DiRT Rally 2.0 Linux NVIDIA vs. AMD Radeon 18-Way Steam Play Benchmarks

[Michael]

Is it a recent regression?

Within past two weeks I would guess

[hakzsam]

Within past two weeks I would guess

Does it work with -rc6? Could be a kernel regression too.

[Michael]

Does it work with -rc6? Could be a kernel regression too.

Can try later this week when back to my comps as away now, but I had asked Alex about it when the issue started coming up and he seemed to think it was Mesa issue.

[johanb]

It's very clear when the R9 Fury hits its 4GB HBM memory limit.
Equal performance as RX 590 on 1080p and 1440p on High settings, but as soon as it's set to Ultra the FPS gets halved. Surprisingly the performance is comparable to a 580 at 4K High settings though which was surprising, why is the VRAM usage more significant between High and Ultra than between 1080p and 4K?

[Brisse]

It's very clear when the R9 Fury hits its 4GB HBM memory limit.
Equal performance as RX 590 on 1080p and 1440p on High settings, but as soon as it's set to Ultra the FPS gets halved. Surprisingly the performance is comparable to a 580 at 4K High settings though which was surprising, why is the VRAM usage more significant between High and Ultra than between 1080p and 4K?

Yea, I have a Fury and I've seen the same happen in a few other games as well, especially when using DXVK which for some reason seem to need more VRAM than native games.

The reason for the big difference between high and ultra is probably that things like textures and shadow maps are at a higher resolution which needs a lot of VRAM. Going from 1080p to 2160p increases VRAM usage because the frame buffer contains more pixels but the difference it makes is not as dramatic as textures.

In practise, 4GiB is still fine. The Ultra presets tend to be a placebo mode anyway and you can barely see the difference in screenshots compared to high, and good luck spotting the difference when things are moving on screen and you're busy playing the game.

[M@yeulC]

It's very clear when the R9 Fury hits its 4GB HBM memory limit.
Equal performance as RX 590 on 1080p and 1440p on High settings, but as soon as it's set to Ultra the FPS gets halved. Surprisingly the performance is comparable to a 580 at 4K High settings though which was surprising, why is the VRAM usage more significant between High and Ultra than between 1080p and 4K?

As the above poster, I like my Fury a lot, though it could be cooler/less power hungry ^^"

I'd also say it comes from the textures (including normal maps, shadow maps, etc) and maybe the more complex geometry. The Fury shouldn't have much problems in pushing more pixels trough the pipe thanks to its high bandwidth memory. The tradeoffs are quite different from a 580. I have no idea how the individual stream processor performance compare, nor which one has more (though that would be easy to find out).

Hmm, I'm tempted to have a look at this Bios hack that could turn your Fury into a Fury X, but worrying a bit about the thermals of it?

[Cool Runnings]

How do you enable Codemasters CA to get MP working? And is it the same procedure on Dirt 4 (and earlier versions)?

[xpander]

How do you enable Codemasters CA to get MP working? And is it the same procedure on Dirt 4 (and earlier versions)?

copy https://www.upload.ee/files/9626680/...sters.pem.html file to /etc/ca-certificates/trust-source/anchors and run update-ca-trust
thats all i needed to do.

[Brisse]

As the above poster, I like my Fury a lot, though it could be cooler/less power hungry ^^"

I'd also say it comes from the textures (including normal maps, shadow maps, etc) and maybe the more complex geometry. The Fury shouldn't have much problems in pushing more pixels trough the pipe thanks to its high bandwidth memory. The tradeoffs are quite different from a 580. I have no idea how the individual stream processor performance compare, nor which one has more (though that would be easy to find out).

Hmm, I'm tempted to have a look at this Bios hack that could turn your Fury into a Fury X, but worrying a bit about the thermals of it?

RX 580: 2304 stream processors, or 36 compute units @ 1340MHz 6.174720 TFLOPs fp32
Fury: 3584 stream processors, or 56 compute units @ 1050MHz which is 7.526400 TFLOPs fp32

Performance in games doesn't scale perfectly with more compute units, similarly to the way they don't scale perfectly with more CPU cores, so the performance difference is smaller than the theoretical fp32 performance would suggest.

Memory bandwidth:
RX 580: 256 GB/s
R9 Fury: 512 GB/s

I did try to unlock the full 64 CU's on my Fury with a BIOS flash but it turned out to be locked in hardware, which many Fury cards are.

[theriddick]

Hmm the difference between Low and High/Ultra settings might be the tessellation effects? Because we see AMD doing great on Low but only matching or falling behind the 1080 at high/ultra. Makes you wonder where the Vega cards are getting hit the worse, thinking tessellation level since AMD has always had a issue with high values which is why NVIDIA used 64 or 128 levels of tessellation in their works suite whatever...