May I inquire whether the 4090 can participate in this project? For instance, in inference tasks？

[wenter]

[HioZx]

I have the same problem

[min-star]

me too

[zdyshine]

me too, can inference on 4090?

[tanghaom]

我用一块3090，可以跑16x256x256和16x512x512的，batchsize设成1

[min-star]

can you share your environments config?

---Original--- From: "Haoming @.> Date: Tue, Mar 19, 2024 15:51 PM To: @.>; Cc: @.@.>; Subject: Re: [hpcaitech/Open-Sora] May I inquire whether the 4090 can participate in this project? For instance, in inference tasks？ (Issue #104)

我用一块3090，可以跑16x256x256的

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>

[aguang1201]

@tanghaom 16x512x512的好像不行吧？把num_frames改成4倒是可以，但效果惨不忍睹

我用一块3090，可以跑16x256x256和16x512x512的，batchsize设成1

[tanghaom]

@tanghaom 16x512x512的好像不行吧？把num_frames改成4倒是可以，但效果惨不忍睹

我用一块3090，可以跑16x256x256和16x512x512的，batchsize设成1

你把vae那个步骤单独跑就行了，前面的生成部分24G显存够用； 你可以拆开跑，比如前面模型用一块卡，vae用另一块卡；或者vae用cpu；或者去掉生成的模型，vae在单独的流程里跑。

[chaojie]

https://github.com/chaojie/ComfyUI-Open-Sora 可以用我这个项目，4090可以跑16x512x512的