#Intel Arc integrated GPU in Intel Core Ultra system and W-Okada

1 messages · Page 1 of 1 (latest)

merry phoenix
#

I know this might not be a typical thread about "how to install and use voice changer" trope like most threads here, but I, as a helper, think there was no one talking about this yet and maybe it's time to ask because I haven't heard any actual answer from this server so far.
So as some of you might know, newer Intel Arc integrated (e.g. Arc B390) GPU within "Intel Core Ultra 3xx" system has been stated to perform almost closer to those entry-level dedicated GPUs (like NVIDIA GeForce RTX 4050 Mobile), can perform games at around 60 - 90 FPS of 1080p, significantly superior to those "Intel UHD Graphics" in traditional Intel Core (e.g. Intel Core i5-1335U) systems, although still borrowing memory from the main PC RAM which is the main characteristic of an integrated GPU. Additionally, there's AMD Radeon 890M the iGPU from AMD Ryzen AI system as well, this one aims to perform better than "AMD Radeon(TM)" integrated GPU in traditional AMD Ryzen systems, surprisingly outperforms older dedicated GPUs like Radeon RX 580 and 480, can performs games also at around 60 - 90 FPS of 1080p, although it has been reported that "Intel Arc B390" integrated outperforms "AMD Radeon 890M" in most scenarios. While integrated GPUs like Intel UHD Graphics and AMD Radeon(TM) have been long known to perform the fraction of their CPUs or worse when running a W-Okada voice changer, respectively, the higher-performance integrated ones like Intel Arc B390 and AMD Radeon 890M got me thinking if these specific integrated GPUs could get any better. There have been an individual here who got their hands on an "AMD Ryzen AI" laptop, which as of now is very rare here, by the way.
So the question is, could Python-related programs like RVC/W-Okada/UVR5, especially their DirectML variant, potentially work with these specific integrated GPUs? The lower performance for these iGPUs is expected. I just wanna hear your thoughts about this.

wide palm
#

I think at least in AMDs case the CPU packages might already be capable of doing that, that is if someone is willing to update PyTorch.
https://www.amd.com/en/resources/support-articles/release-notes/RN-AMDGPU-WINDOWS-PYTORCH-7-1-1.html
AMD makes it clear with posts like these that they're specifically trying to optimize PyTorch on their latest hardware. Most of AMD's AI tech is also based on Ryzen (CPUs), rather than their Radeon series (GPUs).
https://www.amd.com/en/products/processors/consumer/ryzen-ai.html
From what I understand, the way AMD integrates their GPU part, it has direct access to the system RAM and can directly talk to the CPU portion as well, and so likely to the package's NPU. It would need testing I suppose, but I am positive that there isn't much difference in performance between an older gen desktop GPU and these newer AI Ryzen CPU packages.

#

It also looks like ROCm lets you interact with the iGPU. In fact ComfyUI (an image generation tool) apparently makes use the iGPU to generate images. (rocm has full apu support)

#

I haven't kept up with Intel, so no clue there.

wide palm