BLAST RADIUS

Open-weight & owned stack · Quantization efficiency

Fork boosts Qwen decoding on dual Radeon GPUs

Sep 18, 2026

A llama.cpp fork reportedly increased Qwen 3.8 Q8 decoding from about 28 to 82 tokens per second on two Radeon RX 7900 XTX GPUs.

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

More in Open-weight & owned stack