NVIDIA Proposing New Linux API For Dynamic Mux Switching With Modern Dual-GPU Laptops

[phoronix]

Phoronix: NVIDIA Proposing New Linux API For Dynamic Mux Switching With Modern Dual-GPU Laptops

While the VGA_Switcheroo has long been part of the Linux kernel for laptops with hybrid (dual GPU) graphics for switching between the GPUs on platforms with a hardware mux switch, this current API has been found to be ineffective for the latest laptops like those with "NVIDIA Advanced Optimus" support. Thus NVIDIA is working on and proposing a new Linux user-space API around dynamic mux switching...

https://www.phoronix.com/news/NVIDIA-Dynamic-Mux-Switching

[mdedetrich]

Considering that NVidia has huge experience when it comes to dual GPU laptops along with Mux switching, as long as this proposal is generic enough (which appears to be the case since it is in user space) so that AMD can also use it, I see nothing ill of this (in fact its fantastic news as this has historically been a massive sore spot for Linux).

[karolherbst]

Considering that NVidia has huge experience when it comes to dual GPU laptops along with Mux switching, as long as this proposal is generic enough (which appears to be the case since it is in user space) so that AMD can also use it, I see nothing ill of this (in fact its fantastic news as this has historically been a massive sore spot for Linux).

Yeah, it's a very good proposal and I'm glad Nvidia finally made that public. We really need an improved infrastructure when it comes to multi GPU setups. But what's more work is to get compositors to deal with it and generally move to a "composite on the GPU used for displaying" model, even if it means having one rendering context per GPU and stuff.

[lumks]

Without looking at it (company network blocks it for some reason) I hope this also works for

1. Desktop
2. Systems with >3 GPU
3. External graphicscards

Other than that. Yes please do something about this situation. currently it's just baaad.

[WereCatf]

Without looking at it (company network blocks it for some reason) I hope this also works for

1. Desktop
2. Systems with >3 GPU
3. External graphicscards

Other than that. Yes please do something about this situation. currently it's just baaad.

That's not what the proposal is about: you don't have a MUX on a desktop, all those GPUs are fully independent and their outputs are not connected. A MUX is a specific hardware-component to which both the iGPU and the dGPU are connected and the MUX is then connected to the display -- an entirely different scenario.

[schmidtbag]

Considering that NVidia has huge experience when it comes to dual GPU laptops along with Mux switching, as long as this proposal is generic enough (which appears to be the case since it is in user space) so that AMD can also use it, I see nothing ill of this (in fact its fantastic news as this has historically been a massive sore spot for Linux).

Agreed, but if this ends up being something that only works for Nvidia's drivers then this is yet another step in the wrong direction. It'll be interesting to hear more.

[mdedetrich]

Agreed, but if this ends up being something that only works for Nvidia's drivers then this is yet another step in the wrong direction. It'll be interesting to hear more.

Its not going to get anywhere if this is the case, NVidia is proposing adding a userspace API to DRM-KMS so if the changes are NVidia specific then it won't get accepted.

[mdedetrich]

That's not what the proposal is about: you don't have a MUX on a desktop, all those GPUs are fully independent and their outputs are not connected. A MUX is a specific hardware-component to which both the iGPU and the dGPU are connected and the MUX is then connected to the display -- an entirely different scenario.

Indeed, high end dual GPU laptops tend to have a MUX because without you get a performance penalty when running off one of the GPU's due to have to route the display from GPU that is NOT connected to the display via the GPU that IS connected to the display. I think that LTT did a video on this and its roughly a ~10% performance penalty.

[darkbasic]

Indeed, high end dual GPU laptops tend to have a MUX because without you get a performance penalty when running off one of the GPU's due to have to route the display from GPU that is NOT connected to the display via the GPU that IS connected to the display. I think that LTT did a video on this and its roughly a ~10% performance penalty.

Yet mesa developers said it was just because of a bad implementation and you could theoretically achieve way less penalty without all the disadvantages of a MUX.

[openminded]

For me, is too late. I've been waiting for something like this for too long. No more Nvidia GPUs in my laptops.