GTA V on AMD CrossFire - low GPU utilization

AFAIK DX12/Mantle are meant to help users with average CPUs by offloading tasks from CPU to GPU. This could help multi threads too.

 

Oh yeah I agree 150% It's strange indeed.
I don't have the game yet so can only guess looking at the charts.

Maybe the scheduler is skewing things or some of the threads are just being shuffled around and not executed simultaneously.....no idea tbh but yeah pretty weird...

Btw that reminds me I got a question about overscan on Nvidia cards, I'll pm you later.

 

What I have found (albeit in single gpu configuration), is that the in game vsync is (mildly put) bad.

What I did is that I run the game in a borderless window and I force vsync through RadeonPro. It has been running like butter since. No limiters, no triple buffering, nothing.

I have everything maxed except MSAA, Reflection MSAA and the Advanced graphics, and it runs like like a dream on my 7970 at 1080p.

 

If you are still at the start in the city, it does not really reflect the overall performance you will get out of the game.
Even the benchmark is not a true reflection of it as well. Try driving up north into the countryside during daytime, that's where fps will drop. Grass is a killer.

 

Borderless window is very different from everything else.
http://superuser.com/questions/558007/how-does-windows-aero-prevent-screen-tearing

 

Sorry about the delay.

Low utilization is a fact, I am using 15.4 drivers with Default CF profile, resolution is WQHD (2560X1440@60Hz), Vsync is set to off, ULPS is disabled.

The in-game settins are set to HIGH and VERY HIGH, Post FX is set to ULTRA... This thing is a framerate killer, but the game looks great on ULTRA.
I use FXAA or MSAAx2, but never at the same time... MSAA gives very crisp textures and FXAA renders blurred mush.

Anyways, I noticed that indoors or when the character is facing a wall, the usage bumps to around 90%. In all other cases - driving, flying, running around the city... never goes above 70%...

Must be some kind of bottleneck...

 

I know, that's why I did it

The point is what with borderless window but without vsync from RadeonPro, I get jerky movement.

The absolute best settings were borderless window, vsync through RadeonPro, and selecting DirectInput for the mouse in game.

 

Hi mate here is my afterburner settings image :


in gtav the settings i use are all maxed out even the advanced settings all sliders maxed.
the only thing i dont enable is MSAA and use FXAA instead.
@3440x1440 resolution.

i am using 15.4 beta`s
at stock clocks yes 1030/1300 stock @ default 0% powertune, i didnt think i would need more powertune @stock as other games are fine with 0% at stock.. + at stock i use -60% voltages as this card works great undervolted.

as you can see from the image though when overclocked i use 40% power tune.

i dont understand it though why this is happening i will do some more research and report back to try and figure this out also.

 

OK looks like i have figured out the issue.

GTA5 has GPU utilization issues.

with no msaa off and fxaa on gpu fluctuates all over the place.
with 2x msaa enabled gpu usage stays @ 1030/1300 with 99%/97% gpu usage and does not change gpu core speeds but if i look at the sky then i will get gpu cores fluctuating again.
if i put msaa up to 4xaa then clocks will stay put with 1030/1300 @99/99 usage and stay there and if i look at the sky it will stay there also.

seems the game has issues with gpu demand when no msaa is used.
it shouldnt be like that as the game is pretty demanding even without msaa and with msaa the game does not feel smooth

but at least i have figured out the issue its possible this issue needs a patch by rockstar or ati to fix .... as changing msaa level should not make core clocks change.

i tested also default clocks 0% power tune and fxaa and they where bouncing between 1030 on one gpu clock to 300 on the other and 300 on the other back to 1030 on the other with gpu usage boucing also between 80% to 30% and vice versa...

defiantly something fishy going on here.

and it cant be overheating VRMs as these issues happen to me within seconds of starting the game...... and when it was doing it at stock temps where around 48/52c.

 

Okay fair enough.

Are you using Vsync? If so you can smooth out the game by capping FPS at 50. Works well for me and feels the same as 60fps.

Disable MSAA, or at least lower it to x2 AA. Run distance scaling at 0, run grass at very high/high. If you still have smoothness issues lower post FX to very high. A combination of these settings increase stutter i found and the image quality difference between the settings is minimal, at least at 4K which is what i use for my QuadFire setup.

 

So where is the culprit?

FXAA?

I tested with MSAA maxed and still get the GPU fluctuation and here i locked my clocks by AB.

?¿?

 

I thought the thread was about low GPU usage in GTA V on AMD 15.4 drivers not about "smoothness".

I have smoothness, full GPU usage will be great also.

 

We've explained, the game is bottlenecking. All high-end systems get reduced GPU usage outdoors.

 

Yeah but matt mate doing those things still does not solve the underlining issue.
is this issue going to be getting looked into
as its like the game currently has a gpu usage cap when fxaa is used and no no cap when msaa is used and when fxaa is off and msaa x4 is used its perfect but then that defeats the point as most people like to run higher resolution with no need for msaa as its not needed but with fxaa enabled its like our cards are being choked to death lol

really needs fixed....... as these issues are not normal.

 

I understood the game is bottlenecked, it's obvious, i saw it, it's a fact

What is not so clear is what is the bottlenecking culprit using the game in cfx with 15.4 drivers:

- CPU

- DX11 (with DX10 i see 25-55% GPU usage)

- The game itself?

- AMD 15.4 CFX

- A combo of some or all of the previous options


?¿?

 

It would be interesting if somebody would check the game for the same bottlenecks, with the newer leaked Windows 10 drivers.

Apparently they have stutter problems with the texture streaming of the game, BUT they have much higher CPU utilization and efficiency scores on 3dMark for DX11.

You could check if you get the same utilization issues with these.

 

I don't see a big difference.

I tested the game on W10 with 0401 driver and cfx that's what i see:

- GPU usage a little better

- CPU usage almost the same

- FPS a little less

 

I think it's obvious isn't it? In places that are not too heavy on the game engine, such as indoors, Crossfire scales perfectly. Outdoors, as you raise graphical options, GPU usage rises so it cannot be CrossFire bottlenecking. As I said before, SLi systems face the same issue. Look at Titan X benchmarks on GTA V too, it's bottlenecked at 1440p. CPU usage hits nowhere near 100%, so it must be a combination of the game and DX11, no?

 

It must be obvious now it's clear the CPU is not the bottleneck root cause in cfx use.

A combo between DX11(DX10 is even worst 25-55 GPU usage) and the game itself.

AMD CFX and Nvidia SLI drivers can't fix DX11 limitations but they can help Rockstar to fix whatever is posible in GTA V itself.

It's ridicoulus that a quad 290X is unable to max out this game in 4K 60 fps as i saw posted!

 

That's pretty much it. It can be optimized more though if AMD and Rockstar work together but AMD's CPU overhead worse than NVIDIA so it's possible.
It's just a shame that these great games that we have now won't be DX12 compatible. How long before another GTA on the PC?