Dozens of latest supercomputers for scientific computing will quickly hop on-line, powered by NVIDIA’s breakthrough GH200 Grace Hopper Superchip for giant-scale AI and excessive efficiency computing.
The NVIDIA GH200 allows scientists and researchers to deal with the world’s most difficult issues by accelerating complicated AI and HPC purposes working terabytes of knowledge.
On the SC23 supercomputing present, NVIDIA immediately introduced that the superchip is coming to extra techniques worldwide, together with from Dell Applied sciences, Eviden, Hewlett Packard Enterprise (HPE), Lenovo, QCT and Supermicro.
Bringing collectively the Arm-based NVIDIA Grace CPU and Hopper GPU architectures utilizing NVIDIA NVLink-C2C interconnect expertise, GH200 additionally serves because the engine behind scientific supercomputing facilities throughout the globe.
Mixed, these GH200-powered facilities symbolize some 200 exaflops of AI efficiency to drive scientific innovation.
HPE Cray Supercomputers Combine NVIDIA Grace Hopper
On the present in Denver, HPE introduced it’s going to provide HPE Cray EX2500 supercomputers with the NVIDIA Grace Hopper Superchip. The built-in resolution will characteristic quad GH200 processors, scaling as much as tens of 1000’s of Grace Hopper Superchip nodes to offer organizations with unmatched supercomputing agility and faster AI coaching. This configuration can even be a part of a supercomputing resolution for generative AI that HPE launched immediately.
“Organizations are quickly adopting generative AI to speed up enterprise transformations and technological breakthroughs,” mentioned Justin Hotard, govt vp and common supervisor of HPC, AI and Labs at HPE. “Working with NVIDIA, we’re excited to ship a full supercomputing resolution for generative AI, powered by applied sciences like Grace Hopper, which is able to make it straightforward for patrons to speed up large-scale AI mannequin coaching and tuning at new ranges of effectivity.”
Subsequent-Technology AI Supercomputing Facilities
An unlimited array of the world’s supercomputing facilities are powered by NVIDIA Grace Hopper techniques. A number of high facilities introduced at SC23 that they’re now integrating GH200 techniques for his or her supercomputers.
Germany’s Jülich Supercomputing Centre will use GH200 superchips in JUPITER, set to turn into the primary exascale supercomputer in Europe. The supercomputer will assist deal with pressing scientific challenges, similar to mitigating local weather change, combating pandemics and bolstering sustainable power manufacturing.
Japan’s Joint Heart for Superior Excessive Efficiency Computing — established between the Heart for Computational Sciences on the College of Tsukuba and the Info Expertise Heart on the College of Tokyo — promotes superior computational sciences built-in with knowledge analytics, AI and machine studying throughout academia and business. Its next-generation supercomputer shall be powered by NVIDIA Grace Hopper.
The Texas Superior Computing Heart, primarily based in Austin, Texas, designs and operates a few of the world’s strongest computing sources. The middle will energy its Vista supercomputer with NVIDIA GH200 for low energy and high-bandwidth reminiscence to ship extra computation whereas enabling larger fashions to run with better effectivity.
The Nationwide Heart for Supercomputing Functions on the College of Illinois Urbana-Champaign will faucet NVIDIA Grace Hopper superchips to energy DeltaAI, a complicated computing and knowledge useful resource set to triple NCSA’s AI-focused computing capability.
And, the College of Bristol lately acquired funding from the UK authorities to construct Isambard-AI, set to be the nation’s strongest supercomputer, which is able to allow AI-driven breakthroughs in robotics, large knowledge, local weather analysis and drug discovery. The brand new system, being constructed by HPE, shall be geared up with over 5,000 NVIDIA GH200 Grace Hopper Superchips, offering 21 exaflops of AI supercomputing energy able to making 21 quintillion AI calculations per second.
These techniques be a part of beforehand introduced next-generation Grace Hopper techniques from the Swiss Nationwide Supercomputing Centre, Los Alamos Nationwide Laboratory and SoftBank Corp.
GH200 Transport Globally and Out there in Early Entry from CSPs
GH200 is accessible in early entry from choose cloud service suppliers similar to Lambda and Vultr. Oracle Cloud Infrastructure immediately introduced plans to supply GH200 cases, whereas CoreWeave detailed plans for early availability of its GH200 cases beginning in Q1 2024.
Different system producers similar to ASRock Rack, ASUS, GIGABYTE and Ingrasys will start transport servers with the superchips by the top of the 12 months.
NVIDIA Grace Hopper has been adopted in early entry for supercomputing initiatives by greater than 100 enterprises, organizations and authorities companies throughout the globe, together with the NASA Ames Analysis Heart for aeronautics analysis and international power firm TotalEnergies.
As well as, the GH200 will quickly turn into obtainable by NVIDIA LaunchPad, which gives free entry to enterprise NVIDIA {hardware} and software program by an web browser.
Study extra about Grace Hopper and different supercomputing breakthroughs by becoming a member of NVIDIA at SC23.
