Alexey Fateev(@superalesha):I spent 67 hours of model time to find out how much dumber 4 bit really makes Qwen3.8-27B. FP8 vs NVFP4 vs AWQ INT4 vs GGUF Q4_K_M vs NInfer on my 4x RTX 3090. 4,800 tasks, 10,120 requests, 14.5M reasoning tokens, no token caps anywhere. The results surprised me. Big thread, lets go 🧵

Alexey Fateev

@superalesha

⚡I benchmark local LLMs on 4x RTX 3090s. exact configs, tok/s, VRAM, and what broke. ❤️ - 2xDGX Spark 🚀96GB VRAM | Local AI

Joined January 2026

325 Following 3.7K Followers

Alexey Fateev@superalesha

2026.08.20 06:03

I spent 67 hours of model time to find out how much dumber 4 bit really makes Qwen3.8-27B. FP8 vs NVFP4 vs AWQ INT4 vs GGUF Q4_K_M vs NInfer on my 4x RTX 3090. 4,800 tasks, 10,120 requests, 14.5M reasoning tokens, no token caps anywhere. The results surprised me. Big thread, lets go 🧵