The only way to really do this is to geo-fence disable the chips.
The only way to really do this is to geo-fence disable the chips.
For example they could set-up a couple of "mining startups" in Dubai, Egypt, Serbia or India. Then buy 1000s of GPUs under the companies name and then put them in a yacht and ship to Moscow.
Replacing CUDA programs will take work, but they have enough people for that and this is an easier task than designing a GPU.
Russia did that? Right...
China on the other hand, has already designed faster GPUs than those that are banned now for export to them, so they will not be affected. NVIDIA will be more affected by the ban than China, as they have already announced that they will have to delay the deployment of H100 until they will succeed to migrate the assembly of cards to locations outside China.
Other comments have already provided links to benchmarks comparing A100 with the Chinese Biren GPU.
High tech is still high tech for a reason. Sure, in the long term China is going to get there eventually. But in the intermediate decade or two, there's plenty of time for trade sanctions to work to shape markets.
AMD and Google have already beat Nvidia at the ML accelerator hardware game in the past. Google TPUs have had the edge over Nvidia cards for a few years, and AMD spent almost a decade with an edge in compute over Nvidia at every level. It's just that Google TPUs aren't commercially available and AMD GPUs lacked the software stack and critical mass.
Actually building a fast ML accelerator is far from a monumental task. It's difficult, but well within the reach of a country which fully designed a Top10 processor by itself. And now has the means to manufacture super high end chips.
It's not a case of intermediate decade or two. The Chinese are pretty much already there. They already have at least one indigenous exascale supercomputer and their own 7nm process. They can make an ML accelerator comparable to the A100 if they want to. Believe it or not, there are only so many ways of doing matrix multiplication in hardware. The hard part is lithography, and they did it.
EDIT: I forgot to mention, but SMIC wanted to keep their 7nm production secret to reduce incentives for the US to force ASML to stop 28nm lithography machines, keeping the impression that banning the sale of EUV machinery was enough to stop their advances. That's why they mix everything under 28nm in their financial reports, to show their very strong profitability without tipping their hand.
That is already in the past. They have already designed a datacenter GPU faster than NVIDIA A100 (made at TSMC):
https://www.servethehome.com/biren-br100-gpu-for-datacenter-...
Alibaba has already the fastest per socket server CPU for non-floating-point workloads, which has 128 Armv9 cores and DDR5 and PCIe5 before the launch of Intel Sapphire Rapids and AMD Genoa.
At this time, China is at least at parity with USA in chip design, so no export bans of US chips can hurt them.
Their only vulnerability is the dependence on the superior manufacturing capability of TSMC, so only banning exports from TSMC to China can hurt them (like Huawei was hurt when they were no longer able to make chips at TSMC).
This has been discovered by studying with an electron microscope some samples of ASICs made at SMIC for crypto mining.
This story has been reported in a large number of publications and it is possible to buy a detailed report with the result of the investigations about the secret 7 nm process of SMIC.
Because of the similarities with the TSMC process, and because it is known that SMIC has hired former TSMC employees, it is believed that SMIC might have copied various parts of the TSMC process, which could be a reason to not bragg about it.
However such copying is not as simple as non-specialists may believe. I have worked in semiconductor manufacturing, and even if you have all the details about how a technological process is done in another plant, when using other equipment you will not be able to reproduce the results without a lot of tweaking and changes.
Having stolen information about how the competition does it is helpful and it can shorten a lot the development time, but it cannot replace the need for a team of competent engineers, able to develop such a process by themselves, even if in a longer time.
If you look at the SMIC earnings report, you'll notice they added a new category called "28nm or less". That's their way of telling investors they have a money factory without telling the US to sanction them even more! Also, a very large part of SMIC >32% is owned by the Chinese state, and their profits have been skyrocketing in recent years.
Downvoter: Just literally read the news about Jack Ma and BABA and Tencent etc.
Just buy less GPUs and run them for more time. It's not like GPU clusters are running at 100% anyway.
(Also, implying China does not have enough money to buy them like that)
"Just get way better at managing compute resources. You know, that area of research every giant, multi-national cloud provider is pouring tons of money into. Just do that."
[1] https://www.telesurenglish.net/news/CIA-Allegedly-Targets-Tu...