Domestic DF1000 Chip Claimed to Rival Hopper Performance

The domestic DF1000 AI chip utilizes a mature 14nm process but claims a peak computing power of 520 TFLOPS BF16 and 6.4TB/s memory bandwidth. Backed by industry veteran Wei Shaojun, the chip's reconfigurable architecture is described as achieving practical performance close to Nvidia's Hopper architecture.

2026-07-22 ~ 2026-07-23 · 3 related posts