Chinese government shifts focus from x86 and Arm CPUs, gov't promoting RISC-V chips heavily
But what about software?

Being one of the world's biggest markets for PCs and microprocessors, China consumes all kinds of CPUs designed by domestic and foreign companies. Domestic companies like Huawei and Zhaoxin have developed Arm and x86-based CPUs, which enables them to take advantage of industry-leading ecosystems. However, as China moves towards semiconductor self-sustainability, the Chinese government wants to encourage using open-source RISC-V CPUs, reports Reuters.
China is set to introduce new policies promoting the nationwide use of open-source RISC-V chips, aiming to reduce its reliance on foreign technologies, such as x86 and Arm. The initiative was drafted by eight government agencies, including the Cyberspace Administration of China, the Ministry of Industry and Information Technology, the Ministry of Science and Technology, and the China National Intellectual Property Administration. Once enacted, it will mark the first official government push for RISC-V adoption in China. The upcoming policy could be announced as early as this month, though the exact release date remains uncertain.
Several major Chinese companies are already investing in RISC-V. Alibaba's XuanTie and Nuclei System Technology are two of the country's leading providers of commercial RISC-V processors. Eventually, Chinese chip developers could build RISC-V-based processors for AI applications (something that Tenstorrent does in the U.S.). While initially, those processors will hardly be competitive with those from Nvidia in terms of performance, they will be significantly cheaper and designed in China, precisely something the PRC's government encourages. This affordability could encourage wider adoption, particularly among smaller AI firms seeking efficient, lower-cost solutions.
However, while developing RISC-V hardware is important, creating a software ecosystem for RISC-V CPUs and processors is no less critical. Arm and x86 are supported by a wide range of operating systems and programs, unlike RISC-V, which can boast limited compatibility with software. Also, if someone attempts to build AI processors based on RISC-V, they will have to create an ecosystem like Nvidia's CUDA, which will be particularly hard as it took over a decade to make.
RISC-V is an open-source instruction set architecture (ISA) that competes with proprietary technologies such as x86, controlled by Intel and AMD, and Arm, developed by Arm Holdings. Nowadays, RISC-V is often used for microcontrollers and lower-performance CPUs, but since usage of the actual ISA does not require paying licensing fees and its development is not controlled by one or two companies from the West, it is gaining traction worldwide in general and in China in particular. Chinese research institutions and government-backed entities have increasingly embraced the technology, but until now, no official policy has supported its use.
The growing Chinese interest in RISC-V has sparked concerns in the U.S. In 2023, some American lawmakers urged the Biden administration to limit domestic companies from working on RISC-V projects and extending the ISA, fearing China could use its open-source nature to strengthen the capabilities and performance of its processors. With technology at the center of U.S.-China tensions, RISC-V's expansion in China could become another point of conflict.
Following reports of the policy, multiple Chinese semiconductor companies saw their stock prices rise sharply. Companies such as VeriSilicon, ASR Microelectronics, Shanghai Anlogic Infotech, and 3Peak recorded gains ranging from 8.6% to 15.4%, with VeriSilicon hitting the daily 10% increase limit, according to Reuters.
Stay On the Cutting Edge: Get the Tom's Hardware Newsletter
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
Anton Shilov is a contributing writer at Tom’s Hardware. Over the past couple of decades, he has covered everything from CPUs and GPUs to supercomputers and from modern process technologies and latest fab tools to high-tech industry trends.
-
bit_user
I'm pretty sure that, as with the hardware situation, they also don't want to depend on software they don't control. So, that mostly nullifies the concerns about compatibility with legacy software. For everything else, there are emulators that should work well enough.The article said:Arm and x86 are supported by a wide range of operating systems and programs, unlike RISC-V, which can boast limited compatibility with software.
No, you don't need the entirety of CUDA to accelerate AI. Frameworks like PyTorch and Tensorflow already support several different hardware backends and rely on much less functionality than the totality of CUDA.The article said:if someone attempts to build AI processors based on RISC-V, they will have to create an ecosystem like Nvidia's CUDA, which will be particularly hard as it took over a decade to make.
