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

AI is on the coronary heart of humanity’s most transformative improvements — from creating COVID vaccines at unprecedented speeds and diagnosing most cancers to powering autonomous autos and understanding local weather change.

Nearly each business will profit from adopting AI, however the expertise has change into extra useful resource intensive as neural networks have elevated in complexity. To keep away from inserting unsustainable calls for on electrical energy era to run this computing infrastructure, the underlying expertise should be as environment friendly as attainable.

Accelerated computing powered by NVIDIA GPUs and the NVIDIA AI platform provide the effectivity that allows information facilities to sustainably drive the subsequent era of breakthroughs.

And now, timed with the launch of 4th Gen Intel Xeon Scalable processors, NVIDIA and its companions have kicked off a brand new era of accelerated computing techniques which are constructed for energy-efficient AI. When mixed with NVIDIA H100 Tensor Core GPUs, these techniques can ship dramatically greater efficiency, better scale and better effectivity than the prior era, offering extra computation and problem-solving per watt.

The brand new Intel CPUs shall be utilized in NVIDIA DGX H100 techniques, in addition to in additional than 60 servers that includes H100 GPUs from NVIDIA companions around the globe.

Supercharging Pace, Effectivity and Financial savings for Enterprise AI

The approaching NVIDIA and Intel-powered techniques will assist enterprises run workloads a median of 25x extra effectively than conventional CPU-only information heart servers. This unbelievable efficiency per watt means much less energy is required to get jobs performed, which helps guarantee the facility out there to information facilities is used as effectively as attainable to supercharge crucial work.

In comparison with prior-generation accelerated techniques, this new era of NVIDIA-accelerated servers pace coaching and inference to spice up vitality effectivity by 3.5x – which interprets into actual price financial savings, with AI information facilities delivering over 3x decrease whole price of possession.

New 4th Gen Intel Xeon CPUs Transfer Extra Information to Speed up NVIDIA AI

Among the many options of the brand new 4th Gen Intel Xeon CPU is assist for PCIe Gen 5, which might double the information switch charges from CPU to NVIDIA GPUs and networking. Elevated PCIe lanes enable for a better density of GPUs and high-speed networking inside every server.

Sooner reminiscence bandwidth additionally improves the efficiency of data-intensive workloads resembling AI, whereas networking speeds — as much as 400 gigabits per second (Gbps) per connection — assist quicker information transfers between servers and storage.

NVIDIA DGX H100 techniques and servers from NVIDIA companions with H100 PCIe GPUs include a license for NVIDIA AI Enterprise, an end-to-end, safe, cloud-native suite of AI improvement and deployment software program, offering a whole platform for excellence in environment friendly enterprise AI.

NVIDIA DGX H100 Methods Supercharge Effectivity for Supersize AI

Because the fourth era of the world’s premier purpose-built AI infrastructure, NVIDIA DGX H100 techniques present a completely optimized platform powered by the working system of the accelerated information heart, NVIDIA Base Command software program.

Every DGX H100 system options eight NVIDIA H100 GPUs, 10 NVIDIA ConnectX-7 community adapters and twin 4th Gen Intel Xeon Scalable processors to ship the efficiency required to construct massive generative AI fashions, massive language fashions, recommender techniques and extra.

Mixed with NVIDIA networking, this structure supercharges environment friendly computing at scale by delivering as much as 9x extra efficiency than the earlier era and 20x to 40x extra efficiency than unaccelerated X86 dual-socket servers for AI coaching and HPC workloads. If a language mannequin beforehand required 40 days to coach on a cluster of X86-only servers, the NVIDIA DGX H100 utilizing Intel Xeon CPUs and ConnectX-7 powered networking may full the identical work in as little as 1-2 days.

NVIDIA DGX H100 techniques are the constructing blocks of an enterprise-ready, turnkey NVIDIA DGX SuperPOD, which delivers as much as one exaflop of AI efficiency, offering a leap in effectivity for large-scale enterprise AI deployment.

NVIDIA Companions Enhance Information Middle Effectivity 

For AI information heart workloads, NVIDIA H100 GPUs allow enterprises to construct and deploy functions extra effectively.

Bringing a brand new era of efficiency and vitality effectivity to enterprises worldwide, a broad portfolio of techniques with H100 GPUs and 4th Gen Intel Xeon Scalable CPUs are coming quickly from NVIDIA companions, together with ASUS, Atos, Cisco, Dell Applied sciences, Fujitsu, GIGABYTE, Hewlett Packard Enterprise, Lenovo, QCT and Supermicro.

Because the bellwether of the effectivity positive aspects to return, the Flatiron Institute’s Lenovo ThinkSystem with NVIDIA H100 GPUs tops the most recent Green500 checklist — and NVIDIA applied sciences energy 23 of the highest 30 techniques on the checklist. The Flatiron system makes use of prior-generation Intel CPUs, so much more effectivity is anticipated from the techniques now coming to market.

Moreover, connecting servers with NVIDIA ConnectX-7 networking and Intel 4th Gen Xeon Scalable processors will enhance effectivity and cut back infrastructure and energy consumption.

NVIDIA ConnectX-7 adapters assist PCIe Gen 5 and 400 Gbps per connection utilizing Ethernet or InfiniBand, doubling networking throughput between servers and to storage. The adapters assist superior networking, storage and safety offloads. ConnectX-7 reduces the variety of cables and swap ports wanted, saving 17% or extra on electrical energy wanted for the networking of enormous GPU-accelerated HPC and AI clusters and contributing to the higher vitality effectivity of those new servers.

NVIDIA AI Enterprise Software program Delivers Full-Stack AI Resolution

These next-generation techniques additionally ship a leap ahead in operational effectivity as they’re optimized for the NVIDIA AI Enterprise software program suite.

Working on NVIDIA H100, NVIDIA AI Enterprise accelerates the information science pipeline and streamlines the event and deployment of predictive AI fashions to automate important processes and achieve speedy insights from information.

With an intensive library of full-stack software program, together with AI workflows of reference functions, frameworks, pretrained fashions and infrastructure optimization, the software program supplies a perfect basis for scaling enterprise AI success.

To check out NVIDIA H100 operating AI workflows and frameworks supported in NVIDIA AI Enterprise, join NVIDIA LaunchPad freed from cost.

Watch NVIDIA founder and CEO Jensen Huang communicate on the 4th Gen Intel Xeon Scalable processor launch occasion.