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.
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 powered junk.
Personally I don't buy flagship phones because I don't use their full compute capabilities anyway, but a Pixel 11 would cost 2x what I paid for my Pixel 8. For 0 GPU performance improvement.
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.
I just can't wait to see it being tested in an actual phone. So far all the data are done on development board. If they are really as good as everything we have read so far, both Mediatek and Qualcomm are in big trouble.
If only that would mean a better driver story across devices and Android releases.
Unless it is a Pixel or Samsung Sxy flagship device, chances are the driver, which will never get updated, it a box full of interesting developer moments.
Just go to chrome://gpu to check the amount of workarounds enabled in Chrome alone.
thank you, will be using this, it’s far more accurate than simply throwing the devs under the bus and ignoring all the unspoken context of said moments, while not papering over the… interesting developer moments.
Since when do Samsung ship driver updates? I'm being a bit hyperbolic but from my experience Samsung don't ship them very often either. To Google's credit, the Pixels get them quite frequently. Samsung is still better than most other Android vendors, who are somehow even worse.
It's depressing how bad the drivers are. They have 0 market pressure to improve the quality, unlike PC where the whiniest group on the internet (gamers on reddit) will crucify you for driver problems. I've spent too much time investigating and implementing workarounds. Though it does pay my bills...
Moats are obstacles against something you're trying to avoid. It can be competitors coming in, or customers going out. A moat could surround a prison to keep prisoners in. Some companies just choose to treat their customers as prisoners.
Because most game developers aren't religious about APIs, they implement an abstraction layer in their engine, a practice since the heterogeneous 8 and 16 bit days, where all major games were mostly coded in Assembly, and move on.
Any conference talk schedule at one of these conferences (https://gameconfguide.com) will prove the point of how little they care.
It is all about gameplay, taking advantage of gaming IP, franchaises, how to squeeze the ultimate performance out of specific hardware, user engagement,....
You will find an occasional talk about open APIs, probably even done by some Khronos folks, and that's it.
If they open-sourced their userland drivers, and/or released their internal technical documentation, which would have a similar effect, they would open themselves to at least two big problems.
First, they would spill all of their secret sauce for others to "draw inspiration" from. And there is a lot of expensive secret sauce in there. Not all of it is protected by parents either; more on that below.
Second, they would make it a lot easier for competitors and patent trolls alike to seek litigation. One of the criteria thech companies have for filing a patent is how easy is it to detect when it is being violated. Open source your stuff, and you make it much easier for the competition to detect any violation.
And it is not like these companies are violating patents intentionally; much the opposite. But there are only so many ways to implement Vulkan or DirectX, and completely independent teams will come up with similar enough solutions, and with broad enough patents, that there will be overlap every now and then. That is why they file patents in the first place. Only the high cost of litigation and the risk of mutually-assured destruction keeps legitimate vendors from rattling sabers more often.
Then you have NVidia with their immense investment and advantage on the software side, from the CUDA runtime to countless libraries like CUBLAS and cuDNN. That investment goes much beyond a relatively simple Vulkan/DirectX driver.
So the question is less "why don't more of them open-source userland". The question is "why do any of them do it"?
Source: former insider with enough patents under my belt that I lost count.
Do you have any thoughts on the question you posed? Amdgpu and mesa is working out fairly well for amd, but yes, there is also still a proprietary userland for it too. Intel I think is using mesa code for their Windows drivers too. How are they different?
I genuinely don't know. They are smart people, they know the game. They must have figured out that there is an upside for them.
Just speculating here. Maybe they have big customers that communicated to them that they will choose them over the competition if they open more of their software stack. Maybe they are less afraid of being sued because they are the underdog and they believe that the authorities would not be kind to a behemoth suing them, or they believe that the behemoth would not sue them because of the bad press it would generate vs the settlement they would get.
I actually don't know. But I am sure they have good reasons, just like the corps that keep it closed also have good reasons. It is not "good" vs "evil".
> If they open-sourced their userland drivers, and/or released their internal technical documentation, which would have a similar effect, they would open themselves to at least two big problems.
> First, they would spill all of their secret sauce for others to "draw inspiration" from. And there is a lot of expensive secret sauce in there. Not all of it is protected by patents either
Then why does Apple allow Asahi to reverse-engineer their low-level APIs? Are their lawyers not paying attention?
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 technical documentation, not even at the level of NVIDIA, much less something comparable with AMD and Intel.
