TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

Accelerating Stable Diffusion XL Inference with Jax on Cloud TPU v5e

23 点作者 rayshan超过 1 年前

3 条评论

MasterScrat超过 1 年前
TPUs are amazing for Stable Diffusion.<p>We&#x27;ve been doing training (Dreambooth) and inference on TPUs since the beginning of the year at <a href="https:&#x2F;&#x2F;dreamlook.ai" rel="nofollow noreferrer">https:&#x2F;&#x2F;dreamlook.ai</a>.<p>We basically get 2.5x the training speed for Stable Diffusion 1.5 compared to A100, a very nice &quot;unfair advantage&quot;!
brucethemoose2超过 1 年前
That&#x27;s very fast!<p>I just tried it on my RTX 3090, with a riced linux environment + pytorch&#x2F;xformers nightly, and 4 images take 36.7 seconds on the ComfyUI backend (used by Fooocus-MRE).<p>...But the issue is, right now, you can either pick high quality tooling (The ComfyUI&#x2F;automatic1111 backend UIs) or speed (diffusers-based UIs), not both. InvokeAI and VoltaML do not support SDXL as well as Fooocus at the moment, and all the other UIs use the slow Stability backend with no compilation support.
评论 #37758506 未加载
iAkashPaul超过 1 年前
With some CPU offloading I&#x27;m able to run SDXL at 1.5it&#x2F;s on an RTX2070S w&#x2F; 8GB VRAM.<p>When used with ControlNet it still runs but more layers offloaded at 1.4s&#x2F;it