FLUX 3 Klein INT8 tested on RTX 3090/5090: 2x speedup, better accuracy than FP8
CeFurkan · reddit · 2026-10-17
Blogger CeFurkan benchmarked FLUX 3 Klein's ConvRot Int8 HQ quantization on an RTX 3090 and RTX 5090 in the same PC:
- Speed: text-to-image on 3090 drops from 0.384s (BF16) to 0.175s (2.19×), image edit 2.29×; on 5090, 113ms → 52ms (2.18×), consistently over 2x faster
- Accuracy: deviation from BF16 output is only 2.14%, beating official INT8 same-attention (2.26%) and default INT8 (3.04%); FP8 Scaled was actually slower than BF16
- Full ComfyUI tutorial video linked in the post
More from Infra
- Nebius is building the next AWS; CoreWeave is a GPU datacenter OEM — kevinsxu · 2026-10-17
- If cyberspace were Earth, the biggest AI datacenter would exceed the EU — LeviTurk · 2026-10-17
- Ranking seven AI compute plays on survivability: Nebius first, CoreWeave fifth — pdamodaran · 2026-10-17
- Qwen 27B at Full 262k Context on a 16GB RX 9070 XT: 80+ t/s Decode With Adaptive KV Streaming — bottlestreet3d · 2026-10-17
- Cloudflare open-sources Pipecat voice agent on Python Workers with WebRTC — ritakozlov · 2026-10-17
- cuda.fast leaderboard: speculative decoding makes Bonsai 2 9B run 240.9% faster on DGX Spark — HankYeomans · 2026-10-17