Sanctions Change Means Hobbled ‘China-Only’ Nvidia A800 GPUs Now Diverted to Western Markets

There are signs that Nvidia and partners are gearing up to market A800 GPUs to North America, Latin America, Europe, Africa, India, and Japan due to rapidly intensifying US sanctions against China. Nvidia’s A800 was an Ampere architecture GPU designed to lithely limbo beneath a set of US sanction-triggering GPU specifications through the use of a cut-down feature set to comply with US sanctions. Sadly for Nvidia, the sanctions have been intensified in recent weeks, so this CPU can no longer be sold in the Chinese market. 

Tech business news site CRN reports that green team partners like PNY, Colfax International, ASK, and Elsa are all now actively promoting A800 GPUs for AI, data science, and HPC work. We first reported upon the A800 as an A100 alternative for China when news of its existence broke a year ago.

(Image credit: Nvidia)

Nvidia made the A800 to be used instead of the A100, capable of running the same tasks, albeit slower. The export restriction specs might have changed, but the U.S. government says that the goal remains unchanged. These restrictions are in place “to limit PRC access to advanced semiconductors that could fuel breakthroughs in artificial intelligence,” according to Gina Raimondo, the U.S. Commerce Secretary. It is a shame that the government is disrupting commerce, which has been adhering closely to rules clearly set out previously, incurring unexpected inconveniences and expenses.

The chopping and changing export rules enforced by the government might have been a bigger problem for Nvidia and partners if it wasn’t for the strong demand for GPU processing power from businesses investing in A.I. and machine learning.

Nvidia U.S. is touting the newly released U.S. stocks of A800 GPUs as “the ultimate workstation development platform for A.I., data science, and high-performance computing.” The page also has a link to find a vendor, full specifications, a selection of performance metrics for reference, a downloadable data sheet, and so on.

Of course, US customers, for example, also have access to solutions based upon or using Nvidia’s fully-fledged A100 GPU. This can offer up to double the HBM2e memory with choices of 40 or 80 GB SKUs, as well as better raw compute performance (see specs comparison tables above).

Sales of Nvidia A800 40 GB GPU in permissible markets like the US will begin in the next few weeks, according to the source. We don’t have any figures for pricing.

This post was originally published on this site