OpenGL vs. Vulkan On The AMD Ryzen 3

[DarkFoss]

http://www.phoronix.com/scan.php?pag...0xt-vega&num=2

Yeah, the RX480 scored twice what the 1060 did in this run of Dota 2 in that test at the same resolution... and Ryzen 3 is still a reasonably fast single threaded CPU so that doesn't' really explain the difference.

The RX480 score is like 4-5x higher in that test...

In general the opensource Radeon driver should be alot faster than the Nvidia blob except for the very high end cards ie a RX480 still won't beat a 1080 or 1080ti.... despite having a better driver.

Kernel making a difference? Isn't Michael using the one he rolled in the test run you linked ? 4.12.0-vegaphx vs the 4.13.0-999-generic.

[linuxgamer72]

I am utterly surprised to see the AMDGPU/RadeonSI driver in combination with the RX 480 to deliver only a small fraction of the Nvidia performance.
I've been recently pleasantly surprised by how well an old HD7870(GCN1.0) card was doing on RadeonSI in various games, including Dying Light ("Parkour" mission), Rust, Alien Isolation and Metro if compared to an (of course generally faster) GTX 1060. Hence before reading the details of this benchmark test, I was expecting the RX 480 to blow the 1060 out of the water. Something must be wrong. A regression perhaps? Or maybe some in-game video settings the Mesa driver doesn't like?

[linuxgamer72]

Nothing surprising. Driver AMD has always been poor in terms of processor performance.
Very processor capasity. It's even on Windows.
Let Michael try to conduct a similar test on a powerful processor and such difference should not be (so on the old tests and seen)

Not in case of Mesa! This is a great driver stack, on its way to become one of the best performing drivers across all major operating systems (at least from my humble point of view - since two years watching its performance to improve week by week). Normally, the padoka-dev ppa delivers the best possible performance as it incorporates the very latest changes. However it is a development ppa, means non-stable. It can have bugs, regressions etc.. Just last week I had to move away from padoka-dev to stable because of dev was causing in-game crashes of the GPU (Pitcairn). I guess the extremely low performance of the RX480 in this comparison must have been caused by a regression or maybe some in-game video settings that Mesa/GCN4 don't like. With padoka-dev I also experienced in the past that I sometimes (and do to unknown reasons) had to restart a game to get full performance - or even reboot the system.

[mphuZ]

Not in case of Mesa!

Don`t kid youself. Drivers for Linux are doing almost the same team that under Windows. At least, skills and ideas about the same.
Overhead always been huge, so why it shouldn't be on Linux ?
And yes, there is no guarantee that this will be the main driver for all OS. At least, bridgman always leaves the question.

[M@GOid]

A couple weeks ago Michael did a benchmark with the Ryzen 3 and a RX 580, and the result in Dota 2 is the same as the GTX 1060 on this article:

http://www.phoronix.com/scan.php?page=article&item=amd-ryzen-3&num=8

So I believe we are seeing a big regression in Mesa or kernel here.

[rabcor]

get your shit together AMDGPU!

Don`t kid youself. Drivers for Linux are doing almost the same team that under Windows. At least, skills and ideas about the same.
Overhead always been huge, so why it shouldn't be on Linux ?
And yes, there is no guarantee that this will be the main driver for all OS. At least, bridgman always leaves the question.

Also, if you read a bit into it, avoidance of answering the question most likely means that it's possibly being discussed, but no decision has been made on it yet. I mean otherwise he could just say "there has been no discussion on that subject" or "there are no plans for it". I.e. he can neither confirm nor deny it. The alternative possiblity is that it has been decided to eventually move in that direction but he is not at liberty to tell us. Either way if he doesn't wanna answer it, no reason to keep asking.

[eydee]

Also worth noting is with this RX 480 vs. GTX 1060 comparison we are seeing much lower performance out of the RX 480 than we generally have been seeing in the past, at least on higher-end CPUs.

Because Mesa is not too efficient on the CPU side. When I switched from Phenom II to Ryzen 5, performance on my old 6850 increased by up to 400%. In old games, not the latest titles it couldn't run anyway.

[frosth]

Michael, "CPU Usage (Summary) Monitor" is for NVIDIA, AMD or both?

[bug77]

Michael, "CPU Usage (Summary) Monitor" is for NVIDIA, AMD or both?

Nvidia. The graph doesn't say, but it's written in one of the paragraphs near it.

[bridgman]

Don`t kid youself. Drivers for Linux are doing almost the same team that under Windows. At least, skills and ideas about the same.

No, the radeonsi (Mesa) team is completely independent of the Windows team - different people, different ideas.

If you are talking about the AMDGPU-PRO GL/Vulkan drivers then obviously it's not only the same team but the same drivers (other than the layer between common code and OS-specific kernel driver).

Overhead always been huge, so why it shouldn't be on Linux ?

Because it's a different driver, different people, different design, different ideas ?

And yes, there is no guarantee that this will be the main driver for all OS. At least, bridgman always leaves the question.

I have answered the question multiple times. We do look at using the Mesa driver for workstation from time to time and are looking at it again - but adding compatibility mode to Mesa would be a big pile of work and is not accepted as a good idea by much of the Mesa community. The closest thing to a workable solution I have seen so far is keeping the current -PRO GL driver for workstation and using Mesa for consumer Windows drivers, although that would require some changes in DOOM to work with core profile rather than compatibility profile. There are probably some other Windows OGL-only games which require compatibility profiles though, and most of them are not likely to change.

So a big honkin' pile of work for relatively little benefit. Hasn't made sense yet, as I have said multiple times already.