Xiwang released the next-generation inference chip S3

36Kr learned that the domestic GPU manufacturer Sunrise has released the new generation inference chip S3. In terms of computing power and storage design, the S3 supports free switching between FP16 and FP4 precision, and is the first to adopt LPDDR6 memory solutions in domestic GPGPU products, with memory capacity increased by 4 times compared to the previous generation, alleviating the common memory bottleneck issues in large model inference. The unit Token inference cost on mainstream large models such as DeepSeek is reduced by approximately 90% compared to the previous generation