The US government has taken aggressive steps in recent weeks to stifle China’s efforts to forge ahead in AI while bolstering the nation’s computing infrastructure for supercomputing and artificial intelligence.
The US government has banned the export of advanced chips, including specific GPUs from companies like AMD and Nvidia, to China.
The ban cuts off China’s access to artificial intelligence chips and software originating in the United States. The US government hopes the ban will slow China’s rapid progress in AI, which is a national priority.
Nix on Nvidia
The ban on GPU exports will also limit the ability of Chinese entities to program advanced AI software through Nvidia’s CUDA parallel programming framework. CUDA allows programmers to write software that takes advantage of advances in GPUs for faster processing of AI programs.
“The combined strength of CUDA software and Nvidia hardware is a big part of why Nvidia accounts for 95% of AI chip sales in China,” the Center for Strategic and International Studies said in a recently published research note. , citing figures from Fubon Securities Investment Services.
The US government is also cutting off access to Nvidia’s latest GPUs that include the Hopper architecture, which was announced this year.
The ban, which is constantly being amended, also imposes restrictions on the export of software enabling the operation of hardware. This means that there could be a ban on CUDA itself, although that is yet to be determined. Nvidia’s Enterprise AI software suite, which helps organizations write AI applications for vertical industries, is based on CUDA.
Nvidia is expected to release the next version of CUDA, version 12, in the coming months, designed to harness Hopper’s computing power. Chinese companies will be able to use CUDA 12 on older GPUs like the Ampere-based A100, but AI applications will run slower.
Nvidia won’t ship its A100 and H100 chips to China, which could slow CUDA adoption. AMD’s open-source parallel programming framework, called ROCm, could be a beneficiary because it doesn’t lock down AI software to its GPUs.
Chinese entities could also go through parallel frameworks like OpenCL, which is open source and managed by the Khronos group. AI tools like TensorFlow and Pytorch are open source and can be used freely. CUDA-specific code can also be easily removed from programs using tools such as SYCLomatic, which migrates and automates code for execution on a range of CPUs, GPUs and FPGAs.
In the long term, the Chinese aim to reduce their reliance on US-based chips and software tools. Chinese universities and companies are moving away from x86 and developing chips based on the RISC-V architecture, which is an open source instruction set architecture that is licensed for free.
China has tried, but struggled to develop local cutting-edge chips as advanced as those made by Intel and Nvidia. At the Hot Chips Summit in August, a China-based startup called Biren Technology talked about its latest GPU made using the 7nm process, saying it was getting closer to developing accelerators for AI applications.
Earlier this month, researchers from the Chinese Academy of Sciences – which is on the list of US entities – shared a methodology on how Chinese researchers used an agile development method to design their XiangShan RISC chip. -V. The iterative approach was similar to how cloud-native companies use DevOps to write and deploy code. But the paper showed the heavy reliance on FPGA technology from companies like Intel and AMD, which can prototype RISC-V chips before production.
Nvidia once considered China an important market, holding its annual GTC conference in the country. In 2009, the company established the Chinese Academy of Sciences as a CUDA Center of Excellence. In 2011, the Chinese Ministry of Education incorporated CUDA into technology education, and research articles have since been published on the use of CUDA in Chinese industries.
Cut off access
The US government has also cut off China’s access to the latest software tools for developing cutting-edge chips. EDA (Electronics Design Automation) tools are needed early in chip manufacturing – the software allows chip designers to simulate and test a digital copy of the chip, which then helps finalize a physical product for production in a factory.
The American objective is to close access to the tools of companies like Synopsys and Cadence, which dominate the market.
Open source EDA tools such as Google’s OpenROAD are available, but they are mainly used for chip design on older manufacturing nodes.
The restrictions are based on a foreign direct product rule, which was also invoked to ban exports of tech products to Russia after the invasion of Ukraine. The FPDR was previously invoked to deter US companies from doing business with Huawei, whose products are considered a threat to national security.
The export control measures come amid a series of moves by the US government to bolster domestic semiconductor production, research and supply chain. US President Joe Biden recently signed into law the US Semiconductor Production Incentives (CHIPS) Act for America, which unlocks more than $50 billion in public funding to bolster semiconductor infrastructure. American chips.
The CHIPS law has two components focused on chip manufacturing and R&D. Chipmaking initiative opens up $28 million in incentives to a handful of companies that may include Intel, TSMC, and Samsung to build advanced chip factories in the U.S. Intel and TSMC establish manufacturing plants near Phoenix, Arizona, while Samsung is building a factory in Texas. An additional $10 billion in funding is for chipmaking companies such as Texas Instruments to establish factories on older nodes for chips that go into everyday electronics and cars.
The remaining funding of approximately $11 billion will go towards research and development of manufacturing technologies, materials and chip development. The National Institute of Standards and Technology, part of the US Department of Commerce, administers the CHIPS Act programs.
NIST’s CHIPS Program Office now welcomes public comment for manufacturing and research and development initiatives. The contribution is tied to structuring funding, reducing chip counterfeiting and protecting taxpayers by ensuring that incentives are not used for commercial purposes such as share buybacks, which are rampant in technology companies.
NIST also welcomes public comment on the creation of the Manufacturing USA Institutes, which aims to promote private-public collaboration on the latest materials, chips, computer and software technologies. The government is also establishing the institutes as a mechanism for government agencies like DARPA to develop and test the latest chips it can use in equipment like fighter jets.
Intel has already signaled its desire to boost academic research on chips. The company said it would open factories for researchers and academic institutions to obtain physical versions of chips designed in EDA tools. Intel provides access to the Process Development Kit, which will help academics design chips for Intel’s manufacturing nodes, which will then be produced in factories.
“Our intention is to create an environment that is both collaborative and competitive. Think of it like the DARPA Challenge for autonomous driving. It’s a bit like the silicon challenge. We’re trying to encourage all university professors to collaborate on research together but also to build the best intellectual property,” said Bob Brennan, vice president and general manager of Intel Foundry Services, customer solutions engineering.