIDIA AI Enter­pri­se soft­ware suite.

Run­ning on NVIDIA H100, NVIDIA AI Enter­pri­se acce­le­ra­tes the data sci­ence pipe­line and stream­li­nes the deve­lo­p­ment and deploy­ment of pre­dic­ti­ve AI models to auto­ma­te essen­ti­al pro­ces­ses and gain rapid insights from data.

With an exten­si­ve libra­ry of full-stack soft­ware, inclu­ding AI work­flows of refe­rence appli­ca­ti­ons, frame­works, pre­trai­ned models and infra­struc­tu­re opti­miza­ti­on, the soft­ware pro­vi­des an ide­al foun­da­ti­on for sca­ling enter­pri­se AI success.

To try out NVIDIA H100 run­ning AI work­flows and frame­works sup­port­ed in NVIDIA AI Enter­pri­se, sign up for NVIDIA Launch­Pad free of charge.

Watch NVIDIA foun­der and CEO Jen­sen Huang speak at the 4th Gen Intel Xeon Sca­lable pro­ces­sor launch event.