Flying at 5760 x 1080 @ 130 FPS - Long Post

[rbennion 0]

I spent my weekend testing P3DV2.1 performance and just made a really long post over at the LM forum.

 

I hate to cross post, but I think P3DV2 fans might find something interesting in my results as I lurk here all the time and learn a lot from everyone.

 

It's also too long to post again.

 

http://www.prepar3d.com/forum-5/?mingleforumaction=viewtopic&t=5822

 

Best,

 

Rick

 

 

[gboz 157]

I spent my weekend testing P3DV2.1 performance and just made a really long post over at the LM forum.

 

I hate to cross post, but I think P3DV2 fans might find something interesting in my results as I lurk here all the time and learn a lot from everyone.

 

It's also too long to post again.

 

http://www.prepar3d.com/forum-5/?mingleforumaction=viewtopic&t=5822

 

Best,

 

Rick

 

 

Good work Rick.

Great to see some methodical testing.

Word Not Allowed would be proud of you.

 

gb.

[kama2004 74]

Interesting observations...   some thought that come to mind...

 

The difference between surround and single window at 20%. On a system with less capable graphic card, it is almost a 50% hit (reducing by half the fps). I was thinking of changing my 2x580 3 GB and this may be a confirmation of the potential impact of a better card. Just the potential arrival of a SLI support hold me back.

 

I'm surprized at the low impact of AI trafic. Most of the time, AI is considered a FPS hog but that doesnt seem to be the case in your test. Maybe it's because the system was paused?

 

Volumetric fog was not tested ? this would have been interesting as it also consume a lot of processing power. I got a hit of 40% of fps in fog scenario in surround mode on my low end gpu.

 

I'm surprized of the low impact of MSAA (no impact at all). On my system, in surround MSAA at 2x give a 20-30% reduction. MSAA at 4 kill the FPS with a 80% reduction. So seems your system handle that quite well...

 

Tesselation was suppposed to be a big hit on framerate but not here... at 1 fps only.

 

Shadow cast by vegetation and terrain was another setting suspected of reducing fps. In your test little effect from the shadow.

 

So interesting to see those results as some of them make me question some conclusions that were drawn during the first week of p3d 2.0 and 2.1 trial..

[rbennion 0]

 

I'm surprized at the low impact of AI trafic. Most of the time, AI is considered a FPS hog but that doesnt seem to be the case in your test. Maybe it's because the system was paused?

 

Good point!

 

 

 

I'm surprized of the low impact of MSAA (no impact at all). On my system, in surround MSAA at 2x give a 20-30% reduction. MSAA at 4 kill the FPS with a 80% reduction. So seems your system handle that quite well...

 

I was too. Are you using any NI settings?

 

 

 

Tesselation was suppposed to be a big hit on framerate but not here... at 1 fps only.

 

I read Tessellation takes load off your CPU and puts it on the GPU. In some cases I read it increasers FPS.

 

 

 

Shadow cast by vegetation and terrain was another setting suspected of reducing fps. In your test little effect from the shadow.

 

I thought Shadows would be a much bigger drop as well. I have put them back to MAX in my daily profile.

 

 

 

 

 

[kama2004 74]

Yes I was using NI settings. I checked and I saw that Bioshock was still there and I know this is one setting quite demanding and I added it a while a go while testing (I removed it now). The rest was not higher than in P3D.

 

Your test gave me the push to try something on the same lines...

 

I never used benchmarking tool before so I downloaded Fraps andEVGA precision and this was quite interesting. First thing I noticed after install was that my 2 GPU were maxed at 99%. Didn't realized but I pushed my sliders probably to much and the GPU is really the bottleneck. I was in fact adjusting them for the minimum acceptable FPS....

 

 

I tested from all sliders minimum scenario and moved to the right one at the time to see the result. Here some findings.

 

Absolute minimum on all sliders:

FPS = 134, GPU1 and 2 around 43%, Core1 = 20%, Core 2 = 100%, Core 3 = 30%, Core 4 = 30%.

 

First question, why the FPS is "only" 134 if the GPU has still a lot of headroom and the CPU on average is at 23%. Is it because some thread where allocated to Core2 which become saturated at 100% then the rest of the core and GPU processing power is just wasted? I may try to play more with the affinity mask but I don't think you can do more than unlock the core0 for p3d also and control what is sent to Core 2 ......

 

Scenario: All Shadow boxes ticked

FPS = 70, GPU 1 and 2 at 99%, Core1 20%, Core 2= 80% (reduction), Core 3 and 4 = 30% (unchanged)

 

This one is interesting and shows that even with minimum autogen or scenery complexity, the shadow options can saturate the GPU. Even more, we see that the CPU as less load because the GPU is saturated so the CPU usage is reduced with the framerate. This one alone make me think of upgrading my video card...

 

Scenario: Augoten Vegetation max

FPS=70, GPU1 and 2 = 53%, Core 1= 20%, Core2 = 100% maxed, Core 3 and 4 = about 35%.

 

So this one raise again the question. Why the system limit fps to 70 while there is still headroom both on the GPU and 3 cores out of 4. Again, is it core 2 that is the bottleneck?

 

Trafic doesnt cause a huge decrease in FPS in my test also so this seems to confirm your tests. Only road trafic slider cause a hit of 30 fps but I dont see an increase in GPU processing power and not huge for CPU. Is this an indication that trafic is still mainly CPU power processing and little in term of graph power? Makes sense as a lot of trafic handled by the system is just out of view so nothing to send to the GPU to draw...

 

MSAA at 2 or 4x and even 8x also dont cause to much stress on my system in this scenario. Probably because there is minimum autogen so not a lot of things to handle for AA. I lost with 8x only 7 fps and the gpu just increased to 55% from 43%.

 

Interesting to see that 28 of the sliders results in slight reduction of 14 FPS max. 6 sliders only cause an important hit. I realize however that some of the 28 sliders are related to the 6 others and if we crank-up the remaining 6, the performance hit caused by others sliders at max will increase also. 

 

Scenario with 28 out of 34 slider topped:

FPS= 125, GPU 1 and 2 at 82%, Core 0 at 20%, Core 1 at 100%, Core 2 at 30% and 3 at 40%

 

So this is also very interesting as it shows that I got only 9 FPS lost with 28 sliders to max...

 

The 6 remaining sliders that control the rest of FPS reduction and power increase are:

 

- autogen vegetation

- all shadow box ticked

- autogen building

- reflexion 

- Road vehicle

- Detailed cloud

 

The road vehicle is probably not affected by any other sliders (maybe the AA). The detailed cloud will be obviously linked to the cloud coverage and cloud draw distance. The autogen will probably will be linked with the level of detail radius, scenery complexity and AA settings.

 

Anyway, I will pursue the test to understand more how to balance my system for optimum performance under different scenarios ...