Biren Expertise reportedly plans to put off one-third of its employees only a week after Taiwan Semiconductor Manufacturing Co. halted shipments of compute GPUs to Biren due to limitations towards the Chinese language supercomputer sector by the U.S. If the data is correct, this can severely have an effect on Biren’s capacity to develop its GPUs in addition to software program for synthetic intelligence in addition to high-performance computing functions.
Since it’s unclear whether or not Biren will have the ability to get its processors from TSMC any time quickly, the corporate reportedly plans to put off one-third of its personnel, stories ec.ltn.com.tw (opens in new tab) citing sources accustomed to the matter. However slashing headcount severely impacts Biren’s capacity to develop future chips and software program, which finally implies that its aggressive positions towards AMD and Nvidia will weaken significantly.
Biren makes a speciality of AI and HPC GPUs, and its top-of-the-range BR100 providing can compete towards Nvidia’s A100 and H100 compute GPUs in particular workloads. Biren outsources the manufacturing of its processors that comprise tens of billions of transistors to TSMC. Beginning in early October, such chips can’t be imported to China if made utilizing U.S. applied sciences (together with digital design automation instruments, wafer fab gear, and many others.).
| Biren BR104 | Biren BR100 | Nvidia A100 | Nvidia H100 | |
|---|---|---|---|---|
| Type-Issue | FHFL Card | OAM Module | SXM4 | SXM5 |
| Transistor Depend | ? | 77 billion | 54.2 billion | 80 billion |
| Node | N7 | N7 | N7 | 4N |
| Energy | 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 towards Chinese language AI and supercomputer industries are somewhat strict but vaguely written. Particularly, AI and HPC chips shipped to China can’t allow machines with the efficiency of over 100 FP64 PetaFLOPS or over 200 FP32 PetaFLOPS inside 41,600 cubic ft (1178 cubic meters). Moreover, the supercomputer can’t have a throughput of greater than 600 GB/s.
Whereas it isn’t fully clear whether or not Biren’s BR104 playing cards or BR100 OAM modules can allow supercomputers with efficiency that exceeds curbs imposed by the U.S. (to a big diploma as a result of Biren doesn’t disclose all efficiency numbers), TSMC doesn’t need to take any dangers and ship Biren’s supercomputer chips to China earlier than it could possibly confirm that they don’t break any U.S. export guidelines. In the meantime, China is the first marketplace for Biren, and it can’t re-orient to different international locations rapidly. As a consequence, for now, Biren’s future seems to be somewhat gloomy.