>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.
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. :\
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.
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".
The headline seems like a way to set themselves up to fail. The advancements are great, those features are definitely something a lot of developers and players are interested in, but comparing it to desktop immediately just sets the bar too high.
Once one is over the odor of the headline, the announcement reads:
* Neural Super Sampling (NSS) reconstructs higher-resolution images from lower-resolution renders, reducing the amount of conventional rendering work needed to produce a high-quality final image.
* Neural Frame Rate Upscaling (NFRU) generates intermediate frames to increase frame rates and deliver smoother gameplay.
* Neural Super Sampling and Denoising (NSSD) combines neural upscaling with denoising to improve image quality in demanding ray-traced scenes with complex lighting and shadows.
This is probably not about "mobile users" as in Android Phones & iPhone users .. but about the current (trendy) rise of handheld consoles ? Atleast that's what I think
the switch, steam deck and competitors will likely look into ARM, the cheaper ones already use it, flagship and gaming phones, the latter is bigger than you think, normal mobile gaming is big in third world countries and china, and emulation continuous to get better, I can ToTK and The Witcher 3 (windows version) on my phone, and it's a few years old flagship
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.
Google just switched to PowerVR, didn't they?
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:
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 powered junk.
Personally I don't buy flagship phones because I don't use their full compute capabilities anyway, but a Pixel 11 would cost 2x what I paid for my Pixel 8. For 0 GPU performance improvement.
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.
Back when the point of Pixel was to tell the world: THIS is what you can do with Android.
Then Samsung got the message, ran with it, and Google went "Ehh, we're good now."
>Xring O3.
I just can't wait to see it being tested in an actual phone. So far all the data are done on development board. If they are really as good as everything we have read so far, both Mediatek and Qualcomm are in big trouble.
This Xiaomis second or third chip ever so I assume there will be problems.
it's available in retail (the foldable phone variant) in mainland China since today
Why did that happen?
If only that would mean a better driver story across devices and Android releases.
Unless it is a Pixel or Samsung Sxy flagship device, chances are the driver, which will never get updated, it a box full of interesting developer moments.
Just go to chrome://gpu to check the amount of workarounds enabled in Chrome alone.
> …a box full of interesting developer moments.
thank you, will be using this, it’s far more accurate than simply throwing the devs under the bus and ignoring all the unspoken context of said moments, while not papering over the… interesting developer moments.
Since when do Samsung ship driver updates? I'm being a bit hyperbolic but from my experience Samsung don't ship them very often either. To Google's credit, the Pixels get them quite frequently. Samsung is still better than most other Android vendors, who are somehow even worse.
It's depressing how bad the drivers are. They have 0 market pressure to improve the quality, unlike PC where the whiniest group on the internet (gamers on reddit) will crucify you for driver problems. I've spent too much time investigating and implementing workarounds. Though it does pay my bills...
Samsung is using AMD in upcoming premium chips? Small market share again.
[delayed]
Panthor is supposedly pretty good, it's no rtx, but people do game on them.
What about support in mesa?
Give it a couple years before Valve starts sponsoring someone to reverse-engineer it
I do not understand why vendors believe their software stack is a moat. It is an obstacle to me.
opens wikipedia (Emphasis mine:)
> Moats were excavated around castles and other fortifications as part of the defensive system as an obstacle immediately outside the walls.
Unfortunate wording?
Moats are supposed to be obstacles for your competitors, not your consumers.
It's akin to building a moat inside the castle.
Moats are obstacles against something you're trying to avoid. It can be competitors coming in, or customers going out. A moat could surround a prison to keep prisoners in. Some companies just choose to treat their customers as prisoners.
Because most game developers aren't religious about APIs, they implement an abstraction layer in their engine, a practice since the heterogeneous 8 and 16 bit days, where all major games were mostly coded in Assembly, and move on.
Any conference talk schedule at one of these conferences (https://gameconfguide.com) will prove the point of how little they care.
It is all about gameplay, taking advantage of gaming IP, franchaises, how to squeeze the ultimate performance out of specific hardware, user engagement,....
You will find an occasional talk about open APIs, probably even done by some Khronos folks, and that's it.
If they open-sourced their userland drivers, and/or released their internal technical documentation, which would have a similar effect, they would open themselves to at least two big problems.
First, they would spill all of their secret sauce for others to "draw inspiration" from. And there is a lot of expensive secret sauce in there. Not all of it is protected by parents either; more on that below.
