[phoronix.com] NVIDIA vs. Radeon Vulkan & OpenGL Performance With A Celeron, Pentium & Core i7

[Muhammed]

5 games tested in both Vulkan and OpenGL modes, on both the GTX 1060 and RX 580 and with 3 differenct CPUs (Celeron G, Pentium G, and Core i7 7700K)

Results:
1-NVIDIA is generally faster in both OpenGL and Vulkan
2-NVIDIA's lead is extended anywhere between 20~50%, depending on the game
3-NVIDIA's CPU overhead is smaller in both Vulkan and OpenGL, GTX 1060 had better performance under Celeron G than the RX 580.

https://www.phoronix.com/scan.php?page=article&item=kblcpus-gl-vlk&num=2

Use proper article citation format when linking to content at other sites (and stop deliberately provoking confrontation with unnecessary rhetoric).
-- stahlhart

 

[whm1974]

I'm not surprised, after all Nvidia does have a solid lead already in Linux graphics performance and has had it since the early 2000's.

 

[Guru]

Nothing new here, Nvidia has more resources and thus more people to optimize for Linux.

 

[EXCellR8]

Linux? you mean that free OS that six people use for gaming?

 

[Spjut]

That's just AMD's open source driver though?

 

[Krteq]

Yep, the used MESA Dev free/open-source driver. Anyway, AMDGPU-Pro driver seems to get same perf gains every month. I'm quite happy to see this kind of progress from AMD side.

nV always dominated on Linux under OpenGL. AMD's Vulkan support on Linux is still in beta phase, so let's wait for final release.

 

[whm1974]

Linux? you mean that free OS that six people use for gaming?

You are thinking of TrueOS.

 

[Elixer]

5 games tested in both Vulkan and OpenGL modes, on both the GTX 1060 and RX 580 and with 3 differenct CPUs (Celeron G, Pentium G, and Core i7 7700K)

Results:
1-NVIDIA is generally faster in both OpenGL and Vulkan
2-NVIDIA is so fast in Vulkan, it's lead is extended anywhere between 20~50%, depending on the game
3-NVIDIA's CPU overhead is smaller in both Vulkan and OpenGL, GTX 1060 had better performance under the castrated Celeron G, than the RX 580.


https://www.phoronix.com/scan.php?page=article&item=kblcpus-gl-vlk&num=2

That was a very misleading test, but, that is typical for that site.

The first issue is, they were NOT using Nouveau vs Mesa to make it a bit more fair, though, even that isn't entirely fair for other reasons.

They used Nvidia's binary blob, but didn't use AMD's binary blob.
So, overall, this only shows that Nvidia's binary blob is very well optimized for their hardware, and the open sourced Mesa drivers needs lots of work. As if that needed to be proved.

 

[TheELF]

That was a very misleading test, but, that is typical for that site.

The first issue is, they were NOT using Nouveau vs Mesa to make it a bit more fair, though, even that isn't entirely fair for other reasons.

They used Nvidia's binary blob, but didn't use AMD's binary blob.
So, overall, this only shows that Nvidia's binary blob is very well optimized for their hardware, and the open sourced Mesa drivers needs lots of work. As if that needed to be proved.

Errr,would a typical linux user know these things and be able to set them up properly?
Honest question, sounds very technical, I know about computers at least windows but I didn't understand a word of what you said.
If what they tested is the default software that will be installed without further knowledge then I don't think that it's miss leading.

 

[Elixer]

Errr,would a typical linux user know these things and be able to set them up properly?
Honest question, sounds very technical, I know about computers at least windows but I didn't understand a word of what you said.
If what they tested is the default software that will be installed without further knowledge then I don't think that it's miss leading.

Typical linux user? I would assume they know the difference between what I said.

Let me explain it this way, Nvidia has proprietary drivers (like on windows) where you basically install the binary, and be off.
AMD also has this option as well.

There are other drivers that are open sourced, where people contribute to them to make either nvidia (Nouveau) or AMD (Mesa) or Intel (Mesa) work.

The ones made by Nvidia & AMD (The binary blobs) are usually significantly faster, since they have access to the full tech specs, and they can optimize much, much better than the people working on the open sourced ones where there isn't much tech specs, and they end up reverse engineering to make things work.

So, it is up to the distro in question whether they pick the default drivers for the cards as open sourced (which most do) or the proprietary binary blobs (which some ask if you wish to install).
Then the user themselves can install either version as well.

Hope that clears it up.

 

[Muhammed]

They used Nvidia's binary blob, but didn't use AMD's binary blob.

They used the faster option for both. No foul play here. And AMD's original binary blob are total trash, with crashes, performance drops and bugs. Mesa is updated constantly and is the best option for AMD cards on Linux.

Also Phoronix is a highly respected site, it's literally the only technology source for high end CPUs, GPUs and games comparisons on Linux.

 

[lobz]

They used the faster option for both. No foul play here. And AMD's original binary blob are total trash, with crashes, performance drops and bugs. Mesa is updated constantly and is the best option for AMD cards on Linux.

Also Phoronix is a highly respected site, it's literally the only technology source for high end CPUs, GPUs and games comparisons on Linux.

Then go ahead and respect them all you want. The test is a total waste of time, as AMD doesn't even have proper Vulkan support for linux, how about spare hours of testing and some brain cells of the poor chaps reading it and test it when AMD actually releases their Vulkan driver for Linux?

