Is it the GPU or CPU?

[benargee]

Hello everybody im fairly new here. I just bought myself a new Graphics card, a Galaxy GTX 670 1006Mhz (67NPH6DV6KXZ) and im not quite getting the results i had hoped for. I currently have a I7 930 Processor stock clocked at 2.8GHz.

 

i was wondering if anybody would know what in my system would be the bottle neck. how can i get better performance out of DCS World and A-10C

 

My other Specs are

Triple Channel 6GB ram (2GBx3) 1600mhz

WD Black Series 640GB HDD

750w Corsair PSU

ASUS P6X58D-E Mother Board

Ben Benchmark 123.trk

Edited March 25, 2013 by benargee
Added benchmark track

[Lange_666]

So what is the result you are getting?

[pacotito]

The easiest test for cpu/gpu bottleneck would be to compare fps between high and low graphics options with the same mission(a track would be better). If there is not much change its a CPU bottleneck. Also check you nvidia control panel and make sure your not forcing aa.

Edited March 14, 2013 by pacotito

[Sorin]

Overclock that 930, it will make a big difference. 670 GPU should be plenty fast for one monitor.

[Erforce]

definitively CPU.

 

My 670 is about 80% load in cockpit and 100% external view. (everything maxed out)

spec i7 3770k@4Ghz

 

my 2006 Xeons @2.66 got a 7870 this week, and still same FPS than my 4890. (ie : pretty low)

GPU load at about 20-30%

[EtherealN]

Overclock that 930, it will make a big difference. 670 GPU should be plenty fast for one monitor.

 

This.

 

Obviously, it's an operation that is done at own risk (and mobo performance really matters here as well), but I've seen those CPU's do 4GHz on 40 dollar air coolers. What you get specifically may vary obviously, since no two CPU's even on the same model are equally resilient.

 

Hell, I've actually seen one of those do 4GHz while UNDERvolted. But that must have been a lucky specimen. :P

 

Since most aftermarket coolers are multisocket, it's definitely worth trying this before deciding on whether to get a new CPU, since a new CPU would definitely mean a new motherboard (and _possibly_ new RAM) anyhow. If the experiment doesn't get you where you want to go, just continue with the original plan and use that cooler on the new equipment. (But in that case, ensure that you get some new thermal paste for the reapplication, never EVER re-use thermal paste.)

[benargee]

Alright so i know this isn’t an ideal situation but im spectating an ai mission with 20 A-10Cs on the ramp 4 taking off and 4 SU-25Ts and im getting like 7-12 fps

Settings

Default High Settings+

Civ Traffic: Off

Textures Scenes Water and Shadows: High

Preload Radius: 15Km

Tree Distance: 6Km

Bushes: 500m

VSync: On (I know, I know, but I hate tearing)

MSAA: Off

 

Checked my CPU HDD performance according to Windows 7 Resource manager and my CPU 0 is peaking at %80 and averaging %60. the HDD peaks at around 1Mbit/s as im playing if that means anything to you

 

And for my GPU Using EVGA Precision during this whole fiesta, is only peaking at %56 average %50.

 

On another note it seems that Fps is a little bit better before all the jets start their engines, and stays low the entire time they’re running (of course until they all take off and lessen the system load) I have heard this game has issues with the sound engine before, or could it be all those polygons spinning on the engine animation?

 

No you may ask, Ben why such an extreme situation? Well i just got this “awesome” new card and i’m wondering what my next upgrade would be, or if this is normal for a gtx 670. So i'm essentially doing my own little benchmark here (or i just want to see a lot of russian A-10Cs get slaughtered by the patriot system 4 at a time >:) ). The concern came up when i was running the massacre and it ran slow.

 

Another instance of slow fps was a 4 on 4 AI P-51D dogfight (US vs CAN) and at the first engagement of .50 fire i took quite an FPS hit (less than 30)... this was only 8 planes now. (USA won 3/3 times until a mighty CF-18 came to the rescue :P)

 

My thoughts are I either need SSD, RamDisk on more RAM or CPU cooler + OverClock, that is if the video card is at spec. What do you think? sorry if i rambled. thanks.

 

ask for more info if needed, i have my social insurance number handy...

