The Current NVIDIA vs. AMD Radeon Linux Gaming Performance Ahead Of Ampere

[gfunk]

Is there any chance we'll get performance per dollar graphs in the future as well?

Agreed, all these test clearly say "nVidia cards are better" but then you look at prices and see the RX 5700 XT is really good value for money

[JMB9]

My countless GPU driver-related system freezes over the years beg to differ, and I'm sure Navi owners (especially early adopters) also have something to say on that. Haven't had a single crash with NVIDIA though with Maxwell and Pascal.

you cant say that the ppl in this forum will hate you

Well, I would not chose the word hate - it may just be to have different experiences, though.

I have a Navi 10 (aka RX 5700 XT) and you should not use a kernel below 5.6 or Mesa below 20.0 {currently I am using: 5.7.19 & 20.1.5} - that's for sure. But hey, it is 1st gen. RDNA - so there had been much more work than one have to expect now for RDNA2 and RDNA3. But thus the 1st stable distro for me will be Kubuntu 20.10 STS 'Groovy Gorilla' (waiting for Mesa 20.2 and X.org 1.20.9 to arrive before installing it) - while Navi 10 was released 07/2019 ... so there is a lot to improve - especially AMD should be concentrating much more on mainline (Linux & Mesa) as that is what counts for GNU/Linux distros.

When starting to look for an AMD system the many reported problems with crashes (browser/gaming) was not nice ... and I prepared to experience instabilities.
But currently everything works perfectly - using the system more than 10 hr per day without any problem (OK, 2 crashes with a sound echo and sudden reboot ... but after more than two months incl. testing a lot of games ... so for me this is OK for a brand new system).

And even not wanting to giving the impression Nvidia has more problems - but I have experienced problems with games of a fresh Nvidia card on a friend's PC while both of my systems (old Intel Haswell iGPU and the new AMD Navi 10) just works as expected. So the idea of Nvidia being always more stable and superior is even now a lie.
Nvidia may have an advantage for those not riding mainline (which AMD users may be forced to at the beginning - if wanting to have a fresh card at least - which was a pleasant ride for me via Ubuntu PPA - this is no problem for typical Linux users - no compiling etc. necessary). But if something requires following mainline you are much better off with AMD - right now. Nvidia is no good choice if one really uses Linux professionally (not in terms of making money but getting most out of the system).

We will see what is in reach for AMD with RDNA2 and RDNA3 ... and I personally still hope for an RDNA APU (for > 40W desktop CPU, no laptop crap) fully capable of 8k in near future, as efficiency is important (and AMD is currently equal to Nvidia concerning that - and I don't even need the power of Navi 10 ... was just the best supported card for my current and future use cases) - and efficiency will get even more important in future.

[gardotd426]

A separate article next week has RADV vs. AMDVLK vs. AMDGPU-PRO metrics... But long story short, RADV won by far the most (or I think all, don't have the results in front of me at the moment)...

That's not what I meant (or what I said, really).

I'm well aware that you also do RADV vs. AMDVLK vs. Vulkan-AMDGPU-PRO benchmarks. That has nothing to do with this. What I'M saying, is that when doing AMD vs. NVIDIA, for each game, the AMD result should be whichever one performed the best. Like, with my 5700 XT, in Hitman 2, I get better performance with vulkan-amdgpu-pro than I do with RADV+ACO. So, in this article you just posted, I would have had the Hitman 2 result be the vulkan-amdgpu-pro one.

The point is that on Linux we actually have three choices for Vulkan driver, and they can all coexist and be chosen on a per-game basis, depending on which one performs better. Your benchmark choices are honestly pretty flat-out terrible, so they don't get too much of a picture, but in reality there are actually all sorts of games that either won't run with RADV, or run much worse on RADV compared to AMDVLK and/or vulkan-amdgpu-pro. Doom Eternal is one example, RADV is about 30% worse. MK11 is another example, RADV can't even run the game, you have to use AMDVLK or -pro.

But the thing is, AMD users on Linux actually have the benefit of three drivers specifically for instances where a game runs much better with one over the others.

And if you actually used the BEST PERFORMING driver result for each game when doing your AMD VS NVIDIA results, I think that would definitely make a difference, and either way, it's a much more accurate representation of the real situation, considering the fact that we don't just have RADV (and RADV is objectively not the best all the time).

I'm not sure how you read what I wrote and thought I was saying "hey do an article for RADV vs AMDVLK vs vulkan-amdgpu-pro for the 30th time!" when that's not even remotely what I said.

[theriddick]

What's wrong with that RTX 2070? it should be better?

I wish we had a open-source way to benchmark ray tracing performance (obviously NVIDIA would win assuming benchmark capitalizes on RTX feature).

once navi2 comes around it will definitely be a interesting metric to measure on top of the general polygon/raster performance.

