Nvidia redirected accelerators developed for China to Western markets

The classic Nvidia A100 and Nvidia H800 models could not be legally exported to China after the US sanctions, so Nvidia developed models according to the specifications defined by these sanctions and called them A800 and H800. However, the new sanctions of October 16 came with significantly stricter export criteria.

Practically immediately after the publication of the parameters, an addendum arrived in the form of the immediate application of these sanctions, so that Nvidia was left with ready-made hardware in stock and probably also pieces in progress in production. To solve this situation, the company chose to release them on Western markets.

Nvidia A100 with Ampere GA100 core (Nvidia)

Companies that distribute these products, such as PNY, have quickly started offering them in North and Latin America, Europe, Africa, and even India. In Japan, they appeared in the offerings of ASK Corporation and ELSA.

Only the marketing has changed. While to China Nvidia presented these models as “capable GPUs for data centers”, they are now “the ultimate AI development platforms for workstations”. The A800 does not differ from the A100 in processing power or memory, but in the data throughput of the NVLink interface reduced from 600 GB/s to 400 GB/s. Nvidia offers it in a version with an active cooler reminiscent of classic graphics cards, but differs from them in the absence of a graphics output. So it really remains an additional accelerator, for which you need to have a separate graphics card providing output to the monitor.