[Lange_666]

My thoughts are I either need SSD, RamDisk on more RAM or CPU cooler + OverClock, that is if the video card is at spec. What do you think? sorry if i rambled. thanks.

 

SSD will only make your game load faster, you gain no FPS with it.

RAMDisk is cool but you don't have enough RAM to do this since you have copy the entire game to RAM first before you can run it. Used to do that with the A-10C standalone version to prevent loading stutter but now my DCS World install is 10,9 GB big. If you would add 6 extra GB, you would still only have 1.1 GB left for running OS + DCS. And, setting up a RAMDisk + copy your entire DCS World dir over takes time to. Imo, it's not worth the effort and it will only tackle loading stutter.

Overclock is an option, no clue on how much it will bring.

 

I would trade in some eye-candy for performance if i where you.

[_Dredd]

I went from a i7 920 @ 4Ghz and 6970 to my current i7 3930k @ 4.6Ghz and 7970, and I would estimate the performance gain was 25-50% more.

 

Bear in Mind the 3930k has similar specs to the 980X but has more bells and whistles.

 

If you can't afford a massive upgrade, look around for a cheap 980X

[Corrigan]

CPU. In my view it's quite a waste of money to run a modern Intel CPU at stock frequencies.

[Sorin]

Benargee, you must have Win7 64 bit and DCS World 64 bit. You do, right?

 

Can you post here your mission file? I will try your benchmark on my system.

[benargee]

Benargee, you must have Win7 64 bit and DCS World 64 bit. You do, right?

 

Can you post here your mission file? I will try your benchmark on my system.

 

yes i have 64 bit windows... not sure if i have 64 bit dcs, is it a separate download or just an installation option? i will post the track in my original post in a bit

[Hubris]

yes i have 64 bit windows... not sure if i have 64 bit dcs, is it a separate download or just an installation option? i will post the track in my original post in a bit

 

Pretty sure DCS is only 64 bit.

[cichlidfan]

Currently DCS installs the appropriate files (32 or 64 bit) based on OS. It has been announced that this will change to 64 bit only in the near future.

[benargee]

Alright, i added the track on my main post, after all engines have started up i get around 7-10fps when most of the aircraft are in view

[Sorin]

Ben, I have tried your track but to make a benchmark from it we must agree on a camera to use. I recommend F5 camera. So basically we do these steps: start replay, Ctr-Y twice to remove the status bar and then press F5.

 

Next would be to record the FPS, CPU usage (all cores), GPU usage, GPU memory, system memory, etc. Minimum FPS and average FPS should be shown as well.

Ideally all the parameters should be plotted on the same chart. That is something I cannot do right now but I will figure a way because we must have a benchmark for DCS and a proper way to record all the details.

 

Of course all parameters in hardware, DCS options, the video card 3D settings, the operating system, display resolution, etc must be the same so we compare apples with apples - but that is impossible.

So we should just list our details so it is at least known what we have and what are the results. We have to come up with a table format to report the hardware parameters and software configuration.

 

So far I do not have results as mentioned above but preliminary on my Windows XP 64-bit the performance is quite lame, it gets to 6 FPS sometimes. I will post some proper results soon.

[benargee]

Ben, I have tried your track but to make a benchmark from it we must agree on a camera to use. I recommend F5 camera. So basically we do these steps: start replay, Ctr-Y twice to remove the status bar and then press F5.

 

Next would be to record the FPS, CPU usage (all cores), GPU usage, GPU memory, system memory, etc. Minimum FPS and average FPS should be shown as well.

Ideally all the parameters should be plotted on the same chart. That is something I cannot do right now but I will figure a way because we must have a benchmark for DCS and a proper way to record all the details.

 

Of course all parameters in hardware, DCS options, the video card 3D settings, the operating system, display resolution, etc must be the same so we compare apples with apples - but that is impossible.

So we should just list our details so it is at least known what we have and what are the results. We have to come up with a table format to report the hardware parameters and software configuration.

 

So far I do not have results as mentioned above but preliminary on my Windows XP 64-bit the performance is quite lame, it gets to 6 FPS sometimes. I will post some proper results soon.

 

agree, will do, but doesn't the camera go automatic for you, the same that i used when i recorded the track?

