Posted on 11/08/2022 1:52:43 PM PST by FarCenter
Systems that once contained Nvidia and TSMC chips, which are now restricted by the US government, are popping up this week with slower specs to meet US export controls to China and evade the hassles of obtaining special licenses.
Chinese server maker Inspur was spotted swapping out the its A100 Graphic Processing Unit (GPU) in its NF5688MG [PDF] for an A800 in updated promotional material.
H3C, the China-based Tsinghua Group and Hewlett Packard mashup, are also suddenly featuring the A800 on select servers.
According to specs provided online by Chinese server-maker Omnisky, which also uses the chip, it remains eerily similar to the A100, [PDF] but with a 40GB PCIe option available in the A800, in addition to a 80GB PCIe and 80GB SXM available in both models.
The A800 has a degraded transfer rate of 400GBps across all three variations, as opposed to the 600GBps in the A100, thus falling in line with US law that restrict 600GBps and above being shipped to the Middle Kingdom. The Register asked Nvidia about the A800 and will update if substantial information arises.
...
China's domestic chip facilities still have a long way to go to be globally competitive. However, Biren did have a frontrunner in the making, but amid uncertainty about its potential legality, TSMC had reportedly put the brakes on the silicon.
Biren’s BR100 was pitched as an Nvidia H100 rival. The drag of the chip’s transfer rate from 640GBps to 576GBps, at least in its marketing material if not in real life, would allow TSMC to produce the GPU without facing US scrutiny.
(Excerpt) Read more at theregister.com ...
China is America’s biggest long-term rival.
And to oppose them, we embroiled ourselves, Europe and Ukraine in a mutually-destructive war with Russia.
Makes perfect sense....
This isn't news, this is how it's supposed to work.
According to an analysis from the Center for Strategic and International Studies, a bipartisan think tank, “By only targeting chips with very high interconnect speeds, the White House is attempting to limit the controls to chips that are designed to be networked together in the data centers or supercomputing facilities that train and run large AI models.”
https://techcrunch.com/2022/11/07/nvidia-us-china-ban-alternative/
Maybe a 33% reduction in interconnect speed "cripples" something, even though the chip does math internally just as fast. But that depends a lot on overall system design and programming techniques.
It’s a bit late now. This should have been done 20+ years ago.
Also, the stupid part is that this only affects products sold by US companies and/or made in the US, in practical terms. Since we decided we wanted to give our chip manufacturing lead away, well, those foreign companies aren’t really affected and if they want to evade it, all they have to do is sell to another nation which then turns around and sells it covertly to China.
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.