Second, they would make it a lot easier for competitors and patent trolls alike to seek litigation. One of the criteria thech companies have for filing a patent is how easy is it to detect when it is being violated. Open source your stuff, and you make it much easier for the competition to detect any violation.
And it is not like these companies are violating patents intentionally; much the opposite. But there are only so many ways to implement Vulkan or DirectX, and completely independent teams will come up with similar enough solutions, and with broad enough patents, that there will be overlap every now and then. That is why they file patents in the first place. Only the high cost of litigation and the risk of mutually-assured destruction keeps legitimate vendors from rattling sabers more often.
Then you have NVidia with their immense investment and advantage on the software side, from the CUDA runtime to countless libraries like CUBLAS and cuDNN. That investment goes much beyond a relatively simple Vulkan/DirectX driver.
So the question is less "why don't more of them open-source userland". The question is "why do any of them do it"?
Source: former insider with enough patents under my belt that I lost count.
Do you have any thoughts on the question you posed? Amdgpu and mesa is working out fairly well for amd, but yes, there is also still a proprietary userland for it too. Intel I think is using mesa code for their Windows drivers too. How are they different?
I genuinely don't know. They are smart people, they know the game. They must have figured out that there is an upside for them.
Just speculating here. Maybe they have big customers that communicated to them that they will choose them over the competition if they open more of their software stack. Maybe they are less afraid of being sued because they are the underdog and they believe that the authorities would not be kind to a behemoth suing them, or they believe that the behemoth would not sue them because of the bad press it would generate vs the settlement they would get.
I actually don't know. But I am sure they have good reasons, just like the corps that keep it closed also have good reasons. It is not "good" vs "evil".
Thank you for your input - it genuinely is very interesting.
> If they open-sourced their userland drivers, and/or released their internal technical documentation, which would have a similar effect, they would open themselves to at least two big problems.
> First, they would spill all of their secret sauce for others to "draw inspiration" from. And there is a lot of expensive secret sauce in there. Not all of it is protected by patents either
Then why does Apple allow Asahi to reverse-engineer their low-level APIs? Are their lawyers not paying attention?
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 technical documentation, not even at the level of NVIDIA, much less something comparable with AMD and Intel.
>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.
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. :\
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.
Any foundry in Europe able or going to manufacture a chip with this on ?
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".
Pretty much all of graphics research now revolves around machine learning, whether it’s neural networks or something like Gaussian splatting.
You can always run with those features disabled if you find it less disappointing.
The headline seems like a way to set themselves up to fail. The advancements are great, those features are definitely something a lot of developers and players are interested in, but comparing it to desktop immediately just sets the bar too high.
The average desktop has an anaemic integrated GPU - so it sort of depends how one defines "desktop-class"
"Ai-NaTiVe"
"AI-native graphics" good lord, this just reeks of "we're doing something with AI, give us some of that money".
Fun thing, both SA and ChipsAndChesse similarly dunked on that presentation:
https://www.semiaccurate.com/2026/09/07/arms-2026-tech-day-i...
https://chipsandcheese.com/p/arms-c2-ultra-g2-ultra-nx-and-c...
Once one is over the odor of the headline, the announcement reads:
* Neural Super Sampling (NSS) reconstructs higher-resolution images from lower-resolution renders, reducing the amount of conventional rendering work needed to produce a high-quality final image.
* Neural Frame Rate Upscaling (NFRU) generates intermediate frames to increase frame rates and deliver smoother gameplay.
* Neural Super Sampling and Denoising (NSSD) combines neural upscaling with denoising to improve image quality in demanding ray-traced scenes with complex lighting and shadows.
So yeah, more uncanny valley garbage. Yuck.
So, image processing and no intelligence.
And nor is reconstructs true.
This should have happened years ago.
The worst opening ever "Demands from mobile users continue to grow.". I stopped reading bullshit
This is probably not about "mobile users" as in Android Phones & iPhone users .. but about the current (trendy) rise of handheld consoles ? Atleast that's what I think
Why is it bullshit ? Just because you dont need does not mean it’s not real
What's up with that? Younger generations don't even play on PCs anymore.
the switch, steam deck and competitors will likely look into ARM, the cheaper ones already use it, flagship and gaming phones, the latter is bigger than you think, normal mobile gaming is big in third world countries and china, and emulation continuous to get better, I can ToTK and The Witcher 3 (windows version) on my phone, and it's a few years old flagship