[emenance]

CPU. In my view it's quite a waste of money to run a modern Intel CPU at stock frequencies.Win7 x64 | 2 120 Gb SSDs |i5 2500K @ 4.5 GHz | 8 GB RAM | GTX 580 | |2 120 Hz 22'' monitors | TM Warthog HOTAS | Saitek pedals | TIR5

 

 

This Corrigan config or my config will get you into a sweet spot.

[Sorin]

agree, will do, but doesn't the camera go automatic for you, the same that i used when i recorded the track?

 

The replay does repeat what you have done but it is F2 camera which is affected by the mouse (any mouse movement during the replay changes the camera angle). We must have a reproducible camera, that is not affected by accidental mouse movement.

 

Also F2 is too close to the aircraft. F5 shows much more and it is more demanding.

As an example F2 does 23 FPS average while F5 camera does 9 on my system (measured with Fraps, one monitor 2560x1600, GTX680, 3.9GHz CPU, 8G RAM, SSD).

Try F5 and then tell me about slow...

As a compensation you get to see the SAM kill one of the A-10C and the pilot parachute :)

 

Anyway, let's find another track for benchmark because this one is not good. The drastic FPS drop that happens when the engines start is not due to the sound or engines simulation, it is due to the landing lights (on the nose wheel). Having so many light sources in the 3D scene totally kills the FPS.

[AngryBird]

To put it in perspective I have an I7 3770K Currently at stock 3.5ghz an EVGA GTX 670 SIG 2 with 16gb 1600mhz Corsair memory an 120gb ssd an a couple 1tb mechanical drives. On a bad day on max settings I hit maybe 50fps with an average of 80-90 and max about 130.

 

In your case yes its the CPU which is obviously bottle necking.

 

An i5 3570k or if you dare go to the dark side an AMD bulldozer 8350 both would do you just fine. Or you could always just wait till haswell?

[Sorin]

AngryBird please try Ben's track on F5 camera and let us know the results. Preferably measured with Fraps. What monitor resolution do you use?

[EtherealN]

Agreed on that track: it won't give any worthwhile information since it's basically made to ensure that any and all computers with bottleneck terribly on the CPU, basically telling us that "yes, it is possible to design situations in the mission editor that no computer will handle well".

 

 

(In this case, my machine is as per signature but without the overclocking. Some variability in the exact curve locations is caused by the manual start/stop of frame tracking.)

 

To get useable information, what we would need is either separate tracks for different types of situations/locations, or a track that within one or two minutes transitions between several such situations. Most especially since it would allow identifying specific issues: for example, if we notice a dip with one configuration but not in another, we can thus find any specific setting that will give our specific computer greater protection against performance thieves. :)

 

(TSSAA is one of those on my computer, being asked to run MSAA, two screens with Aero PLUS the game window is a bit harsh on a 560Ti, though I do feel it does an admirable job of it. :P )

[Sorin]

Thanks EtherealN for your analysis; it mirrors my findings.

 

I think the attached track would be more suitable for a simple benchmark (only 80 seconds).

It is done with F11 camera but it may be close enough to a helicopter flight. There is no need to change camera, just play and avoid any input.

 

We should try to make a track that is demanding but reasonable.

Benchmark.trk

[vodkapom]

Very intersting topic. I have the same issue with an I7 2600K@4.4 Ghz, AMD 7970, 16GO RAM and SSD 120 GO.

I have usually 30/40 FPS, but when AI is in front of me, FPS drop to 5/7 FPS.

I will test this track tonight and i will report here my results.

[EtherealN]

Here's my results Sorin.

 

High preset, no TSSAA.

Theen I set Water to Low.

Then I also turn off HDR.

 

 

Going to get a few rounds with much lower settings in later, I suspect object tracking is CPU-bombing it during most of that track, but need to test a few more variables first to be sure.

 

Below is a testrun I'm working on based on a test scenario I've devised that does a little of everything, doing combat in-cockpit A-10C with TGP on, but this here is preliminary only - as you can see there definitely is something that is happening there and I'm looking at figuring out what it is. (I have my suspects, as you will probably guess if you run it. 5 minutes long unfortunately.)

 

FPS-test_330sec.trk

Edited March 27, 2013 by EtherealN