— Shanghai Macro Strategist (@ShanghaiMacro) August 31, 2022 US officials have asked Nvidia to stop exporting their A100 and H100 data center GPU cards. For AMD, the ban is limited to their MI250 GPUs. Nvidia states in a regulatory filing, “The USG indicated that the new license requirement will address the risk that the covered products may be used in, or diverted to, a ‘military end use’ or ‘military end user’ in China and Russia,” The GPUs in question are one of the best cards these companies offer for AI and HPC workloads. According to Nvidia the “ H100 enables next-generation AI and HPC breakthroughs“. The H100 GPU uses fourth generation tensor cores which are up to 6x faster chip-to-chip compared to the A100. Even AMD’s MI250 is a super performer when it comes to AI and HPC workloads, delivering about 1.4 – 2.4 times the performance of the A100. China has been seizing the HPC ground from the US, and even plans to build 10 exascale systems by 2025. While these supercomputers can be used in many useful ways, such as weather detection and studying complex models. They can also be used in many militaristic ways, such as breaking encryption, and running complex war simulations. – US officials to Reuters But China is obviously not the one to give up, and they have heavily ramped up R&D spending in their own domestic chip capacity. Even their current exascale systems were based on a domestic chip design. The recent GPU export ban from the US is definitely going to hurt, but only temporarily as China continues to bridge the gap.
