lshqqytiger
A short reflection and thank-you to ZLUDA and a clear recommendation for newcomers
I want to sincerely thank ZLUDA for their long-standing support and the incredible effort behind the project. Without it, I probably wouldn't have been able to get into the world of AI so quickly. Thanks to ZLUDA, I was able to work with ComfyUI and related tools at a usable speed even on an AMD GPU. It made the initial entry into this field possible for me.
But after I really started understanding how things work, building my own workflows, and spending a lot of time experimenting, I slowly realized how frustrating the experience on AMD really is in the long run. Things "work" but just barely. Almost nothing is truly optimized, and everything feels like a workaround.
I’ve been using a GeForce RTX 5070 Ti for the past two months now, and I can honestly say: there’s no comparison to AMD. The difference is massive in speed, compatibility, and especially reliability. Everything works out of the box. No patching, no strange error messages, no guessing.
For example, in Flux Dev 1.0, I can generate a 1920x1080 photo in 45 seconds, without any caching tools like Teacache. Just SageAttention and Fast FP16 enabled. In WaN 2.1, I get 81 tokens at 832x480 in under a minute. Even SDXL models run in mere seconds now, where my former RX 6800 XT took 40–80 seconds, depending on resolution, sampler, CFG, etc.
So if you’re new and thinking about getting into AI: Don’t do it with AMD, unless you have absolutely no other option. Yes, it works somehow, but you're constantly fighting with broken support, strange limitations, and wasted time.
If you have the money or opportunity: go with NVIDIA. Prices have come down again, and the gain in productivity and enjoyment is absolutely worth it.
Still and I mean this sincerely I want to thank ZLUDA once more. If you have to stick with AMD, ZLUDA at least gives you the chance to get something going. It’s not perfect, but it’s something. Without it, I probably would’ve given up early on.
