Biren Technology reportedly plans to lay off one-third of its staff just a week after Taiwan Semiconductor Manufacturing Co. halted shipments of compute GPUs to Biren because of limitations against the Chinese supercomputer sector by the U.S. If the information is accurate, this will severely affect Biren’s ability to develop its GPUs as well as software for artificial intelligence as well as high-performance computing applications.
Since it is unclear whether Biren will be able to get its processors from TSMC any time soon, the company reportedly plans to lay off one-third of its personnel, reports ec.ltn.com.tw (opens in new tab) citing sources familiar with the matter. But slashing headcount severely affects Biren’s ability to develop future chips and software, which ultimately means that its competitive positions against AMD and Nvidia will weaken considerably.
Biren specializes in AI and HPC GPUs, and its top-of-the-range BR100 offering can compete against Nvidia’s A100 and H100 compute GPUs in specific workloads. Biren outsources the production of its processors that contain tens of billions of transistors to TSMC. Starting in early October, such chips cannot be imported to China if made using U.S. technologies (including electronic design automation tools, wafer fab equipment, etc.).
| Biren BR104 | Biren BR100 | Nvidia A100 | Nvidia H100 | |
|---|---|---|---|---|
| Form-Factor | FHFL Card | OAM Module | SXM4 | SXM5 |
| Transistor Count | ? | 77 billion | 54.2 billion | 80 billion |
| Node | N7 | N7 | N7 | 4N |
| Power | 300W | 550W | 400W | 700W |
| FP32 TFLOPS | 128 | 256 | 19.5 | 60 |
| TF32+ TFLOPS | 256 | 512 | ? | ? |
| TF32 TFLOPS | ? | ? | 156/312* | 500/1000* |
| FP16 TFLOPS | ? | ? | 78 | 120 |
| FP16 TFLOPS Tensor | ? | ? | 312/624* | 1000/2000* |
| BF16 TFLOPS | 512 | 1024 | 39 | 120 |
| BF16 TFLOPS Tensor | ? | ? | 312/624* | 1000/2000* |
| INT8 | 1024 | 2048 | ? | ? |
| INT8 TFLOPS Tensor | ? | ? | 624/1248* | 2000/4000* |
The U.S. sanctions against Chinese AI and supercomputer industries are rather strict yet vaguely written. Specifically, AI and HPC chips shipped to China cannot enable machines with the performance of over 100 FP64 PetaFLOPS or over 200 FP32 PetaFLOPS within 41,600 cubic feet (1178 cubic meters). Furthermore, the supercomputer cannot have a throughput of more than 600 GB/s.
While it is not completely clear whether Biren’s BR104 cards or BR100 OAM modules can enable supercomputers with performance that exceeds curbs imposed by the U.S. (to a large degree because Biren does not disclose all performance numbers), TSMC does not want to take any risks and ship Biren’s supercomputer chips to China before it can verify that they do not break any U.S. export rules. Meanwhile, China is the primary market for Biren, and it cannot re-orient to other countries quickly. As a consequence, for now, Biren’s future looks rather gloomy.