NVIDIA GeForce: Windows 10 vs. Ubuntu 15.04 Linux OpenGL Benchmarks

[eydee]

Not even close. What makes it a disaster? Got some examples or some sources on that?

How about the article you're commenting below? Check page 2, you'll immediately see how higher resolution gives a higher fps. (Same as fglrx scoring higher in Heaven with extreme tessellation compared to no tessellation at all...)

For other examples, start reading the Steam game hubs. Most of them are full of nvidia-W10 complaints.

[blackout23]

wccftech was able to identify that there are pretty significant performance regressions on Fermi cards in Windows 10. I'm sure it'll be fixed eventually.

Interesting to hear that. I also heard that there isn't a DirectX12 driver for Fermi, yet which they promised (DX12 for all DX11 cards). Hope my Fermi will get Vulkan support at least which it should be capable of in theory. I guess we'll hear about that next week. There seems to be an BOF Talk about Vulkan and one topic is called "

• Platform and hardware support: Members and community"

EDIT: Even better there is an NVIDIA Talk "Vulkan on NVIDIA GPUs" on August 9th.

https://www.khronos.org/news/events/2015-siggraph

[dungeon]

It's hard to believe that 4K shows the same performance as 1080p (on both platforms), and that 4K is not GPU-limited.

It is the same, because on both cases it actually runs at 4K ... 1080p is just false advertised.

I have that bug with Xonotic 0.8 (not with 0.7), where running benchmark or just using command line to set resoultion lower then native simply does not work

[johnc]

Not even close. What makes it a disaster? Got some examples or some sources on that?

wccftech was able to identify that there are pretty significant performance regressions on Fermi cards in Windows 10. I'm sure it'll be fixed eventually.

[johnc]

The fact the synthetics give equivalent results would tend to rule out driver side issues.

Again, as I said in the SoM thread, I firmly believe you're seeing a difference in the way threads, in particular, the thread that handles most of the GPU workload, is getting scheduled. If that thread ever gets bumped, you are going to lose significant performance. Windows biases high workload threads to prevent this from happening, Linux does not. Even if the thread gets bumped for 1ms, that's still 6.25% your render window.

Point being, a purely synthetic benchmark may run on just a few threads. A full fledged game, at least on windows, easily pushes 80, even if only a handful do meaningful work. But how those high workload threads get scheduled is of the upmost importance.

The benchmarks we typically see here though seem to indicate that games that are CPU-limited show a performance advantage on Linux. And games that are GPU-limited are about identical. SoM is almost out on its own it seems. (At least in terms of games benched here.) If it was a scheduling issue, wouldn't it show in other games as well?


The Xonotic results in this post seem to be off-kilter. It's hard to believe that 4K shows the same performance as 1080p (on both platforms), and that 4K is not GPU-limited.

[Dukenukemx]

I think he's talking about Windows update and Nvidia's latest downloadable driver conflicting.

[grenadecx]

People say the current state of nvidia win10 drivers is a disaster. Think of it like running fglrx on linux...

Not even close. What makes it a disaster? Got some examples or some sources on that?

[F i L]

something i like to see, cgo dx windows 10 cgo opengl ubuntu

CS:GO on Win10/DX would probably kill CS:GO on Linux/GL... not because Windows or DirectX are faster in general (they're not) but because CS:GO on Linux still has some really bad performance issues in my experience. I just got a new computer with pretty good specs (AMD FX, GTX960 4Gb, 16Gb Ram) and CS:GO has horrible frame-drops 'randomly' during gameplay every few seconds (I've tried everything to fix it, even found a bug-report, nothing works for me yet).. it makes the game completely unplayable on this machine which is just stupid..

And while I'm ranting... I love Chivalry and it plays OK on Linux (at least on this machine) but it still has these bad frame-rate issues even on low settings/resolution on some levels. Compared to my Win7 Computer (Phenom II X4, Radeon HD 7850 1Gb, 8Gb Ram) which runs the game butter smooth on the highest of settings @ 1080p... *sigh*.. I can't wait until Steam Machines are in full force and Linux X11/Wayland/Vulkan drivers & platforms are considered a priority target from more game/driver devs. I can't fault them for not spending time improving the situation for a platform with very little marketshare, but at the same time having a system which can easily handle these games, in terms of raw horsepower, but fails due to lack of interest from the developers is very frustrating...

[eydee]

Something is hilariously broken in your Xonotic test there. Windows scores are flat, increasing the resolution improves performance, and a 780 Ti does better than a 980 Ti...

People say the current state of nvidia win10 drivers is a disaster. Think of it like running fglrx on linux...

[gamerk2]

The fact the synthetics give equivalent results would tend to rule out driver side issues.

Again, as I said in the SoM thread, I firmly believe you're seeing a difference in the way threads, in particular, the thread that handles most of the GPU workload, is getting scheduled. If that thread ever gets bumped, you are going to lose significant performance. Windows biases high workload threads to prevent this from happening, Linux does not. Even if the thread gets bumped for 1ms, that's still 6.25% your render window.

Point being, a purely synthetic benchmark may run on just a few threads. A full fledged game, at least on windows, easily pushes 80, even if only a handful do meaningful work. But how those high workload threads get scheduled is of the upmost importance.