Live data from Hacker News

Arm Mali G2-Ultra NX GPU: desktop-class mobile gameplay with AI-native graphics

newsroom.arm.com

31–40 of 69 posts

Re: Arm Mali G2-Ultra NX GPU: desktop-class mobile gameplay with AI-native graphics

#31

Earlier quoted context omitted.

Why? I thought these GPU were open source with mainline kernel support?

All the mainline kernel support is reverse-engineered. No help from Arm. Thus there is a considerable lag for the support of newer Mali generations. Some Linux distributions for various embedded computers with Arm-based CPUs and GPUs use closed-source GPU drivers from Arm, not the reverse-engineered Linux drivers, when the Mali versions are newer than supported by the kernel and Mesa. The Arm GPUs do not have public…

>All the mainline kernel support is reverse-engineered. No help from Arm.

That's not true. Arm has been supporting the open source driver effort for the last couple years, but it's still behind the closed source one in terms of performance, features, and HW support.

Re: Arm Mali G2-Ultra NX GPU: desktop-class mobile gameplay with AI-native graphics

#32

Earlier quoted context omitted.

Why? I thought these GPU were open source with mainline kernel support?

All the mainline kernel support is reverse-engineered. No help from Arm. Thus there is a considerable lag for the support of newer Mali generations. Some Linux distributions for various embedded computers with Arm-based CPUs and GPUs use closed-source GPU drivers from Arm, not the reverse-engineered Linux drivers, when the Mali versions are newer than supported by the kernel and Mesa. The Arm GPUs do not have public…

I liked it better when electronic components which didn't have very thorough detailed documentation were instantly recognizable as completely unsuitable for a production BOM. :\

Re: Arm Mali G2-Ultra NX GPU: desktop-class mobile gameplay with AI-native graphics

#33

Earlier quoted context omitted.

Why? I thought these GPU were open source with mainline kernel support?

All the mainline kernel support is reverse-engineered. No help from Arm. Thus there is a considerable lag for the support of newer Mali generations. Some Linux distributions for various embedded computers with Arm-based CPUs and GPUs use closed-source GPU drivers from Arm, not the reverse-engineered Linux drivers, when the Mali versions are newer than supported by the kernel and Mesa. The Arm GPUs do not have public…

I think you are confusing Apple and Arm.

Mali drivers are contributed by Arm:

https://github.com/torvalds/linux/blob/master/MAINTAINERS

Re: Arm Mali G2-Ultra NX GPU: desktop-class mobile gameplay with AI-native graphics

#34
post #6

With IMG PowerVR mostly gone from Mediatek roadmap. Which means ARM Mali is effectively the GPU IP on Android, excluding Qualcomm Adreno which its market continue to shrink and concentrate on premium segment. I wonder how far or how good Mali is now compared to others. I wish Chip and Cheese do compassion on GPU as well.

Samsung is using AMD in upcoming premium chips? Small market share again.

Re: Arm Mali G2-Ultra NX GPU: desktop-class mobile gameplay with AI-native graphics

#35
All the hoopla about "neural graphics" just seems to mean various forms of upscaling (spatial, temporal, denoising).

I guess that's nice and makes sense on mobile, but it somehow feels a bit disappointing that realtime 3D is shifting from modeling to slop filters.

Re: Arm Mali G2-Ultra NX GPU: desktop-class mobile gameplay with AI-native graphics

#36
post #35

All the hoopla about "neural graphics" just seems to mean various forms of upscaling (spatial, temporal, denoising). I guess that's nice and makes sense on mobile, but it somehow feels a bit disappointing that realtime 3D is shifting from modeling to slop filters.

These three forms are trained in order to approximate the ground truth, and they perform significantly better than their non-neural counterparts which do the same thing but worse. Moreover, neither of them is a pure postprocess, since they receive various rendering passes as input, like the depth buffer. They don't really hallucinate anything, just interpolate.

This is another story with DLSS 5, which doesn't reconstruct some ground truth and which is a pure post-processing effect. It just makes the final image look more photo-like in a vague sense. This is more of a "slop filter".

Re: Arm Mali G2-Ultra NX GPU: desktop-class mobile gameplay with AI-native graphics

#39

Earlier quoted context omitted.

All the mainline kernel support is reverse-engineered. No help from Arm. Thus there is a considerable lag for the support of newer Mali generations. Some Linux distributions for various embedded computers with Arm-based CPUs and GPUs use closed-source GPU drivers from Arm, not the reverse-engineered Linux drivers, when the Mali versions are newer than supported by the kernel and Mesa. The Arm GPUs do not have public…

I think you are confusing Apple and Arm. Mali drivers are contributed by Arm: https://github.com/torvalds/linux/blob/master/MAINTAINERS

Most Mali drivers from the kernel are not contributed by Arm, but you are right that nowadays some contributors from Arm have appeared for the older reverse-engineered GPU drivers, like Panfrost and Panthor, and now there are also a few GPU-related drivers, like HDLCD and MALI-DP, which have indeed been contributed by Arm.

Besides the Mali generations that are supported in the mainline kernel, there are many other Mali generations for which Arm has not provided anything else except closed-source drivers for Android, in a few cases also closed-source drivers for Linux. However, there are many other reverse-engineered Mali drivers, besides those from the mainline kernel.

It is good that Arm has begun to contribute to the kernel DRM drivers, but these drivers do little else except providing the means to communicate with the GPUs. The complex work is done in the user-space GPU drivers from Mesa.

The Mesa Lima driver, for older generations of Mali, does not have any contribution from Arm.

In the past, Mesa Panfrost was also completely reverse-engineered. Seeing that now Arm has begun to contribute to the kernel Panfrost DRM, I do not know whether they have started to contribute also to the Mesa Panfrost.

In any case, while some contributions to the previously reverse-engineered drivers are much better than nothing, what would have been much better is to publish the technical documentation of the Mali GPUs, in which case it would not have mattered whether Arm contributes something or not, as anyone could have written the drivers.

Re: Arm Mali G2-Ultra NX GPU: desktop-class mobile gameplay with AI-native graphics

#40

Earlier quoted context omitted.

Given the horrible GPU performance as well as thermal issues in Tensor G5 and G6, Google will have to switch back to Arm in G7. Example benchmark showing how extremely far Google is behind: GeekBench 6 Compute Score (GPU): Xring O3. 36901. (the GPU in this announcement) Snapdragon 8 Elite G5 27501. (Adreno 840) Dimensity 9500M. 23753. (Mali-G1 Ultra MP11). Tensor G6. 6216. (PowerVR CXTP-48-1536)

The Pixel 10 and 11 GPUs are offensively bad. Not only are the Power VR GPUs slow, and less capable feature wise than Mali and Adreno (Xclipse is still around too!), but the tile Google put in the Pixel 11 is actually an _older_ design than what they put in the Pixel 10. They just clocked it higher to make up the performance difference. And then Google has the gall to charge basically flagship prices for their under…

I have fond memories from back when Pixels were actually flagship. The Pixel 2 is my nostalgic pinnacle. I'd totally forgotten about that squeeze gesture until I got my old Pixel 2 out to erase.
Post reply on HN