New Patches Aim To Improve Smoothness & Latency Of NVIDIA On GNOME

[mdedetrich]

Yes, that's exactly what I meant. There are some, but the choice is grim.

Thats putting it diplomatically. Up until very recently, for the past few years there were basically no non-junk tier laptops that had AMD discrete graphics. Almost all laptops with discrete graphics that were mid tier (or higher) were NVidia.

[Paradigm Shifter]

the nvidia drivers are top notch, the amd drivers improve a lot in last years in linux, intel drivers improve a lot

I'm basically trapped using nVidia drivers as I require CUDA for just about everything I do. I've not really noticed much of an improvement from nVidia over the years, although I have noticed that various distros have made getting the drivers installed correctly without needing to perform strange sacrifices to the phase of the moon a lot easier. This is good, but I don't think it's so much an nVidia driven thing, as distros wanting to have people get a good out-of-the-box experience.

The one system I have (for e-mail/general office work) without an nVidia card has an Intel iGPU. It has only becomes worse as kernels have moved forward. I'm basically stuck on kernel 4.15 unless I want random soft GUI locks. It's sort of livable with 4.18, in that it happens once or twice a day. With kernels 5 and up, it's once or twice every few minutes.

That is not an improvement.

Admittedly I've not tried kernel 5.7 or the 5.8rc, so it's possible that it has finally been solved.

My AMD (more correctly, Radeon) experience on Linux is horribly out of date, so I can't really comment there beyond saying that I never really had any problems - even with an old FireGL 8800.

[royce]

The one system I have (for e-mail/general office work) without an nVidia card has an Intel iGPU. It has only becomes worse as kernels have moved forward. I'm basically stuck on kernel 4.15 unless I want random soft GUI locks. It's sort of livable with 4.18, in that it happens once or twice a day. With kernels 5 and up, it's once or twice every few minutes.

I would wager you have a hardware problem instead of dodgy intel drivers.

[timrichardson]

Getting off-topic here, but the users I've dealt with that own Intel/AMD hybrids verify that DRI_PRIME works (glxinfo checks out), but are disappointed with the performance of the AMD GPU. Ideas?
Note: I can't remember for sure, but I think the same thing might occur on AMD/AMD setups - dGPU performance isn't much better than iGPU.

Apple has been AMD-only in its most expensive laptops since 2016, and they sell those to people who need the GPU and pay for it. The new one launched today looks like it will be very powerful; but Apple has exclusivity on it for a while. The specs look pretty good.

[DanL]

Apple has been AMD-only in its most expensive laptops since 2016,

Either you meant to reply to someone else or I'm having trouble playing connect the dots.

[MrCooper]

Funny thing is, the patch actually didn't mention NVIDIA at all. It's strictly an improvement to the generic case where the low-level-provided presentation time is useless; it helps NVIDIA because NVIDIA blob falls under the generic (read: useless) case.

RTFA. It's not just Nvidia that benefits from the change(s).

The nvidia drivers are probably the only relevant ones, for non-virtual hardware, which don't provide the functionality needed for proper frame timing in the first place. This code in mutter is basically a band-aid for that omission.

(There are other surprising omissions, e.g. the nvidia drivers still seem incapable of saving and restoring VRAM contents across suspend/resume. Not what I'd expect from "top notch" drivers...)

[DanL]

The nvidia drivers are probably the only relevant ones, for non-virtual hardware

Ah, so you agree with me that the changes benefit more than Nvidia hardware, and that people should RTFA before crying/ranting about Nvidia?