Overall, I think this was very predictable and actually a lot better for everyone than if they'd thrown their weight behind Loongarch. This way, at least there's still room for collaboration and code compatibility between Chinese systems and western ones. -
ekio Well, that's the smartest way to go.Reply
RISC-V is the obvious choice for technological freedom, advanced software support, and even for potential chip efficiency. -
blitz120 bit_user said:No, you don't need the entirety of CUDA to accelerate AI. Frameworks like PyTorch and Tensorflow already support several different hardware backends and rely on much less functionality than the totality of CUDA.
I'd also add that there's separate work underway to implement CUDA on RISC-V:
https://github.com/vortexgpgpu/vortex_tutorials -
bit_user
FWIW, Think-Silicon (now owned by Applied Materials) announced embedded GPUs based on RISC-V ISA almost 3 years ago:blitz120 said:I'd also add that there's separate work underway to implement CUDA on RISC-V:
https://github.com/vortexgpgpu/vortex_tutorials
https://www.cnx-software.com/2022/06/20/think-silicon-neox-risc-v-gpu-offer-3d-graphics-or-ai-acceleration/ -
nookoool Eventual wide usage on the server side won't be in doubt with linux and porting over of development tools. The trillion dollar question is will anything be able to dethrone win/x86 on desktop in china or elsewhereReply -
bit_user
Development tools have been there for a while. I guess things like profiling and performance counters are still in early daysnookoool said:Eventual wide usage on the server side won't be in doubt with linux and porting over of development tools.
Maybe Arm could get there first:nookoool said:The trillion dollar question is will anything be able to dethrone win/x86 on desktop in china or elsewhere
https://www.tomshardware.com/pc-components/motherboards/worlds-first-open-source-armv9-motherboard-surfaces-radxa-orion-o6s-pricing-starts-at-usd200-for-the-8gb-ram-model
Also, Nvidia is partnering with Mediatek on Arm PCs. We should learn more, later this year. They announced one product that sounds fairly impressive (although rather on the expensive side, IMO):
https://www.tomshardware.com/pc-components/gpus/nvidias-project-digits-desktop-ai-supercomputer-fits-in-the-palm-of-your-hand-usd3-000-to-bring-1-pflops-of-performance-home -
TerryLaze
With that logic, intel released x86 smartphones in 2012....tremble in fear arm, your days are counted....bit_user said:Maybe Arm could get there first:
https://www.tomshardware.com/pc-components/motherboards/worlds-first-open-source-armv9-motherboard-surfaces-radxa-orion-o6s-pricing-starts-at-usd200-for-the-8gb-ram-model
https://en.wikipedia.org/wiki/Motorola_RAZR_i
You people have to get that there is no throne, risc and x86 have existed together since the 80ies with companies using whatever is best suited for any job. It has always been and it will always be that there will be things better suited for risc and others better for x86. -
bit_user
But there haven't been recent examples of an ARM-based motherboard in a PC form factor, like the Radxa one I linked. Yes, you can find a couple (Firefly and LattePanda) that have SoMs with a host board, but those are more for industrial/embedded applications, whereas the Radxa seems aimed more at the AI PC market (indeed, they even have an AI PC kit option).TerryLaze said:You people have to get that there is no throne, risc and x86 have existed together since the 80ies
In my other example is a powerful mini AI PC made by a pair of companies who have also announced plans to deliver a Windows-on-ARM solution. If it's anywhere near as powerful as their DIGITS box, it should be interesting.
We haven't seen examples like these, before. That's what's changed. -
TerryLaze
What's your point with all of this?!bit_user said:But there haven't been recent examples of an ARM-based motherboard in a PC form factor, like the Radxa one I linked. Yes, you can find a couple (Firefly and LattePanda) that have SoMs with a host board, but those are more for industrial/embedded applications, whereas the Radxa seems aimed more at the AI PC market (indeed, they even have an AI PC kit option).
In my other example is a powerful mini AI PC made by a pair of companies who have also announced plans to deliver a Windows-on-ARM solution. If it's anywhere near as powerful as their DIGITS box, it should be interesting.
We haven't seen examples like these, before. That's what's changed.
That people with an x86 based PC need a second one just for AI?
How is that going to dethrone or do anything to x86?
I don't get your logic here.