 

[Elixer]

They used the faster option for both. No foul play here. And AMD's original binary blob are total trash, with crashes, performance drops and bugs. Mesa is updated constantly and is the best option for AMD cards on Linux.

Also Phoronix is a highly respected site, it's literally the only technology source for high end CPUs, GPUs and games comparisons on Linux.

I *am* using AMD's binary blob on my linux box, and haven't run into any issue yet, but, I have had issues with Mesa. It all depends on what you are doing.

As for being 'highly respected', not in my book. They are more tabloid style than anything else, with the occasional good post.

 

[Muhammed]

I *am* using AMD's binary blob on my linux box, and haven't run into any issue yet, but, I have had issues with Mesa. It all depends on what you are doing.

Maybe I can direct you here:

The open-source Radeon Linux driver is much faster now than where it was last year and is becoming competitive with NVIDIA's Linux driver performance.

Overall, the open-source Radeon driver stack is finally closing in on the Windows 10 Radeon Software performance for being able to deliver roughly the same performance potential on both platforms... Of course, NVIDIA's proprietary driver stack has been capable of doing the same features and performance across Windows / BSD / Solaris / Linux for as long as I can possibly remember, but Radeon hasn't been in the same boat. It wasn't too far back that AMD developers were hoping to just get 80~90% of the performance of their Windows/proprietary drivers. The Radeon performance for Linux vs. Windows is looking good now with the newest code, but there are also features lacking like a number of display-related features being blocked by DC/DAL, there not yet being any Radeon Software Settings for Linux, etc. But hopefully more of these features will be to parity in the not too distant future.

http://www.phoronix.com/scan.php?page=article&item=radeonsi-win-catching&num=1

 

[Hitman928]

The reason the AMD binary drivers weren't used is because they are only officially supported on LTS releases. The last LTS release for Ubuntu was 16.04 but they had 17.04 installed. They didn't want to have to roll back to 16.04 for every test so they just used the open source drivers.

 

[whm1974]

The reason the AMD binary drivers weren't used is because they are only officially supported on LTS releases. The last LTS release for Ubuntu was 16.04 but they had 17.04 installed. They didn't want to have to roll back to 16.04 for every test so they just used the open source drivers.

That does make sense and besides they should use the drivers that work the best and offers the best performance when test anyway. After all Linux users will use what works the best.

I use the close source Nvidia driver for this reason. If I was using an AMD GPU, then I would use the open source driver.

Thank you @stahlhart for opening this thread back up.

 

[richaron]

Errr,would a typical linux user know these things and be able to set them up properly?
Honest question, sounds very technical, I know about computers at least windows but I didn't understand a word of what you said.
If what they tested is the default software that will be installed without further knowledge then I don't think that it's miss leading.

Every Linux user will be aware of these issues. It comes down to understanding the open source ethos and the "teething issues" people have when switching over. For example it vexed me greatly in my early days that I couldn't just play mp3s on a fresh linux install, and that the bundled graphics drivers were crappy compared to the alternative. But a little searching around taught me easy lessons about what could (and could not) be included in a standard Linux distribution because of licensing.

Once you've done a tiny bit of Linux research it's also very apparent there's a large base of people who for idealistic or technical reasons only want open source software (i.e. no proprietary software or binary "blobs"), and AMD has really been trying to win friends in this area over the past couple of years. But iirc the AMDGPU drivers have been the main beneficiaries of these open source efforts, and these weren't the drivers tested.

Maybe I can direct you here:
http://www.phoronix.com/scan.php?page=article&item=radeonsi-win-catching&num=1

^It appears you still don't understand why this is an apples to oranges comparison. It's a binary blob vs one of the open source options. Sure there may be reasons why, and it may in fact be indicative of other comparisons, but nothing you linked to makes this comparison any more valid.

 

[richaron]

That does make sense and besides they should use the drivers that work the best and offers the best performance when test anyway.

Perhaps they should, but they didn't. Did they?

 

[whm1974]

Perhaps they should, but they didn't. Did they?

I will have to read the article, but I'm sure he did. But he usually uses drivers that work the best over performance.

 

[DidelisDiskas]

Linux? you mean that free OS that six people use for gaming?

I am one of those 6 people! The odds that we would end up in the same discussion forum.

 

[whm1974]

I am one of those 6 people! The odds that we would end up in the same discussion forum.

As is am I. I just beat Civ: Beyond Earth: Rising Tide with a Domination victory.

 

[NTMBK]

The best way to boost gaming performance on a Linux PC is to install Windows.

 

[Grep_Linux]

What NTMBK said is pretty much true.

There should be a prerequisite that you use "linux" before commenting. Unlike with windows, Anything can be different in linux, kernel version, window manager, shell, versions of anything, I've always got different results from Michael as probably you would too with the same hardware just from having different software somewhere... Nvidia users should probably use the open source driver more as it would spur development but that isn't the case since most are happy with the blobs performance. Most AMD users will opt for the open sauce flavor. Apples to oranges test as others have pointed out, using RADV for vulkan tests is like using tiny c vs ICC. However since it's the most used for both it's a valid article just meaningless is many ways. Like testing Linux games vs Windows games in general. If you're using Linux there's hopefully a reason (I hate windows isn't a reason, no one likes windows) and many of us concede certain things to have the OS we want, frame rates be damned.

I game on Linux, performance is fine, lower than windows, I use open source driver, makes my machine a high premium for lower performance but that's because I want the OS I want.