Factorio Forums

Hi all,

I'm noticing bad graphics performance on Windows 10 compared to Linux 4.20.3 with an AMD RX580 8GB.

Playing on Linux the performance is butter smooth, the game seems completely CPU/RAM bound. FPS is always equal to UPS, even when completely zoomed out. UPS only sometimes dips between 55 and 60, due to the fact my base is approaching a size on which my CPU/RAM start struggle. (ever so slightly tho)
However on Windows 10 graphics performance is pretty bad, especially zoomed out. FPS is always lower then UPS, and often hangs around 30 fps zoomed out, UPS is the same as on Linux, 95% of the time straight 60 and sometimes between 55 and 60.

I'm using the same (graphics) settings on both Windows 10 and Linux.

I'm not experiencing any other GPU problems on Windows 10, games demanding much more, like the Witcher 3, run smooth 60 fps with maxed out (ultra) settings.

I ran some GPU profiling on Windows 10 and Linux while playing the game.
some screenshots are listed below for reference
keep in mind that maximum specified clocks for an/my RX580 are:
GPU core clocks: 1340mhz
GPU memory: 2000mhz

Typical Windows 10 GPU utilization during play:
GPU core clocks: between 300 and 500mhz
GPU memory clocks: up to 500mhz
GPU util: never exceeds 50%
FPS: between 30 to 60, feels stutterly to play.

Typical Linux GPU utilization during play:
GPU core clocks: between 1000 and 1200mhz
GPU memory clocks: between 1000mhz and 2000mhz
GPU util: n.a. (used 3rd party software, gpu util calculation doesn't to work)
FPS: always maxed/UPS bound.

hardware specifications
i5-3570K,
8 GB DDR3-1600mhz (single slot)
AMD RX580 8GB

drivers:
Windows 10 1809 with AMD Adrenaline 18.12.2 and 19.1.1 (tested both drivers, both show the same issue)
Arch Linux 4.20.3 with AMD DDX 18.1.0

For the time being i can just play on Linux, but as you can see the GPU is quite under utilized on windows. The GPU could be pushed a lot more. I'm unsure if this is a game or a driver issue or there is another mechanic at hand i have no notice of. Nevertheless I thought I put it on the forum to report the issue as I only experience it with Factorio. Other cross platform games don't show this issues and other GPU demanding games run as expected on Windows 10.

With regards, Visione.


Screenshots
Linux GPU profile: Linux ingame performance: Linux settings: Windows ingame performance and GPU profile: Windows settings:

Have you tried using opengl on windows (--force-opengl launch option)?

Hi Bilka,

Thanx for the quick reply.
I just tried launching factorio with opengl (--force-opengl). The same issue persists. GPU utilization is also still low.

Have you tried --force-d3d instead of --force-opengl (just in case your graphics settings defaulted to OpenGL renderer back in 0.12 days)

I think the rendering is CPU bound in this case and maybe the game is even dropping frames to keep UPS at 60. If that's the case, it should be much better in 0.17 as that area was very optimized. Anyway there is debug overlay under F4, that can show some basic timings - see: viewtopic.php?f=7&t=3638#p299295

If there is lot of time spent in Flip, rendering is GPU bound, if lot of time is spent in Render or Render preparation, it's most likely CPU bound.

Aah, thank you.

Render preparation is fluctuating between 3.00 and 9.00, render (game render) is roughly at 17.00 and Flip is around 2.00.
I guess it is most likely CPU bound after all, cool to see it plays out that way, lesson learned.
Also interesting to see how the difference in OS affects in-game performance so much.

edit: just to round it off
started up Linux,
render preparation is at ~3.00
render is at ~10.00
flip is at ~0.2

Thanks for the help , I guess I have to advance my CPU/RAM upgrade sooner then later. And stick with Linux which was my preferred OS anyway.

8 GB DDR3-1600mhz (single slot)

Oh, man, add another stick, you need two to get the Double data rate. Doubling memory bandwidth never ever sucks.

quyxkh wrote: Sun Jan 20, 2019 6:36 am Oh, man, add another stick, you need two to get the Double data rate. Doubling memory bandwidth never ever sucks.

Yes, I agree, it is a flaw in this build, but I consider it too late by now, I'll be building a new rig in a few months, no use in buying old DDR3 when ill be switching to DDR4 very soon i can wait that long

a couple thoughts come to mind with this, does the performance improve by setting the GPU to stay at the higher speed in windows(if possible to set)? also, I think you'd be surprised(probably appalled) at how much background CPU usage there is in windows vs linux