NVIDIA 375.10 vs. Linux 4.8 + Mesa 13.1-dev AMD GPU Benchmarks

[dungeon]

exactly :P
the count then is: 5x4k vs 8x1080p

So exactly you ingored there GPUtest results for some reason and didn't bother answering why

[Shevchen]

Wow, this topic is hot...

The OS-Vulkan driver is early and will see improvements. Other than that, there is only one game out there with a decent Vulkan implementation, and this is DOOM. On Windows benchmarks, we see exactly the behavior we expect: AMDs is beating NVidia via the pure horsepower-difference. The RX 480 is battling the GTX 1070 there.

Now, a single game is not really worth a benchmark-suite and we will see a lot of games being released which are not on the DOOM-level of Vulkan implementation (also, its good but not finished, even DOOM has some reserves in its engine code to be further optimized)

Once more Vulkan games are out, we will have:
- a better OS-Vulkan driver
- Nvidia will have a new architecture that *will* be capable of async-compute and all the nice stuff Vulkan offers
- SQ42 of Star Citizen will be out

I'm talking like about a 3/4 year from now on. This is the rough time-frame the OS-drivers have time to get close (like not more than 10% difference in raw-power) to the CS-driver performance, because this is the time where the curtain will fall. The results in this benchmark are still promising. We get those results or even better ones, when we don't hit a bottleneck and the next step is to "simply" solve those bottlenecks.

[cj.wijtmans]

1060 looks like a great card for the money.

all the 10x0 cards here are 200 euro + insane prices. (even1050)

[bridgman]

So the real problem here is: We need more demanding benchmarks / AAA games on linux to test.

... and in the meantime run those games at higher resolutions where possible.

Metro LL does have very low frame-rates at 4K especially on the lower-end cards. For BioShock that is one of the games I need to re-check, it's either BioShock or DiRT Showdown where I run into problems at 4K where sometimes the game engine dynamically resets to 1080p, I think that might be DiRT though thinking more about it.

Thanks Michael.

[Shevchen]

We know that Star Citizen will be build on Vulkan and it will be released on Linux. I'm pretty sure, that this will be the ultimate benchmark we might see and to test drivers against. (As they are using every core on the CPU and not just stop at 4 and will also do a good chuck on physics for the GPU to perform)

[dungeon]

... and in the meantime run those games at higher resolutions where possible.

I know why you want it higher preferably , but really if inteded to the most readers then benhmarking should be done at resolution most people use.

[dungeon]

But yeah more demainding AAA games make sense to test on higher resolution possibile... again only on High End GPUs, as only there that make perfect sense

[duby229]

Ohh stop.

You people won't be satisfied until the colors on those graphs are completely reversed, then you'll be praising how representative of reality these benchmarks are.

For April 1 Michael should run the same benchmarks but switch the video card labels and colors. The forum thread would be hilarious.

The truth has always been that if you want open drivers on a discrete GPU you need to buy AMD. But you will have to accept a penalty in terms of performance and features. If you're willing to do that, fine. But don't lie to yourself and everyone else here and pretend that sacrifice isn't being made.

I have absolutely no idea what you are trying to portray? Colors? I don't understand.

And the truth has always been just as much that if you want to buy nVidia hardware you must accept a penalty in terms of visual fidelity and quality of the rendered output. What you ignore is the issue of the "right" performance vs the "most" performance. You continue making that mistake and you will continue being wrong.

[Xen0sys]

even nvidia results are kind of cpu bound at the tested resolution and the 1080/70 get slapped by the 980/70.
All that matters for such a game is 60fps (its not counter strike).
So the real problem here is: We need more demanding benchmarks / AAA games on linux to test.

There's still some noticeable scaling whether newer cards are well optimized yet or not.

Yes all that matters when playing the games arguably, is about 30+fps but 60+fps preferably. I regularly play games in the 30-60fps range (avg 45fps) simply because that's the best my card can do (for example with ARK). I never made any claims that the performance itself was bad for the game (B:I or otherwise). I also never claimed I needed lots of fps.

Even if games could go >60fps I still keep them down to 60fps with V-Sync because that is just more efficient usage of the GPU. V-Sync on/off could be the difference between >100 Watts at the wall. I track all my PC's Watt usage.

That being said, your completely right but that is besides the point I am making which is that the performance/$ is not level and not where it should be. That matters for justifying a card purchase and that's about it - has nothing to do with playing games.

That argument is weakened by the fact that many games scale well and have a relatively low fps drop going from 1080p>4k (AMD) for example. It's certainly peculiar but not nearly as damning as I made it sound initially. I was also operating under the assumption that the game was Linux-ready ~3 years ago instead of the 1 or so year ago that it really was.

The secondary problem is as you described in that there needs to be a larger pool of demanding games in general to spur further development efforts.

[johnc]

And the truth has always been just as much that if you want to buy nVidia hardware you must accept a penalty in terms of visual fidelity and quality of the rendered output. What you ignore is the issue of the "right" performance vs the "most" performance. You continue making that mistake and you will continue being wrong.

There is no "penalty" in "visual fidelity" and "quality of rendered output".

Try making up some other excuse.