BLAST RADIUS

Open-weight & owned stack · Quantization efficiency

Developer Releases Ampere-Focused llama.cpp Fork

Sep 15, 2026

A developer released llamAmpere and a Qwen3.8-27B quant claiming more than 90 tokens per second on RTX 3090 systems.

Read the original at reddit.comOpens the publisher's site in a new tab

More in Open-weight & owned stack