China's AI Industry Barely Slowed By US Chip Export Rules (reuters.com) 24
Export controls imposed by the U.S. on microchips, aiming to hinder China's technological advancements, have had minimal effects on the country's tech sector. While the restrictions have slowed down variants of Nvidia's chips for the Chinese market, it has not halted China's progress in areas like AI, as the reduced performance is still an improvement for Chinese firms, and researchers are finding ways to overcome the limitations. Reuters reports: Nvidia has created variants of its chips for the Chinese market that are slowed down to meet U.S. rules. Industry experts told Reuters the newest one - the Nvidia H800, announced in March - will likely take 10% to 30% longer to carry out some AI tasks and could double some costs compared with Nvidia's fastest U.S. chips. Even the slowed Nvidia chips represent an improvement for Chinese firms. Tencent Holdings, one of China's largest tech companies, in April estimated that systems using Nvidia's H800 will cut the time it takes to train its largest AI system by more than half, from 11 days to four days. "The AI companies that we talk to seem to see the handicap as relatively small and manageable," said Charlie Chai, a Shanghai-based analyst with 86Research.
Part of the U.S. strategy in setting the rules was to avoid such a shock that the Chinese would ditch U.S. chips altogether and redouble their own chip-development efforts. "They had to draw the line somewhere, and wherever they drew it, they were going to run into the challenge of how to not be immediately disruptive, but how to also over time degrade China's capability," said one chip industry executive who requested anonymity to talk about private discussions with regulators. The export restrictions have two parts. The first puts a ceiling on a chip's ability to calculate extremely precise numbers, a measure designed to limit supercomputers that can be used in military research. Chip industry sources said that was an effective action. But calculating extremely precise numbers is less relevant in AI work like large language models where the amount of data the chip can chew through is more important. [...] The second U.S. limit is on chip-to-chip transfer speeds, which does affect AI. The models behind technologies such as ChatGPT are too large to fit onto a single chip. Instead, they must be spread over many chips - often thousands at a time -- which all need to communicate with one another.
Nvidia has not disclosed the China-only H800 chip's performance details, but a specification sheet seen by Reuters shows a chip-to-chip speed of 400 gigabytes per second, less than half the peak speed of 900 gigabytes per second for Nvidia's flagship H100 chip available outside China. Some in the AI industry believe that is still plenty of speed. Naveen Rao, chief executive of a startup called MosaicML that specializes in helping AI models to run better on limited hardware, estimated a 10-30% system slowdown. "There are ways to get around all this algorithmically," he said. "I don't see this being a boundary for a very long time -- like 10 years." Moreover, AI researchers are trying to slim down the massive systems they have built to cut the cost of training products similar to ChatGPT and other processes. Those will require fewer chips, reducing chip-to-chip communications and lessening the impact of the U.S. speed limits.
Part of the U.S. strategy in setting the rules was to avoid such a shock that the Chinese would ditch U.S. chips altogether and redouble their own chip-development efforts. "They had to draw the line somewhere, and wherever they drew it, they were going to run into the challenge of how to not be immediately disruptive, but how to also over time degrade China's capability," said one chip industry executive who requested anonymity to talk about private discussions with regulators. The export restrictions have two parts. The first puts a ceiling on a chip's ability to calculate extremely precise numbers, a measure designed to limit supercomputers that can be used in military research. Chip industry sources said that was an effective action. But calculating extremely precise numbers is less relevant in AI work like large language models where the amount of data the chip can chew through is more important. [...] The second U.S. limit is on chip-to-chip transfer speeds, which does affect AI. The models behind technologies such as ChatGPT are too large to fit onto a single chip. Instead, they must be spread over many chips - often thousands at a time -- which all need to communicate with one another.
Nvidia has not disclosed the China-only H800 chip's performance details, but a specification sheet seen by Reuters shows a chip-to-chip speed of 400 gigabytes per second, less than half the peak speed of 900 gigabytes per second for Nvidia's flagship H100 chip available outside China. Some in the AI industry believe that is still plenty of speed. Naveen Rao, chief executive of a startup called MosaicML that specializes in helping AI models to run better on limited hardware, estimated a 10-30% system slowdown. "There are ways to get around all this algorithmically," he said. "I don't see this being a boundary for a very long time -- like 10 years." Moreover, AI researchers are trying to slim down the massive systems they have built to cut the cost of training products similar to ChatGPT and other processes. Those will require fewer chips, reducing chip-to-chip communications and lessening the impact of the U.S. speed limits.
Protectionism does not work (Score:4, Interesting)
It makes your own people fat and lazy and stupid and gives the other guy a lot of incentives to catch up and remove the dependency on your stuff in the first place. This cannot be avoided. Hence all it does is accelerate evolution of the competition you are trying to restrict. Of course, a country where Evolution is "debated" or seen as "just a theory" may not have the collective intellectual capacity to understand how this works.
Re: (Score:2)
Re: (Score:1)
Re: (Score:2)
Re: (Score:2)
...locked out US companies.
The US company I work for makes nearly $1 billion in profits from it's Chinese holdings every year.
Re: (Score:2)
It's called making a *deal* with the devil. Of course you get a benefit, but at what cost? Ask the Japanese and German rail industry on how they feel.
Re: (Score:2)
They seem to quite like the billions in profits they've had over the last 20 years or so. You might be mad at the wrong people.
Re: (Score:2)
China is/was a developing economy. The US did the same thing when it was developing. The goal of protecting developing industries should be to produce businesses that are competitive.
Re:Protectionism does not work (Score:5, Insightful)
Protectionism for economic reasons does not work and is harmful to both sides.
The trade restrictions on China are not primarily about economics, but about national security. They are economically damaging to everyone but hurt China more if enough countries join in.
Realistically, the sanctions are unlikely to improve China's behavior and may make it worse. There's a lot of misunderstanding between our culture and theirs.
Re: (Score:2)
All this was ever going to do was raise costs for Chinese AI development slightly, which is motivation to develop domestic alternatives. Nvidia is wasting money developing extra SKUs for export, and handling all the paperwork and legal requirements.
It's like in a movie when the hero is trying to diffuse a bomb, pulls the wrong wire and sees the countdown is now running twice as fast.
We need to invest in R&D. Unfortunately right now companies are instead laying off hundreds, even thousands of people, bec
Re: (Score:2)
Indeed. The only way to keep an adversary under control that is slowly catching up in capabilities is to develop your own capabilities faster. Of course that means less profits in the short run and hence is hard to do in a form of capitalism where everybody is obsessed with short-term profits.
The second effect all these layoffs have is that a lot of smart people will not go into R&D and instead look for a career with less risk. And that is a loss for more than a few years.
Re: (Score:2)
Trouble dealing with the truth? Your examples are bogus. While protectionism may have been involved, none of your examples rose on protectionism.
Re: (Score:2)
Re: (Score:2)
Wait, you're saying that competition is a good thing?
Someone should figure out some kind of economic system based on that idea.
I'm old enough to remember (Score:3)
Back in the 70's the US put export restrictions on mainframes and mini-computers, PCs were just hobbyist toys.
This meant DEC's PDP-11 couldn't be exported to say India without approval from quite a few US agencies. This didn't stop the Soviets and Indian entrepreneurs from reverse engineering them and building clones.
We're now in an AI arms race.
Good! (Score:1)
I love it when this kind of short-sighted, jealous and evil political goal backfires, just like with (overused and abused) sanctions against a vast number of countries.
Well, the bans were just what China needed (Score:4, Insightful)
To kick their own chip development efforts into high gear. Which they will now do.