How Nouveau Compares To NVIDIA's Linux Driver When Kepler Re-Clocking Works

[smitty3268]

Michael what differences do you see in valley? I just tested locally with ultra settings + 8x msaa and I didn't see a difference at all, but for me nouveau was also much slower (11 fps vs 19 fps)

I think Marek has said the problem is due to having outdated drirc files present on the system.

[karolherbst]

Michael what differences do you see in valley? I just tested locally with ultra settings + 8x msaa and I didn't see a difference at all, but for me nouveau was also much slower (11 fps vs 19 fps)

[SystemCrasher]

Not relevant, unless you want to contribute and/or test. Or if you think installing a piece of software has something to do with asserting a certain philosophy.

Actually, it is. But it's not even about philosophy, just about being able to compute possible outomes more than 1 step further.

When Nvidia releases new driver and it turns out they used sharp axe to cut "extra" monitors support, to make it on par with windows, there is nothing you can do. You can rollback to old driver, but its a dead end. Over time, new kernels, Xorg, etc... would be released and your old version of driver would not work with these. So either you have to stick to ancient software forever, or you have to face unpleasant change. That's what we dislike about proprietary software. Not sure if this qualifies as phylosophy or not. IMHO its rather pragmatic approach. If opensource code faces some change I do not like, I can go and roll it back, or even patch my newer version of code to match old version behavior. So making "unpopular" or "nasty" changes happens to be hard and rather pointless thing to do. Somehow, recently some companies got idea its okay to treat customers like a shit, and nvidia is clearly one of such companies.

Then, even if you do not want to test, modern hardware and software are complicated and you can count there are going to be some bugs. Some of these are insignificant, but you can alsway encounter some pest which makes your life worse than it should be. In case of opensource driver it is rather pleasant and efficient process when you can report annoying crap and take some part in eliminating this pest (as much as your skill and knowledge permits). But in case proprietary software is just does not works. You just hit some dead end. And, uhm, replacing hardware after facing each and every bug is costly, and you'll go this way you'll end up ditching ALL hardware. So being able to report bug to adequate bug tracker and follow its fate is a really nice bonus.

The proprietary driver can emulate most things KMS does.

But it would have plenty of moron issues and bugs, and many things would not work. When KDB debugger halts whole kernel, it is really matters if you got it right, so GPU driver stays "alive enough". Else you can halt important part of driver, and guess what would follow. And it also sad if you've been hit by bug, kernel crashed and ... failed to switch mode to be able to display panic message, regardless of what X did before it, etc. Its sad when consoles working like a crap and full of bugs, which has been hallmark of nvidia for a while. Then it also suxx when you can't use recent kernel or X with some good new features.

Basically, opensource got its own advantages, and if you do not want these, why not just GTFO and use windows? Sure, MS would pwn you. Nvidia would pwn you. You'll be unable to deal with even most annoying bugs. And if you'll read EULA you'll get idea how exactly you're being treated. But you deserve such treatment for sure. Because you're dumb enough to be "agree" with it. So it happens.

And the one that don't work are purely cosmetic, won't break your workflow.

It really depends. Speaking for myself I do not want some nvidia faggots to decide what breaks my workflow and what is not. Whatever, I've got fed up with it in Windows ~ 8 years ago. This approach suxx and does not leads to anything good. And I do not need yet another Windows, thanks.

[smitty3268]

It is great seeing the progress being made on the OpenGL 4.x front, but 4.5 was released 1 year, 3 months and 16 days ago. I am afraid the open source drivers are chasing a moving target. Vulkan is about to be released is that something the open source drivers will be able to support relatively quickly?

I'd probably expect Vulkan support from Intel and AMD by the end of next year, which I think counts as relatively quick support. AMD is planning to opensource their proprietary vulkan code, and Intel is already working on support for their hardware, just behind the scenes right now. Not sure about nouveau - they'll probably be able to hook into some of the code being added by Intel and/or LLVM, but their schedule will be heavily driven by how much time their volunteer developers put into it.

[imirkin]

But, I spent major bucks on a GTX 980 because I needed to drive my 4K screen for complex graphics tasks, and it still looks like I could only expect from 15% to 25% of the performance I get with the binary driver. That's not enough to make it even a usable option, let alone something to get excited about. Sorry.

Actually you can expect about 0.01% of perf on GTX 980. There's no accel with nouveau for now on GM20x GPUs. They've locked down the firmware, it has to be signed in order to be uploaded.

[hubick]

I really appreciate the work these guys are putting into the Nouveau driver. It's a huge and complicated undertaking, and akin to doing it while blindfolded (no hardware specs). I run Linux because I value open source, and someday really hope to be able to run on the open driver.

But, I spent major bucks on a GTX 980 because I needed to drive my 4K screen for complex graphics tasks, and it still looks like I could only expect from 15% to 25% of the performance I get with the binary driver. That's not enough to make it even a usable option, let alone something to get excited about. Sorry.

[jf33]

DRI_PRIME is another reason for open-source drivers.

[pixo]

Hmm, LunarG and Valve are already working on a open source Vulkan driver for Intel hardware, right? Can that be used as a basis / starting off point for similar drivers for Nvidia and AMD hardware?

AMD is planning to opensource their Vulkan implementation after some time. It will be interesting to see how long it will take them.

[shmerl]

It is great seeing the progress being made on the OpenGL 4.x front, but 4.5 was released 1 year, 3 months and 16 days ago. I am afraid the open source drivers are chasing a moving target. Vulkan is about to be released is that something the open source drivers will be able to support relatively quickly? Will they even have to? I am not sure how those sorts of things work.

It's indeed somewhat of a wasted effort in the long term. Because once Vulkan will be out, I expect someone starting a new project for OpenGL implementation atop of Vulkan. That would basically render all those hardware specific OpenGL implementations obsolete. But it would take quite a while. At least it will be one uniform implementation that can be optimized going forward. It will be useful for legacy games (which are the vast majority now). Future games and applications will already use Vulkan.

[justmy2cents]

Thanks for the answers guys. The open source part I get 100% and I absolutely wish the open source drivers were 100% competitive with their proprietary counterparts but from the perspective of a gamer as low as 65% performance is not really acceptable, in a scenario with 30 FPS that puts you at 19.5 FPS with the open source driver, which is far from playable. Significant reductions in graphics quality isn't something that gamers want either.

It is great seeing the progress being made on the OpenGL 4.x front, but 4.5 was released 1 year, 3 months and 16 days ago. I am afraid the open source drivers are chasing a moving target. Vulkan is about to be released is that something the open source drivers will be able to support relatively quickly? Will they even have to? I am not sure how those sorts of things work.

Edit:

Hmm, LunarG and Valve are already working on a open source Vulkan driver for Intel hardware, right? Can that be used as a basis / starting off point for similar drivers for Nvidia and AMD hardware?

you have to take it with other perception. a lot of us are not just gamers and integration of OSS drivers means much more than meager 35%. it is easy to buy one rank better card than one otherwise would, except right now NVidia makes this logic hard because they don't provide firmware they promised

as far as chasing moving target, with Vulkan there will probably be much less problem than with OpenGL. for Vulkan there will be at least one (Intel) OSS full implementation they can follow and quite a lot of parts are not in driver anymore. for OpenGL there is no such thing and has to be reinvented part by part