China presents its first competitive 7 nm data center GPU
Boosted by the trade wars with the United States started during the Trump administration, China is now more than ever eager to prove that it too can produce powerful computer processing tech. The rise of the Chinese tech was first led by the production of affordable yet powerful smartphones from Huawei, which “coincidentally” became the biggest target for bans in the U.S. Nevertheless, China is also striving to diminish the reliance on U.S. tech, and has been trying to produce its own computer processors and GPUs for the past few years. While the Chinese foundries are still playing catch-up on the node miniaturization side, they are certainly not too far behind industry leaders like TSMC. As far as performance goes, the X86 CPUs produced in China are admittedly not yet impressive enough and certainly cannot compete with what Intel and AMD currently have, but we are seeing promising results with ARM-based processors and, more recently, with GPUs like the ones produced by Tianshu Zhixin.
Unlike the Jingjia Micro JM5400 GPUs announced back in 2019, the new Big Island models from Tianshu Zhixin are not specifically designed to compete with the gaming Nvidia RTX and AMD Radeon models, as they are more tailored towards AI and HPC applications, plus other general purpose uses for the education, medicine and security sectors. Tianshu Zhixin’s Big Island GPGPUs were developed between 2018 and 2020, and the original plan was to release them in 2H 2020, but the pandemic pushed back the release to 2021. According to Tom’s Hardware, the Big Island integrates 24 billion transistors and is said to be produced on a 7 nm process node with 2.5D CoWoS packaging that most likely originates at TSMC’s fabs. As with all home-grown products, Zhixin’ new GPGPUs offer an enticing performance/cost ratio compared to Nvidia’s A100 and AMD’s Instinct MI100 solutions, and include support for a wide array of floating point formats including FP32, FP16, BF16, INT32, INT16, INT8 etc.
Most of the home-grown Chinese processors we have seen thus far under-delivered when it comes to direct performance comparisons with mainstream U.S. models, and thus are probably not going to have much of an impact on the global market. However, the Big Island GPU might see global availability if the claims regarding its floating point performance prove to be true. Tianshu Zhixin revealed that the FP16 performance of the Big Island chips can reach 147 TFLOPS, which is comparable to AMD’s Instinct MI100 with 184.6 TFLOPS, but clearly cannot match Nvidia's A100 with 312 TFLOPS. If China keeps up like this on the performance side and also offers aggressive price schemes, we could be seeing some decent alternatives to the U.S. models in just a few years from now.