Oct. 9, 2023 — Russia has plans to assemble as much as ten supercomputers by 2030, aiming to equip every with 10,000 to fifteen,000 Nvidia H100 GPUs, in response to reporting from Tom’s {Hardware}. A system full of so many H100 GPUs might obtain round 450 FP64 TFLOPS, amounting to half of an ExaFLOP, a benchmark solely at the moment met by the U.S.
The ‘Trusted Infrastructure’ staff spearheads this Russian enterprise, striving to redefine the bounds of computational capabilities. Nonetheless, a looming hurdle is the acquisition of the coveted Nvidia H100 GPUs, primarily as a result of Nvidia is an American entity.
Following the conflict Russia initiated in Ukraine, the U.S. imposed stringent tech export restrictions on Russia, rendering the procurement of such GPUs a frightening job. Given the sheer quantity of GPUs, smuggling appears implausible, leaving everybody questioning how Russia intends to navigate these restrictions.
One other vital problem is the monetary side. As of now, the supercomputer undertaking would command a funds near $6 billion. However the ever-evolving tech panorama could result in price reductions, probably bringing the worth tag right down to $500 – $700 million by 2030.
At current, Russia’s computational powerhouse is the Chervonenkis supercomputer, owned by Yandex and fitted with Nvidia A100 GPUs. Rating twenty seventh globally in computational energy, it boasts a efficiency of 21.53 PetaFLOPS. Out of the seven Russian supercomputers listed within the world prime 500, three belong to Yandex, two to Sberbank, and one every to MSU and MTS, all of which depend on Nvidia GPUs from earlier generations.
For a worldwide context, the USA dominates with 150 supercomputers on the checklist, adopted by China’s 134, Germany’s 36, and Japan’s 33. As of June 2023, Russia stood twelfth in world rankings. The nation’s dedication to bolstering its place in high-performance computing is obvious, however the path to reaching this stays riddled with challenges.
Observe this hyperlink for extra from Tom’s {Hardware}.
