Vulkan vs. OpenGL Linux Game CPU Core Scaling

[gamerk2]

Very interesting. You would expect Vulkan games to scale with more cores. I wonder if this is an limitation of radv right now.

Not really. All cases look like classic GPU bottlenecks once you get beyond 3 or so cores. Once you get beyond a certain minimum threshold, CPU power stops being relevant for games.

The reason Vulkan does well on two cores is because it is DESIGNED to use less CPU resources; two cores is less of a bottleneck compared to OpenGL. As core count/CPU power increase, OpenGL ends up being just as fast. This is EXPECTED behavior for both APIs.

[gamerk2]

How does testing proprietary crapware enrich any Linux experience?

How does ignoring the existence of 99.9% of the market enrich the Linux experience?

Sorry, but your argument goes both ways. By ignoring non-FOSS, you've outright ignored MANY glaring problems within Linux that never get addressed properly. And that's why for home usage Linux is relegated to also-ran status.

How about stopping the religious warfare against non-FOSS and instead address some of the problems that they expose? Just a thought.

[Michael]

For those wondering, NVIDIA scaling tests are coming up shortly....

[Veerappan]

Not really. All cases look like classic GPU bottlenecks once you get beyond 3 or so cores. Once you get beyond a certain minimum threshold, CPU power stops being relevant for games.

The reason Vulkan does well on two cores is because it is DESIGNED to use less CPU resources; two cores is less of a bottleneck compared to OpenGL. As core count/CPU power increase, OpenGL ends up being just as fast. This is EXPECTED behavior for both APIs.

But does that explain the tests where OpenGL caps out higher than vulkan? I agree with Michael here. Smells like a thread synchronization bottleneck to me

[Azpegath]

Michael has been spewing fabrications recently. Remember his post recently about how RadeonSI performance was on-par with Windows? As an RX 480 owner, let me tell you that's horse shit. I'm excited about AMD developments and want to hear about updates to Mesa, but he sells AMD drivers as far better than they actually are... they are definitely not good enough to do an OpenGL-Vulkan comparison on Linux.

Well, you're wrong there. They're always good enough to do a RADV (Vulkan) vs RadeonSI (OpenGL) comparison, and this is what this benchmark is.

[bug77]

For those wondering, NVIDIA scaling tests are coming up shortly....

Just as I suspected. TYVMM.

[Xicronic]

You might be one of those who play with DirectX on Windows without being aware of the difference or just use 10 year old drivers in a 5 year old distro version while Michael usually tests with the most recent git version.

Why so quick to discount me and my experience? Do you own an AMD card? Because I have been using the RX 480 since a month after its launch and I don't really like it.
Nvidia's Windows OpenGL performance is about as good as their DirectX performance. AMD's Windows OpenGL performance is worse than their DirectX performance. So because AMD has bad Windows OGL drivers to begin with, it makes their poor Linux OGL performance okay? You can attempt to rationalize this however you want, but effectually, AMD's Linux drivers deliver a fraction of their Windows performance. And no, I ran Mesa-git until I switched back to Nvidia last week.

And you would like to make sure that what you already know is still consistent to show the public? You think people read articles that for? Sounds like a North Korean practice.

... This is a generally accepted fact (and not the subject of this article). You're comparing me to North Koreans for saying Nvidia drivers are better than AMD's?

How does testing proprietary crapware enrich any Linux experience? Sometimes you might not be able to get around but our open source drivers do of course come in the first place and people like Michael are necessary to point out where to look for further improvements.

You might blame Nvidia for Michael not having anything reasonable to test.

Most people using Linux aren't on an open-source crusade, they're here because it's free. Good luck trying to keep pace with new hardware without proprietary UEFI firmware, processor microcode, and other blobs.

[Xicronic]

Well, you're wrong there. They're always good enough to do a RADV (Vulkan) vs RadeonSI (OpenGL) comparison, and this is what this benchmark is.

Then the article ought to be "Vulkan vs OpenGL AMD Linux Game CPU Core Scaling". As mentioned before, AMD drivers are not good enough to do a fair API-API comparison by themselves.

[bug77]

Then the article ought to be "Vulkan vs OpenGL AMD Linux Game CPU Core Scaling". As mentioned before, AMD drivers are not good enough to do a fair API-API comparison by themselves.

Even so, they're testing a common real-life setup, so they're still relevant. In a perfect world we'd had Nvidia results alongside AMD, but both for page hits and for not embarrassing AMD more than he has to, Michael doesn't do that.

[Shevchen]

... Okay? What does that have to do with what I said? My point is that Nvidia has solid OpenGL and Vulkan drivers to do an honest API-API comparison, and AMD does not. I never said anything about the benefits of open-source vs closed-source software.

As other people already pointed out, he does tests on Nvidia GPUs. The number of benchmarks on the AMD side are due to:

• More frequent updates on the drivers (meaning more frequent benchmarks to check them)
• More options to do different types of fine-analysis as all parameters are open for modification
• More than one driver to test and compare against

The Nvidia blob is known to be performing well - thats why its the baseline. The OS-Devs aim at it in order to get driver parity. Right now, we are pretty close to it (in some scenarios, we already reached parity) and this is why it is also interesting for the OS-community to look at. As of such, benchmarks for tuning the OS-drivers are very welcome.