MiniMax H3 AMD Optimization: Sol-Attn and BlockCache Boost Performance
DrBearJ3w · reddit · 2026-08-14
Testing MiniMax H3 on RX 7900 XTX, Sol-Attn and BlockCache reduce Turbo run from 2:59 to 2:36, and 20-step runs from 5:14 to 3:44. Detailed configs and links provided.
More from Infra
- Qwen3.8-27B-FP8 on GH200: 10 concurrent streaming requests, first token in 10ms — MaziyarPanahi · 2026-08-15
- Qwen3.8-27B Serving Configs: DGX Spark vLLM and RTX 4090 llama.cpp — erdaltoprak · 2026-08-15
- AI video billing metered by pixel area, not duration: developer warns to reconcile invoices — DryProgress9179 · 2026-08-15
- Analyst: Memory Cycle Structural, Sentiment Should Shift — BenBajarin · 2026-08-15
- DeepSeek-V4-Pro launches: 1.6T-param MoE cuts inference FLOPs to 27% of V3.2 — AccBalanced · 2026-08-14
- Qwen 3.8-Max Now on Modal: 2.4T Params, 1M Context, Custom DFlash Speculator — AAAzzam · 2026-08-14