[onlyLinuxLuvUBack]

That's not what I meant (or what I said, really).

I'm well aware that you also do RADV vs. AMDVLK vs. Vulkan-AMDGPU-PRO benchmarks. That has nothing to do with this. What I'M saying, is that when doing AMD vs. NVIDIA, for each game, the AMD result should be whichever one performed the best. Like, with my 5700 XT, in Hitman 2, I get better performance with vulkan-amdgpu-pro than I do with RADV+ACO. So, in this article you just posted, I would have had the Hitman 2 result be the vulkan-amdgpu-pro one.

The point is that on Linux we actually have three choices for Vulkan driver, and they can all coexist and be chosen on a per-game basis, depending on which one performs better. Your benchmark choices are honestly pretty flat-out terrible, so they don't get too much of a picture, but in reality there are actually all sorts of games that either won't run with RADV, or run much worse on RADV compared to AMDVLK and/or vulkan-amdgpu-pro. Doom Eternal is one example, RADV is about 30% worse. MK11 is another example, RADV can't even run the game, you have to use AMDVLK or -pro.

But the thing is, AMD users on Linux actually have the benefit of three drivers specifically for instances where a game runs much better with one over the others.

And if you actually used the BEST PERFORMING driver result for each game when doing your AMD VS NVIDIA results, I think that would definitely make a difference, and either way, it's a much more accurate representation of the real situation, considering the fact that we don't just have RADV (and RADV is objectively not the best all the time).

I'm not sure how you read what I wrote and thought I was saying "hey do an article for RADV vs AMDVLK vs vulkan-amdgpu-pro for the 30th time!" when that's not even remotely what I said.

Is it possible to run zink too ?

./benchmark_go --set_renderer=zink
./benchmark_go --set_renderer=radv
./benchmark_go --set_renderer=amdvlk
./benchmark_go --set_renderer=vulkan-AMDGPU-PRO

I mean it's already setup with the pcie insert slot, so just a bit more run time ?

[smitty3268]

Is it possible to run zink too ?

Why are people so obsessed with Zink? I don't get it.

[oleid]

Um... Are you serious?

First of all, dude was talking about Linux. Not Windows. BSODs and Windows in general are irrelevant.

Also, AMD is NOTORIOUS for its driver issues, especially with Navi. It was literally so bad that reviewers temporarily stopped recommending AMD precisely because their drivers were so bad, and it was actually a problem on BOTH Linux and Windows.

I guess what our pal was trying to say in a trollish way is that it's not all roses in team green either. Usually things work as expected, but you might be among the unluckily ones where it doesn't. I recall pretty nasty problems when working as student system administrator in our Linux desktop cluster. You put it as if there are no problems whatsoever.

[gardotd426]

You put it as if there are no problems whatsoever.

No I didn't.

I said Nvidia isn't notorious/known for shitty drivers like AMD is.

Also,

I recall pretty nasty problems when working as student system administrator in our Linux desktop cluster.

The Linux graphics driver situation is literally unrecognizable compared to what it was 3-4 years ago and beyond, for both AMD and Nvidia. That experience is irrelevant for today, unless your days as a student were < 2-3 years ago.

Also, as I said already, I literally own two Navi GPUs (and have owned 4 AMD GPUs in total). I'm well aware of how bad the AMD drivers are. It's great that they're open source, but that absolutely, objectively does not make them automatically good. They're not that good. They're just viewed as amazing because they're open source, and we're so used to having godawful, completely shit open source graphics drivers.

[oleid]

About the same number of people buy the top 3 Nvidia GPUs as buy the bottom 9. And the 2070 Super is the most popular.

To be fair, this is only true if you only consider Nvidia isolated from the rest of the world. If you include AMD, the picture is different. AMD is a lot stronger in low end and mid range than Nvidia is. Which means, if there is a choice, people prefer AMD (probably due to fps/price). Thus, for Nvidia customers, low and mid range is underrepresented.

You said (my interpretation) half of nvidia customers buy high end. Considering AMD has no high end and about 50% of the sales, that would mean:
low/midrange is about 75%, high-end 25% if you consider the whole market (which is the only reasonable thing to do).

[oleid]

No I didn't.

You're right, it wasn't you our pal responded to, I'm sorry.

Also, as I said already, I literally own two Navi GPUs (and have owned 4 AMD GPUs in total). I'm well aware of how bad the AMD drivers are. It's great that they're open source, but that absolutely, objectively does not make them automatically good. They're not that good. They're just viewed as amazing because they're open source, and we're so used to having godawful, completely shit open source graphics drivers.

What do you do with that many GPUs?