Blackwell and DeepSeek Operators Push GLM 5.2 to 300 tps
teortaxesTex · x · 2026-07-04
Technical reviews point out that achieving 300 tps on GLM 5.2 is not difficult, thanks to Blackwell's higher memory and communication bandwidth, combined with DeepSeek's megamoe operator (suspected to only support fp4). The expectation is that 150 tps will become the new normal baseline.
More from Infra
- YC talk on BCI x AI says infrastructure is what really determines speed — garrytan · 2026-07-27
- A 13B model ran on a no-GPU PC by paging weights from SSD via llama.cpp — ID_R_McGregor · 2026-07-27
- llama.cpp warns that GGUFs made before a recent change must be regenerated — EconomySerious · 2026-07-27
- RTX 5090 local tests show Qwen Q6 can drop to 15 tok/s at 80k context — LFAdvice7984 · 2026-07-27
- Surprising Ubuntu Setup: NVIDIA 5090 PC Becomes the Easiest AI Rig — _xjdr · 2026-07-27
- TSMC reportedly plans 5%–10% price hikes in 2027 to cover rising costs — Beth_Kindig · 2026-07-27