Skip to content

Home » China unveils its own Biren BR100 graphics chip

China unveils its own Biren BR100 graphics chip


Earlier this month, we reported that Chinese company Birentech is working on its fastest Biren BR100 GPU to date. Based on what the company has publicly announced, the Biren BR100 aims to be a general-purpose GPU that will provide better performance than NVIDIA A100 GPUs in AI processing. Now at Hot Chips 34, the company is bringing us more details on the specifications and architecture of its Biren GPGPU line. Birentech BR100 is the flagship general purpose GPU that China has to offer, with its own GPU architecture using a 7nm process node and accommodating 77 billion transistors in its die.

The GPU was made with TSMC 2.5D CoWoS design and also comes with 300MB on-chip cache, 64GB HBM2e with 2.3TB/s memory bandwidth and support for PCIe Gen 5.0 (CXL interconnect protocol). The entire chip has an area of ​​1074 mm2, which exceeds the grid limit of the technological node. Speaking of the architecture itself, the Biren BR100 consists of two chipsets, each containing 16 SPCs, or stream processing clusters. Each SPC has 16 EUs, and four of these EUs form an internal Compute Unit or CU that is connected to 64 KB of L1 Cache (LSC), while the SPC has a shared 8 MB L2 Cache for all execution units. So, 32 SPCs in total with 512 execution units, 256 MB L2 cache, and 8 MB L1 cache.

Click here for more Tech News


Leave a Reply

Your email address will not be published.