A Chinese startup founded by a Google engineer claims to have developed its own TPU chip for AI — the custom ASIC is reportedly 1.5 times faster than Nvidia’s A100 GPU 2020, 42% more efficient.

When you make a purchase through links in our articles, Future and its syndication partners may earn a commission.

Chinese startup Zhonghao Xinying is offering a domestically developed General Purpose Tensor Processing Unit (GPTPU) as an alternative to international AI training and inference hardware such as Nvidia’s graphics cards and Google’s TPUs, the South China Morning Post reported. These ASIC chips are said to be 1.5 times faster than Nvidia’s 2020 A100 release based on its Ampere architecture.

Although it is several years and generations behind the capabilities of the latest hardware from its international competition, it shows the increasing competitiveness for global computing power and how China may find a way to future silicon independence as it explores both traditional GPU and ASIC designs as alternatives.

The “Ghana” chip was developed at the company by Yanggong Yifan, who previously attended Stanford and the University of Michigan to study electrical engineering. He also worked on chip architecture at Google and Oracle, with specific design work on several generations of Google’s TPUs. Co-founder Zheng Hanxun previously worked at Oracle and Samsung’s electronics research and development facilities in Texas.

They claim that the new TPU uses only self-contained intellectual property for core design, with no dependence on Western companies, software stacks, or components for development, design, or manufacturing.

“Our chips do not rely on any foreign technology licenses, ensuring security and long-term sustainability from an architectural level,” SCMP quoted Xinying as saying earlier this year, highlighting the understanding that national security is now closely intertwined with access to semiconductors.

They claim that the Ghana chip is capable of 1.5 times the performance of Nvidia’s A100, as well as “able to reduce power consumption by 75 percent using the manufacturer’s process which is an order of magnitude lower than leading foreign GPU chips.”

If true, that would be an impressive feat, but the benefits are not unheard of for an ASIC, which is a purpose-built chip that excels at certain functions by stripping out all the unnecessary computational elements found in general-purpose silicon like GPUs.

Still, if this Chinese TPU design is anything close to what they claim, it should be quite powerful. The A100 was state-of-the-art hardware five years ago, although 1.5 times that performance still puts Ghana behind Hopper designs from 2022, and far behind the latest Blackwell Ultra hardware.

But for the Chinese market which is still being smuggled in these old GPUs? That could be plenty.

All this comes at an interesting time for the AI ​​chip industry. Although Nvidia has been a dominant force and the face of the industry for the past year, Google’s recent announcement to start hiring and then sell Meta its own TPU silicon has opened the possibility, and very idea, to direct competition. It’s a small-fry deal, albeit one worth billions, but as alternatives emerge in the West, they do in the East as China pushes for more domestic chip production and support through energy subsidy carrots and mandatory quota sticks.

GPUs developed by Nvidia and, to a much lesser extent, AMD, will likely remain the most versatile methods for AI training for some time, but ASICs such as Google’s TPUs, and perhaps even from firms like this one, could offer an interesting option for companies looking to break free from Nvidia’s near-monopoly.

Or just to get hardware access. Memory prices, silicon shortages, and business barriers can all get in the way of even accessing the GPUs your company needs. In their absence, unproven ASICs may be a viable alternative.

Follow up Tom’s Hardware on Google Newsor Add us as a preferred resourceGet our latest news, analysis, and reviews in your feed.