3 ms·
As another data point. Running Qwen3.5 122B at 35t/s as a daily driver using Vulcan llama.cpp on kernel 7.0.0rc5 on a Framework Desktop board (Strix Halo 128).
by lrvick 7mo ago
As another data point.
Running Qwen3.5 122B at 35t/s as a daily driver using Vulcan llama.cpp on kernel 7.0.0rc5 on a Framework Desktop board (Strix Halo 128).
Also a pair of AMD AI Pro r9700 cards as my workhorses for zimageturbo, qwen tts/asr and other accessory functions and experiments.
Finally have a Radeon 6900 XT running qwen3.5 32B at 60+t/s for a fast all arounder.
If I buy anything nvidia it will be only for compatibility testing. AMD hardware is 100% the best option now for cost, freedom, and security for home users.
- syntaxing 7mo agoAre the dedicated GPU cards on another machine or you’re using eGPU with the framework?
- lrvick 7mo agoA separate machine.
- plagiarist 7mo agoHow is the performance for Z-Image on the R9700s?
- lrvick 7mo agoAbout 10 seconds for a 1024x1024 on one, but not found a nice way to scale processing a single image across both.