US Revises Rules On Chip Export, AI Tools To China

The US Department of Commerce recently revised the requirements and regulations concerning the export of high performance chips, tools, and AI accelerators to China. The rules are expected to be on 4 April (5 April here) and will see the Biden administration further tighten its noose around the country’s appetite for AI.

The updated revision, 166 pages long, outlines that if a company’s product surpasses 70 TeraFLOPS of compute performance, they will need to apply for a license from the US government.

For all destinations, except those countries in Country Group E:1 or E:2 of Supplement No. 1 to part 740 of the EAR, no license is required (NLR) for computers with an “Adjusted Peak Performance” (“APP”) not exceeding 70 Weighted TeraFLOPS (WT) and for “electronic assemblies” described in 4A003.c that are not capable of exceeding an “Adjusted Peak Performance” (“APP”) exceeding 70 Weighted TeraFLOPS (WT) in aggregation, except certain transfers as set forth in § 746.3 (Iraq).

By virtue of the revised regulations, that would mean NVIDIA may not be able to sell its GeForce RTX 4090D. It’s ironic because the GPU itself was redesigned from the original RTX 4090 to circumvent the original chip export ban. The card, by the way, far exceeds the 70 TeraFLOPS ceiling, running at 74 TeraFLOPS.

As always, it comes as no surprise that Beijing has expressed its discontent at its economic rival over its revised regulations, saying that it has not only set up more obstacles for companies to carry out normal economic trade cooperation, but that it is also harming the global semiconductor industry.

(Source: Reuters, Icedeal, Department of Commerce, SCMP, Videocardz)