No FPS increase after video card upgrade

[Gloom Demon]

I have purchased a Geforce 760 videocard to replace my trusted Geforce 560 Ti 448 cores. Of course, I did not expect a major improvement, however different Internet reviews and comparisons promised something around a 35% FPS increase.

 

I have created a mission over Tbilisi with a lot of moving ground and air units and plotted a course for the Su-25T over it. I allowed the plane to fly without any interference and measured the FPS using FRAPS for 60 seconds. I have all textures quality set to high, no antialiasing or anisotropic filtering, no shadows, no mirrors, blur is on and HDR is set at warm colors. My 8320 CPU is overclocked at 4200 Mhz (at this frequency it is completely stable and there is no need to increase voltage) and my screen resolution is 1920x1080.

 

The results are suprising:

 

Geforce 560 Ti 448

Avg: 28.867 - Min: 21 - Max: 41

 

Geforce 760

Avg: 29.417 - Min: 21 - Max: 42

 

Just in case i wasn't doing something right I returned the CPU to its nominal frequency of 3.5 Ghz and ran 3DMark Vantage on both cards:

 

Geforce 560 Ti 448 - 19447

 

Geforce 760 - 23845

 

The difference for the 3DMark graphics score alone was even bigger (without 3DMark measuring CPU effectiveness):

 

Geforce 560 Ti 448 - 19418.1 (59.1 FPS for the first test and 54.6 FPS for the second)

 

Geforce 760 - 25803.8 (75.5 FPS for the first test and 75.7 FPS for the second)

 

So if 3DMark shows a 33% FPS increase when upgrading from a Geforce 560 Ti 448 to Geforce 760 how can it be, that DCSW does not show virtually any increase in FPS?

 

Is it because of the CPU not being powerful enough to utilize the power of the mainstream Geforce 760, or is it because DCSW uses the CPU to such an extent, that there is not enough left to fully load the video card?

 

I'll be glad for any input clarifying the situation.

Edited July 26, 2013 by Gloom Demon

[Kuky]

CPU bottleneck - remove all those "many ground and air units" as they use lot more CPU cycles then test again, there should be more FPS difference

 

Hint that CPU is the bottleneck is when that min FPS value is the same

 

another thing you can do to show more GPU power is increase AA to max and fly very low (so that trees and grass show close) and that will put lot more strain on the GPU, then measure difference in FPS in % and it should be much closer to 33% you were expecting

Edited July 26, 2013 by Kuky

[Gloom Demon]

Thanks, I'll try that and post the results :-)

[Rhinox]

Answer to your problem is in your signature: AMD 8320

[whartsell]

If you want a good systematic approach to performance in DCS read this thread.

 

Your benchmarking with too many unknowns but if I had to guess on your problem it would be the CPU

[Gloom Demon]

Well, I deleted all of the units from the mission and ran a couple of tests again with both of the cards. The results are as follows:

 

Geforce 560 Ti 448

Avg: 78.600 - Min: 72 - Max: 85

 

Geforce GTX 760

Avg: 80.583 - Min: 75 - Max: 87

 

With almost all settings ramped up to the max the difference increases somewhat:

Geforce 560 Ti 448

Avg: 58.800 - Min: 51 - Max: 63

 

Geforce GTX 760

Avg: 65.350 - Min: 59 - Max: 72

 

Still kinda short of the expected 35%...

 

whartsell That's the whole point - according to numerous reviews 8320 (or 8350) demonstrate an increase in FPS with different video cards in a variety of games... other than DCSW, which is puzzling (that's why i ran the 3DMark tests).

 

P.S. I have attached both missions just in case

Test GPU.miz

Test GPU Simple.miz

Edited July 26, 2013 by Gloom Demon

[Gloom Demon]

Here is how I ran the tests:

 

1. Booted the computer

2. Started Fraps, configured to measure FPS for 60 seconds when the proper combination of keys is pressed

3. Loaded up DCSW

4. Loaded the mission

5. Pressed Ctrl Pause to see the ingame FPS meter

6. Pressed Fly

7. Waited for 10 seconds to pass on the ingame FPS meter to prevent disk swaps (the game is rather unpredictable when it starts in terms of FPS)

8. Pressed the combination of keys to engage FRAPS

9. Waited till the 80th second and turned the whole thing down

[trooph]

Are U running that CPU on stock clocks? Maybe try oc a bit...up to 3,8 maybe? That should help.

