- cross-posted to:
- [email protected]
- [email protected]
- cross-posted to:
- [email protected]
- [email protected]
You must log in or register to comment.
Some might even go so far as to say that it never should have started
That has become less of an issue now because of a recently announced feature called Windows ML 2.0, which does not distinguish between different NPUs, CPUs, GPUs, and AI chips, O’Donnell said. Microsoft earlier this year also added the Phi and Mu small language models (SLMs) for AI applications to run directly on PCs.
While the new AI features announced at Ignite also take advantage of the NPUs, Microsoft’s hard requirement for a performant NPU has been dwindling, analysts said. There are signs Intel may be deprioritizing NPUs and switching back to GPUs as a minimum compute standard for AI PCs.
While my DIY use cases (i.e. not using cloud services) in ML are somewhat limited in their scope (mostly video upscaling, I honestly gave up on local LLMs and local imagine generation), the NPU concept seems like a bad fit for a market that’s still in its early stages (and likely needs a massive corrective decline for us to understand truly valuable use cases).
Why’d you give up on local image generation? With FLUX-based models and tools like ComfyUI, it’s actually better than what you get with cloud-based services. You have a lot more control, and the wide availability of LoRAs makes it much more fun/useful, IMHO.
Having said that, if you don’t have a modem GPU with at least 8GB of VRAM, it’s not going to be a great experience. 16GB is preferable.
My great wish is for there to be affordable, fast GPUs with at least 32GB of VRAM. That would be enough to play a modern AAA game while also running other AI workloads at the same time (e.g. as a secondary aspect of the game).
I have two really fantastic game ideas that can’t really exist without the average gamer having access to that level of hardware. Not for fancy graphics; for the AI possibilities 😁
I have a 3080 with 10 GB VRAM.
I wasn’t getting very good results with Automatic1111, it felt like I was spending more time fiddling with config than getting the images I was looking for (nothing special mostly generic home hardware type objects, but done in a specific and consistent style and view).
32 GB VRAM won’t be common for at least another 5-7 years IMO, perhaps more.
What model? I highly recommend trying AnimePro Flux in ComfyUI. It generates really great images in just six steps which is like 8 seconds for a 768x768 image on my 4060 TI 16GB. It’d be even faster on your 3080.
Don’t really remember which Stable Diffusion model was used. I followed a guide.
I am just looking for generation of simple two tone flat icons and very basic 3D photo images for products. It to save time making presentations. I would usually just cook something up based on free iconsets or Google images. It actually doesn’t take long.
Gemini works pretty well for this (and support multi -model prompts), but I would prefer to use local models.
