Chinese computing GPU developer Biren is considering an initial public offering (IPO) in Hong Kong this year. That’s because domestic customers are increasingly preferring the company’s AI chips to Nvidia’s, which is expensive and in short supply, Bloomberg reports. Hoping to seize this opportunity, tech start-ups are poised to capitalize on the increased demand for their products.
Citing unnamed sources, the report said Biren plans to file for its first stake sale in the coming weeks. At the same time, Biren is in talks with potential investors, including a government-backed fund in Guangzhou. These discussions focus on another independent funding round that could bring in about 2 billion yuan ($279 million). Biren was looking to raise a valuation of 17 billion yuan (about $2.4 billion) last year. As of now, Biren has yet to determine the scope and exact duration of his IPO.
Biren is confident in its valuation because its products appear competitive (on paper, at least) compared to Nvidia’s compute GPUs, and the market for AI-enabled compute GPUs has recently increased. Because it’s booming.
Biren’s debut computing GPU family will consist of two options, the BR100 and BR104. A “baseline” BR104 offers up to 128 FP32 TFLOPS or 1 INT8 PetaFLOPS of performance, while a high-end BR100 (basically two BR104s on one silicon interposer) delivers up to 256 FP32 TFLOPS or 2 INT8 PetaFLOPS. Provides FLOPS performance. The middle tier BR104 has 32 GB of HBM2E memory and uses a 2048-bit interface providing 819 GB/s bandwidth. In contrast, the premium BR100 comes with 64 GB of HBM2E memory and has a 4096-bit interface with a bandwidth of 1.64 TB/s.
| Header Cell – Column 0 | 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* |
There is another reason for Mr. Biren’s optimism. The company’s fundraising activities are consistent with the Chinese government’s vigorous push to develop the domestic semiconductor industry. The move is a response to a US-led campaign to prevent Chinese companies from acquiring a number of computing GPUs from AMD, Intel and Nvidia, which compete with Biren’s offerings. Media reports say that Nvidia’s products are expensive and in short supply, so Biren can sell more GPUs, at least to companies that aren’t using Nvidia’s CUDA software stack for AI workloads.
But Biren also faces many challenges. Last year, TSMC temporarily halted shipments of computing GPUs to Biren to ensure they meet US export regulations in terms of performance and features. As a result, the company was forced to cut staff to cut costs. Apparently Biren can procure enough silicon for now and his main job at the moment is to make sure his software stack is competitive compared to his Nvidia, Intel and AMD software stacks. is.
It’s very difficult to beat Nvidia in this area. The company has spent nearly two decades improving his CUDA, and in recent years he has invested hundreds of millions of dollars in creating his CUDA platform best suited for AI development. For now, many Chinese hyperscalers prefer to use his Nvidia GPUs in their AI products due to CUDA’s superiority and the amount of money they have already invested in this ecosystem. increase.
