14-Way NVIDIA GPU Comparison With Civilization VI On Linux

[dungeon]

The fact that it's running at almost precisely 60 FPS when not GPU bound leads me to believe that the game is not actually CPU bound, it's instead been limited to 60 FPS on purpose by the developers.

A quick look at Windows benchmarks, where this limit is not present, would suggest this limit is Aspyr's handy work and not Firaxis'.

Just ask Michael to run bench on some lower resolutions like 720p, if it is same @60 fps bound there that way your cap theory might be proven as true.

But i think that is not the case, as here GoL did it on 800p also to not much difference.... his results with 980Ti even on minimum are even much less then Michael's 60 fps

https://www.gamingonlinux.com/articl...rt-report.9075

And only thing obviosly different is CPU model, looks exactly the same difference as if someone compare i7 7700K vs i7 5960X singlethread IPC... Michael's CPU is about ~25% faster on that

Not sure how we can prove 60 fps cap as no one have that fast CPU yet, OC maybe?

[L_A_G]

Not CPU bound? This piece of crap runs at 40 fps with good cards at 1080p.

At the maximum settings where you can still see a difference between GPUs. When you drop the settings you see everything from the 950 to the 1080 performing at almost precisely 60 FPS and 60 is a bit too perfect to be a coincidence.

In my experience doing GPGPU work the fact that you don't get the performance difference between different GPUs than what you expected doesn't always mean you're CPU bound. Apart from just badly written shader code different GPUs often have the same amount of fixed function hardware, which when used will at a certain point will bottleneck additional device/shader threads. Found this out the hard way when I was writing a CUDA application and noticed that my 970 and 680 were performing almost the same. Turns out Nvidia GPUs turn to the fixed function hardware when doing modulo (did this to figure out what part of the work each thread should do) and Nvidia hadn't boosted the amount of fixed function hardware by anywhere near as other hardware resources between the two cards. Came up with a trick that used the downward rounding of integer math when doing divisions to do modulo without actually using modulo and shaved off a decent amount of the runtime on both cards, but a much bigger amount on the GTX 970.

In other words: When it's not GPU bound it's probably purposefully limited to 60 FPS.

[dungeon]

Game is not 60 fps limited, read on GoL comment that it goes 80-100 fps down to 30 fps on Ultra... it is just that benchmark scene is very demanding, probably is there just to show you random sort of average worse case scenario

And of course it wants to bite all of your singlethread IPC

[bug77]

No difference between FHD and 4k, now that's a crappy port.
Granted, this can be played nicely at 30fps, but having your CPU work that hard when it shouldn't is going to add (a little) to your electrical bill.

[Mike Frett]

So it's a bad port then? That's just unfortunate but we are partially to blame for pushing them about releasing this Game...

[L_A_G]

Just ask Michael to run bench on some lower resolutions like 720p, if it is same @60 fps bound there that way your cap theory might be proven as true.

I don't think that's actually necessary when Michael's results for 4k at low settings were also 60 (+/- 2) FPS from the GTX 680 to the 1080.

As for the benches over at GoL, the supposed 80 FPS readings were from actually playing the game (with little going on) and not the benchmark mode. The benchmark figures over there are all less than 60 FPS.

[AsuMagic]

There's nothing wrong with the extra occasional needless redundant word.

yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep. yep.

[liamdawe]

Game is not 60 fps limited, read on GoL comment that it goes 80-100 fps down to 30 fps on Ultra... it is just that benchmark scene is very demanding, probably is there just to show you random sort of average worse case scenario

And of course it wants to bite all of your singlethread IPC

To confirm this: https://www.gamingonlinux.com/articl...rt-report.9075

It is not limited to 60FPS, so don't go down that road. The video I did also show it going well above 60 in places. The majority of it once you get two-three hours in is much lower though.

[DJViking]

Would very much like to see benchmarking with a SLI setup of GTX 1080. Aspyr has said that SLI should work.

[killyou]

I have a 9 years old "gaming" PC which I recently updated with quad core Xeon E5472 3.0 GHz, 8GB RAM and GTX 1050. Probably the whole thing costs like 300$ or something. Granted, I was not testing it under a heavy scenario but it runs Civ 6 on ultra at constant 60 FPS on Windows at 1680x1050. If playing Civ 6 on linux means I would have 17.99FPS according to the benchmark, or less because of the old CPU, then it's terrible, cheez. Civ 6 is somewhat playable on my Windows work laptop, Lenovo T440s with intel HD GPU so the Linux performance is not something to be proud of. The macOS port is slow as well as it's barely playable on mid-2014 MBP with Intel Iris Pro.