China’s Largest GPU Developer Slashes Headcount Due to U.S. Sanctions
Biren Technology has announced it will lay off a third of its staff just a week after Taiwan Semiconductor Manufacturing Co. stopped shipping computing GPUs to Biren due to US restrictions on China’s supercomputing sector. reportedly planned. Biren’s ability to develop software for artificial intelligence and high-performance computing applications, not just GPUs.
It’s unclear if Biren will be able to get processors from TSMC anytime soon, so the company is reportedly planning to lay off a third of its workforce. ec.ltn.com.tw (opens in new tab) Citing sources familiar with the matter. But the job cuts will severely impact his Biren’s ability to develop future chips and software, ultimately meaning it will be significantly less competitive against AMD and Nvidia.
Biren specializes in AI and HPC GPUs, and its top-of-the-line BR100 product can compete with Nvidia’s A100 and H100 computing GPUs for certain workloads. Biren outsources the production of processors containing tens of billions of transistors to his TSMC. Starting in early October, such chips will not be allowed to be imported into China if manufactured using US technology (including electronic design automation tools, wafer fabrication equipment, etc.).
Billen BR104 | Billen BR100 | NVIDIA A100 | NVIDIA H100 | |
---|---|---|---|---|
form factor | FHFLCard | OAM module | SXM4 | SXM5 |
number of transistors | ? | 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* |
US sanctions on China’s AI and supercomputing industries are fairly draconian, but vaguely written. Specifically, AI and HPC chips shipped to China cannot enable machines with performance greater than 100 FP64 PetaFLOPS or greater than 200 FP32 PetaFLOPS within 41,600 cubic feet (1178 cubic meters). Furthermore, the throughput of a supercomputer cannot exceed 600 GB/s.
It’s not entirely clear whether Biren’s BR104 card or BR100 OAM module can deliver a supercomputer with performance exceeding the limits imposed by the United States (because Biren does not disclose all performance figures). , significantly), TSMC does not want to take any risks. Biren’s supercomputer chips ship to China before making sure they don’t violate US export regulations. China, on the other hand, is Biren’s main market and cannot be turned to other countries anytime soon.