Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

NVIDIA and nouveau

lwn.net

Post reply on HN
Print view View original

31–39 of 39 posts

Re: NVIDIA and nouveau

#31

Post by belval »

collegeburner wrote: post #27
erwincoumans wrote: post #16

>> GPU may allow creating a cross-platform compute stack That was the intention of OpenCL.

except writing in opencl is a dogshit experience compared to cuda. khronos is incapable of producing an API that's even remotely pleasant to use, vulkan suffers from the same thing. it is one of the worst "design by committee" groups.

+1 to that. Nvidia isn't winning the AI accelerator race because it's making obviously superior products to AMD/Intel. It is winning because CUDA is easy to use and reason about.

OpenCL's API is terrible to use and even if you know all of it's kinks there is much more boilerplate which turns a small script into a tangled mess.

Re: NVIDIA and nouveau

#32

Post by xnzakg »

Probably in the minority here but I've recently had a better experience with nouveau than the proprietary nvidia driver on a laptop with the external video outputs wired to the nvidia gpu. The proprietary driver doesn't work well with sway, and recently nouveau got support for powering down the discrete gpu when not used, meaning I still get good battery life instead of it draining my battery ~4x as fast as with it disabled.

Re: NVIDIA and nouveau

#33

Post by wing-_-nuts »

Reading this only reinforces my conviction to go with an AMD gpu for my next PC build (my last build was ~ 20y ago with an ATI 9700). Nvidia doesn't seem to be a good citizen in the linux ecosystem.

Re: NVIDIA and nouveau

#34

Post by brokenmachine »

wing-_-nuts wrote: post #33

Reading this only reinforces my conviction to go with an AMD gpu for my next PC build (my last build was ~ 20y ago with an ATI 9700). Nvidia doesn't seem to be a good citizen in the linux ecosystem.

I recently did, it has been mostly a good transition.

The only thing I am slightly miffed about is I couldn't get recording working in HDR games. The colors come out completely ruined. This is on a windows pc.

Re: NVIDIA and nouveau

#35

Post by mmis1000 »

collegeburner wrote: post #27
erwincoumans wrote: post #16

>> GPU may allow creating a cross-platform compute stack That was the intention of OpenCL.

except writing in opencl is a dogshit experience compared to cuda. khronos is incapable of producing an API that's even remotely pleasant to use, vulkan suffers from the same thing. it is one of the worst "design by committee" groups.

I think vulkan is intentionally to be like that. It's by design a `not so abstract` abstraction that allow you to extract most power from the device while keep your most setup the same.(you still need to handle different vendor device specifically) It never mean to be a `write once and behaves the same on every vendor device` API.

That isn't what vulkan mean to do. If you want a experience like that, opengl is the way to go.(At the cost of performance)

Re: NVIDIA and nouveau

#36

Post by mmis1000 »

brokenmachine wrote: post #34
wing-_-nuts wrote: post #33

Reading this only reinforces my conviction to go with an AMD gpu for my next PC build (my last build was ~ 20y ago with an ATI 9700). Nvidia doesn't seem to be a good citizen in the linux ecosystem.

I recently did, it has been mostly a good transition. The only thing I am slightly miffed about is I couldn't get recording working in HDR games. The colors come out completely ruined. This is on a windows pc.

> I couldn't get recording working in HDR games

Use latest obs with Color space Rec2100 in advanced tab in main setting and Color space Rec2100 selected in game recording setting.

That works perfectly on my machine when trying to record forza horizon 5 with hdr enabled.

(The didn't work for streaming though, And it is probably not possible because rtmp didn't have support for it)

Re: NVIDIA and nouveau

#37

Post by dontlaugh »

mmis1000 wrote: post #35
collegeburner wrote: post #27

Earlier quoted context omitted.

except writing in opencl is a dogshit experience compared to cuda. khronos is incapable of producing an API that's even remotely pleasant to use, vulkan suffers from the same thing. it is one of the worst "design by committee" groups.

I think vulkan is intentionally to be like that. It's by design a `not so abstract` abstraction that allow you to extract most power from the device while keep your most setup the same.(you still need to handle different vendor device specifically) It never mean to be a `write once and behaves the same on every vendor device` API. That isn't what vulkan mean to do. If you want a experience like that, opengl is the wa…

The more ergonomic APIs with similar capabilities are Metal and it’s near-clone WebGPU.

Re: NVIDIA and nouveau

#38

Post by mort96 »

mmis1000 wrote: post #35
collegeburner wrote: post #27

Earlier quoted context omitted.

except writing in opencl is a dogshit experience compared to cuda. khronos is incapable of producing an API that's even remotely pleasant to use, vulkan suffers from the same thing. it is one of the worst "design by committee" groups.

I think vulkan is intentionally to be like that. It's by design a `not so abstract` abstraction that allow you to extract most power from the device while keep your most setup the same.(you still need to handle different vendor device specifically) It never mean to be a `write once and behaves the same on every vendor device` API. That isn't what vulkan mean to do. If you want a experience like that, opengl is the wa…

OpenGL is officially deprecated on all of Apple's platforms, and seems unofficially deprecated from Kronos; the last update to the standard was 4.6 in 2017. It will keep working (at least on most non-Apple platforms) for a long time still, but it's not going to necessarily give you access to everything your hardware can do.

That's not to say it's not the "way to go", but it's not in a great state.

Re: NVIDIA and nouveau

#39

Post by mmis1000 »

dontlaugh wrote: post #37
mmis1000 wrote: post #35

Earlier quoted context omitted.

I think vulkan is intentionally to be like that. It's by design a `not so abstract` abstraction that allow you to extract most power from the device while keep your most setup the same.(you still need to handle different vendor device specifically) It never mean to be a `write once and behaves the same on every vendor device` API. That isn't what vulkan mean to do. If you want a experience like that, opengl is the wa…

The more ergonomic APIs with similar capabilities are Metal and it’s near-clone WebGPU.

I think the simplicity of metal(and the apple version webgpu, now webmetal) relies on the fact that it only has to support apple hardware (With much less edge case need to handle).

I would be more curious about if the whatwg version webgpu can still be simple like that and has similar performance like the original one apple proposed.

Besides that, I think DX12 is a more proper competitor about access low level primitive from devices from different vendor.

Post reply on HN