Remember that DCs uses only 2 cores only.

 

I had a bit of a problem with my old time rig :) X4 965be at stock clocks was bottleneck for my hd 7850. Now i dip bellow 30 fps just with CBUs expoding and all the DCS fps killers...

[Gloom Demon]

trooph Nope - its overclocked to 4,2 Ghz from 3,5 Ghz.

[EtherealN]

Unless the review and test you read SPECIFICALLY tested DCS World, don't use it as a reference. Most computer games are extremely GPU-centric. Not so with DCS World, which is a serious simulation product that needs to do a lot of maths that GPU's aren't well suited for, meaning that they'll happen in CPU.

 

Quite possibly, you bottlenecked outside GPU in several situations even before your graphics card upgrade. At that point, literally ANY graphics card upgrade (say, quad-SLI 780's) will give very very little benefit.

 

Remember: "framerate" is NOT something that happens in the GPU. The GPU needs to be told what to do - which is something the CPU does. And the simulation itself also runs in "frames", so if your computer can only do X simulation frames per second due to other issues, no amount of upgrades on the graphics card will give you a boost, because the graphics card will only get told what to do that same X amount of times per second.

[whartsell]

Well, I deleted all of the units from the mission and ran a couple of tests again with both of the cards. The results are as follows:

 

Geforce 560 Ti 448

Avg: 78.600 - Min: 72 - Max: 85

 

Geforce GTX 760

Avg: 80.583 - Min: 75 - Max: 87

 

With almost all settings ramped up to the max the difference increases somewhat:

Geforce 560 Ti 448

Avg: 58.800 - Min: 51 - Max: 63

 

Geforce GTX 760

Avg: 65.350 - Min: 59 - Max: 72

 

Still kinda short of the expected 35%...

 

whartsell That's the whole point - according to numerous reviews 8320 (or 8350) demonstrate an increase in FPS with different video cards in a variety of games... other than DCSW, which is puzzling (that's why i ran the 3DMark tests).

 

P.S. I have attached both missions just in case

 

What were the % utilization of GPU, CPU,Vram and RAM during these tests. Without any diagnostic data there is now way to know how/if this can be tuned away

[SkateZilla]

DCSW Favors Intel Chips, Simply Because Intel Chips have better Single Core/Thread Performance.

 

I Had to Clock my CPU at 5.15 Ghz to match a friends Performance in DCSW, and he's Running a 2nd Gen i5 at 3.4 Ghz.

 

 

 

Also 4 GB of Ram is more of a Bottleneck, because you'll be constantly spooling to VRAM.

[whartsell]

On my machine DCS.exe has a commit size of over 5GB so in your case with only 4GB of total system ram you will be swapping out a bit. I also want to point out that in your "test" mission even without units your flying over a city that has objects. To get a baseline you should compare the two cards on a mission in the middle of nowhere and record FPS/%GPU Util/%CPU Util/%VRAM/%System RAM. This will be your best case scenario and you know it will never get better than this as there are no objects affecting anything. From there we can see if there are any bottlenecks in the most basic of tests. If there are they will be even worse and possibly be masked by other issues.

 

That is if you want to conclusively see:

1. What is the bottleneck on your system

2. measure the difference between hardware

3. measure difference between settings changes

4. measure differences between different objects in the world

 

You should really read through that link i posted again and run all the tests mentioned there and capture the data i mentioned above. With slow FPS is a symptom but not indicative of the cause

 

also what EtherialN said is dead on. I never pay attention to synthetic tests and game benchmarks unless they are specific to simulations as they are very different beasts

[Gloom Demon]

EtherealN Unfortunately, I have not been able to find a review of DCSW performance on different CPU's and GPU's (i sincerely doubt that it exists in the first place), however what you are saying is exactly what I was thinking - DCSW is VERY CPU-dependent, which is not characteristic for most computer games.

 

I am not trying to be a smart ass, or anything, but, given the fact, that DCSW already runs better on nVidia's video cards, maybe in the future CUDA cores could be utilized, to help with the CPU burden (though frankly, I think DCSW would draw a larger audience if it's performance was not that much dependent on specific components - be it CPU's or GPU's).

 

I am still happy with my upgrade, though it's always a shame to say farewell to a trusted component that served you well, isn't it? ;-) Being able to have a better video quality is definitely a feast to the eye and, hopefully, with the new EDGE engine DCSW will become better at utilizing hardware resources.

