Not available in English yet
RTX 3090 单卡跑 Qwen 跑到 140 token/s — 老硬件的天花板又被掀高
Related Reading
More on #Qwen
$2,800 Builds a 256GB VRAM AI Rig — The Cheap Local LLM Era Opens
$2,800 buys a 256GB VRAM AI rig: 8 used AMD Radeon cards, Claude writes the RDNA2 kernel, 3000+ tokens/sec.
Qwen user says local small model feels like Claude — one sample, one signal
Reddit user: quantized Qwen 3.8 Flash Next 'feels like Claude 4.6' writing three.js. One anecdote — open/local models may be closing the API gap.
Qwen 8B Hits 400 Tokens/Sec — Self-Hosted AI Cost Inflection Point
Qwen3-8B-Flash hits 400 tokens/sec on an RTX 6000; 8K prefilling nears 14K. The self-hosted AI affordability threshold is arriving for enterprises.
RTX 5090 May Be Discontinued: NVIDIA Reserves Top Chips for Enterprise
RTX 5090 was the go-to consumer GPU for local large-model inference. A new rumor claims NVIDIA will discontinue it to reserve GB202 for enterprise.
Strata Wipes Claude Co-Author Tags — Industry Split on AI Ghostwriting
Strata rewrote its Git history to scrub every 'Co-Authored by Claude' tag — small incident, big gray debate on AI ghostwriting disclosure.
76 GPUs Benchmarked: Memory Bus Width Outweighs Newer Silicon for Local AI
76 GPUs tested against second-hand prices, generation speed, and VRAM capacity. Wide memory buses beat newer cards on local LLM value.