News Intel's new graphics driver lets you dedicate up to 87% of laptop memory capacity to the iGPU for VRAM — Core Ultra CPUs get "Shared GPU Memory Ov...

Admin

Administrator
Staff member
Intel’s latest graphics driver (32.0.101.6987) introduces a “Shared GPU Memory Override” feature for certain Core Ultra CPUs with Arc graphics, letting users manually choose the amount of VRAM that's reserved from system RAM. The update also claims to offer smoother Windows animations, modest gaming FPS gains, and reduced resource usage.

Intel's new graphics driver lets you dedicate up to 87% of laptop memory capacity to the iGPU for VRAM — Core Ultra CPUs get "Shared GPU Memory Ov... : Read more
 
And it also BREAKS Battlefield 6 from running...even AFTER the developer told them in advance of the release that it did.


It's becoming more and more obvious that Intel lost it's mojo, and that the problems they are going through are not going to be resolved any time soon....cuts to staff aren't going to help, either.
 
>And it also BREAKS Battlefield 6 from running...even AFTER the developer told them in advance of the release that it did.

This isn't about gaming, but for LLM localhosting. Intel is following AMD's footsteps here. As mentioned, AMD has had Variable Graphics Memory (VGM) for Strix Point since last year (but notably not for Hawk or Phoenix Point). AGM allows GPU allocation of up to 75% of system memory. Just recently, that upped to 87% for Strix Halo, coincidentally the same as what Intel now allows.

Again, this isn't for better gaming performance. Large memory size would encourage loading of large textures, which on iGPU would actually decrease performance.

As for breaking BF6 beta, the world doesn't revolve around BF6. Intel cares more about AI than it does for PC gaming.

From the horse's mouth,

View: https://x.com/bobduffy/status/1956141865037938814

Bob Duffy (Director Graphics & AI Evangelism at Intel): "If you have Intel Core Ultra and are doing AI, you're going to want to update to latest Intel Arc driver... because this"
 
Alas, the ability to use RAM for both CPU and GPU workloads was part of AMD's original Heterogeneous Systems Architecture (HSA), well described in Anandtech articles on those earlier APUs now lost to the public.

But there the current somewhat artifical divide between CPU and GPU RAM was meant to basically disappear completely, in its ultimate form allowing the granularity of a functional call for a control transfer between CPU and GPU code, blurring the distinction between ISAs and computing paradigms.

Today that might impose more limitations than provide benefits as GPUs and CPUs have grown ever further apart in terms of how transistor budgets are better spent on extra cache layers vs register files or extra function blocks.

And even then there just wasn't any easy way to support HSA in a high-level programming language or via a single compiler.

In any case that barrier was always just values in control registers and drivers supporting that. And in theory drivers and the OS should be able to manage that with a high degree of flexibility.

The main reason this felt fixed in the past is that software like they people who wrote it, often can't manage freedom without getting into conflict with others, trying to do the same. So vendors just didn't expose the flexibility designed into the hardware.

These days, where everyone thinks iGPUs working on much bigger chunks of system RAM for something AI is worth paying extra for the hardware you own already, exposing a golden gilded lever paid in real money, has vendors quickly push their marketing AI bots into overdrive.
 
And it also BREAKS Battlefield 6 from running...even AFTER the developer told them in advance of the release that it did.


It's becoming more and more obvious that Intel lost it's mojo, and that the problems they are going through are not going to be resolved any time soon....cuts to staff aren't going to help, either.
I know you're basically an anti-Intel troll at this point so my words will probably be wasted but... did you actually read either article?

During the BF6 beta there was a driver bug which Intel had to address for Alchemist based graphics. That has nothing to do with this feature being added. A beta driver fix was already published for the BF6 issues a week ago as well.
 
And it also BREAKS Battlefield 6 from running...even AFTER the developer told them in advance of the release that it did.


It's becoming more and more obvious that Intel lost it's mojo, and that the problems they are going through are not going to be resolved any time soon....cuts to staff aren't going to help, either.
Wrong. The launch driver did have some issues with Alchemist cards but they quickly tossed out a hot fix which quickly remedied that. VRAM allocation had nothing to do with the crashes.
 
  • Like
Reactions: SirStephenH
Who is this even for? Windows already allows iGPUs to use up to 50% of your system RAM by default without fussing with BIOS settings (The typical 512MB BIOS allocation, sometimes higher or lower, is for backwards compatibility and doesn't restrict the total amount of RAM available to the iGPU. It pre-allocates the RAM and tells games and apps that the iGPU has a minimum of that amount of RAM when they do VRAM checks. The OS normally controls the maximum available RAM for the iGPU.). If you're already RAM starved then you can't spare the extra system memory, if you have plenty of RAM then your iGPU probably already has more memory than it knows what to do with with. No amount of shared RAM is going to make running a large LLM on an iGPU worthwhile and games are held back by the iGPU and RAM speed, not the available shared RAM.