A very good overview of the generations of Chinese chips, but I feel like matching performance isn’t really the goal for domestic Chinese chips? Sure, they’d love them to be as fast or faster than western cpus, but ultimately they ensure China is not starved of microprocessors in a war.
> ensure China is not starved of microprocessors in a war
Partially true, but the other issue is high performance CPUs and GPUs are needed for simulations applications such as nuclear weapons testing, jet engine design, airframe design, etc.
Much of this requires a highly performance HPC infra, and while China has built a strong distributed systems base (kinda hard not to when UIUC and UCB CS has such a strong alumni network in China), the compute resources needed are still dependent on foreign inputs from Intel and Nvidia, whose top-of-the-line offerings all face import restrictions from the US.
A lot of the impetus around domestic high performance CPU research came when the Intel Xeon ban entered the fold, which was specifically because of the worries from above.
> such as nuclear weapons testing, jet engine design, airframe design
People were doing that 30 years ago with the chips available at the time. F-22 first flew in 1997. If the Chinese can match the F-22, I think they'll take that as a win.
> People were doing that 30 years ago with the chips available at the time
I agree to a certain extent, but this isn't something done in isolation, and I think you're falling into the classic trap of assuming a hard problem is actually easy.
Just having chips alone isn't enough for simulations - you need an entire ecosystem of SDKs, BLAS implementations, etc that are very difficult and very hardware dependent.
Most of the existing stack and ecosystem in the space was developed with Intel, AMD, and increasingly Nvidia hardware in mind and with their input, and to even reach that point you need to build multiple domestic champions if you're the kind of country that's increasingly facing tech sanctions which imo are increasingly comparable to those imposed on Russia post-Crimea.
And this is why there has been a massive push on this kind of domestic ecosystem in China after the Xeon import sanctions began.
Also, as I mentioned previously, the biggest reason is nuclear simulations, as China has de facto chosen to transition from defensive nuclear capabilities to offensive as well [0] (sources in this comment thread).
There is this mythical thing about the nuclear simulations. The national labs run the baddest supercomputers out there, and they are involved in the nuclear stockpile stewardship, so huge simulations must somehow be necessary. Either to maintain the nukes themselves, or at least to maintain them while respecting the comprehensive nuclear test ban treaty. Treaty that was signed but never ratified by the US, Russia or China, by the way.
The reality is something like that. A fission bomb, when detonated, goes through about 50 or 60 generations of neutrons. Because of the exponential nature of the chain reaction, most of the energy comes from the last 10 or 20 generations. You can adjust the power of the implosion so that you trigger only 10 or 20 or 30 generations of neutrons. At some point you get enough heat from the reaction that the core melts; such a test is called "hydronuclear" and CIA has strong evidence that Russia performs hydronuclear tests.
You can also adjust the implosion that you get a chain reaction that is subcritical (each new generation has fewer neutrons that the one before). All nuclear powers perform subcritical tests.
The difference between subcritical and hydronuclear tests is only a theoretical one. There are no means to monitor a state for hydronuclear tests (other than human intelligence of course).
China does not need supercomputers to maintain a cutting edge nuclear force. After all, the US and the USSR had such nuclear forces 50 years ago, when the entire computing power of the world was less than what my iPhone can provide to me right now.
The Chinese have more honor students than the us has students. I’m sure they can figure it out… it’s not easy sure but given the enormous amount of resources at their disposal it’s certainly possible
I mean in a way it worked pretty well for Russia. Otherwise we wouldn't have surreal pictures of Space Shuttles in dumps or clones of the Concorde even though inferior still able to fly. It works and it's quick to do and catch up as a national pride item.
Seems to work a lot better when we, the fools that we are, fund it.
For the record, I am confident that when Xi appointed himself emperor, it was a clear sign of their system having failed.
edit: As a counterpoint, temporary central planning of the US economy is exactly how the USA won WWII. I often think about this when I consider what can be done about anthropogenic climate change.
Temporary central planning in pursuit of a single common goal works at the cost of individual freedoms.
Centrally planning an economy for (hundreds of) millions of free people just going around their lives doesn't. Central planning requires rationality, but people do things just because they feel so.
> Temporary central planning in pursuit of a single common goal works at the cost of individual freedoms.
Yes, in the case of WWII, central planning led to many US car manufactures losing the freedom to manufacture cars. In exchange, we beat the Nazis, freed Europe, and stopped the holocaust.
I'd say that was a pretty good deal.
note to anyone reading this: Please attempt not to use simplified, polarized thinking for a moment. Communism can be bad/evil at the same time that temporary central planning to achieve a goal can be good.
Seems like the problem is that the leadership in the US is half fascist, half bumbling morons (in both parties), not so much whether or not central planning works or doesn't.
Eh, Germany, Japan, and the US all extensively used central planning after WW2. The results speak for themselves. The Chinese government, for it's many faults, has a good track record when it comes to modernization.