[jazjar]

Dude I am sorry but I think amd cpus are just not cut out to run games like dcs world. My Intel i7 2600 running on almost itself ( the. HD6770 sucks so bad it might as well not exist) is giving reasonably smooth performance, replace that CPU and I think your issues will vanish.

[whartsell]

I also wouldnt call it a problem. I would love to Average in the 60s but thats unlikely as i run multiple monitors (no full screen as I havent setup SoftTH yet).

[EtherealN]

maybe in the future CUDA cores could be utilized, to help with the CPU burden (though frankly, I think DCSW would draw a larger audience if it's performance was not that much dependent on specific components - be it CPU's or GPU's).

 

Indeed, developing hardware-specific is most likely not an option. (In fact, developers that do that often get kickbacks from the hardware manufacturer in question, as part of that manufacturer's marketing. And not like nVidia is going to care about giving ED money - they've only recently started paying a minimal amount of attention regarding SLI profiles.)

 

Also, physX and so on isn't valuable, and from what I undestand most of the things that are happening in the CPU would actually not benefit from running on the GPU anyhow - GPU's are very powerful, but they are slow, which means that any operations that have data dependencies from other operations won't run effectively on it; they "live" on being able to operate on many small, independent, jobs simultaneously; for example a bunch of pixels. (You typically doesn't have to know anything about any other pixel to decide what coler a given pixel should be, with some minor exceptions.)

 

This is why you see these things heavily used for graphical effects and some types of scientific applications (like Folding@Home), but not for doing simulation.

 

Better is to be able to separate more routines to more threads, to better utilize the resources that do already exist on the CPU but can't be used at present. This is something that is being worked on, but it is no small task to break out threads from a monolithic codebase.

[xracer]

EtherealN Unfortunately, I have not been able to find a review of DCSW performance on different CPU's and GPU's (i sincerely doubt that it exists in the first place), however what you are saying is exactly what I was thinking - DCSW is VERY CPU-dependent, which is not characteristic for most computer games.

 

I am not trying to be a smart ass, or anything, but, given the fact, that DCSW already runs better on nVidia's video cards, maybe in the future CUDA cores could be utilized, to help with the CPU burden (though frankly, I think DCSW would draw a larger audience if it's performance was not that much dependent on specific components - be it CPU's or GPU's).

 

I am still happy with my upgrade, though it's always a shame to say farewell to a trusted component that served you well, isn't it? ;-) Being able to have a better video quality is definitely a feast to the eye and, hopefully, with the new EDGE engine DCSW will become better at utilizing hardware resources.

 

I dont really understand all this fuzz about CPU dependent etc. etc. DCS as of today utilizes the graphics card very well. THE difference between DCS and most other games is the drawing distance combined with very detailed objects and by the look of thiings the GPU is taking care of this to large extent allready. I know what i saw before my GC upgrade and what i see now at 3.8 GHz. Two totally different worlds. Simple as that. If you have a 3.8 Ghz Intel CPU either i5 or i7 then your more or less there if the graphics card is good. The difference shows when you go above GTX-770 or 680. With EDGE hopefully we will enjoy something greater, but if its gonna be that much faster i really doubt it. I think multitreading is the key though.

[Gloom Demon]

xracer No fuss at all :-)

 

If you upgrade a component like a video card or CPU and do not get a performance increase that means there is a bottleneck. In my case it's the CPU not being able to provide enough data to the video card in order to be processed. So no matter witch video card I add I will still get the same FPS (more or less). That is with DCSW, other applications scale quite well.

[xracer]

xracer No fuss at all :-)

 

If you upgrade a component like a video card or CPU and do not get a performance increase that means there is a bottleneck. In my case it's the CPU not being able to provide enough data to the video card in order to be processed. So no matter witch video card I add I will still get the same FPS (more or less). That is with DCSW, other applications scale quite well.

 

Sorry, sorry and sorry my bad, overlooked that 8320 you have. If you look around the forum maybe some other guy here has the same GC as you but a faster Intel CPU, but havent looked into the tread. Maybe he can tell you how his system performs with a similar mission? Also what was mentioned in a post earlier in the tread, do check the GPU utilzation closely while running DCS and see how it does. Use HWInfo64 and log utilization and D3D in a DCS session to a file. Anyway i think that if you go Intel soon and an allround good one at least you should be able to get pretty good performance in DCS. If you go Intel buy a USED IB syst. 2600K or 3770k then OC it. I think you will be happy.