Shanghai Tianzhu Zhi Zhi Core Semiconductor's new 7nm GPU is now up and running. This new chip is a beast, with no less than 24 billion transistors. Incidentally, Nvidia's GA102 GPU in the RTX 3080 and 3090 graphics boards has 28 billion transistors.

Known as Big Island (BI), this new GPU is capable of up to 147 TFLOPS of FP16 performance, which, as Tom's Hardware points out, is comparable to similar products from Nvidia and AMD. Nvidia's A100 is 77.5 TFLOPS. 97TFLOPS and the AMD Instinct MI100 delivers 184.6TFLOPS of FP16 performance, but the Nvidia A100 also includes a Tensor core with an additional 312TFLOPS of FP16 performance.

Regarding production technology, Shanghai Tianshu Zhixin Semiconductor stated that the BI will be manufactured at 7nm. While the details provided appear to be consistent with TSMC's 7nm node, no foundry partners have been identified, and it is unclear whether China has domestic chip manufacturing technology in that class. Images of the new GPUs suggest a multi-chip package in which the GPU itself is housed along with on-board memory and cache.

Of course, BI chips and competing products from AMD and Nvidia are mentioned, but these are all GPUs built specifically for enterprise applications, not desktop PCs or gaming. GPUs for gaming require a wide range of hardware beyond the compute cores that produce TFLOPS, including texture units, pixel processing, and a full pipeline of render outputs.

It is unclear whether BI GPUs will have these features, and indeed, given their mission, there is little reason to include them. Therefore, it is a safe guess that Shanghai Tianzhou Zhixin Semiconductor's new GPU will not pose a threat to established competitors in game graphics.

Nevertheless, the simple fact that a large, ultra-complex GPU has been produced that is comparable to existing players in terms of floating-point performance is important. It is a big step from there to a gaming GPU. But it is definitely a start.