A Few Benchmarks for GPU Upgrade and Limited/Unlimited Frames

[MarkSC 122]

I thought I would post some interesting benchmarks.  I decided to upgrade my GTX 760 to a 780.  The 760 is a good card, but I wanted to see what the 780 could do. (For those who don't know, the 760 is roughly equivalent to the 670, so roughly 40%-50% less powerful than the 780 in gaming benchmarks)  

 

I ran some benchmarks with FRAPS.  With 4x MSAA and 4X SGSS, I flew a circuit around Langley with FTX Global installed.  Sliders were set to medium high.  LOD was high.  Shadows were set to medium with cast/receive on for plane, buildings, and terrain.  Weather was partly cloudy.

 

With the 760 and frames limited to 33:

Avg 24

Min 20

Max 28

 

With the 760 and frames set to unlimited:

Avg 23

Min 20

Max 28

 

With the 780 and frames set to 33:

Avg 31

Min 24

Max 33

 

With the 780 and frames set to unlimited:

Avg 40

Min 29

Max 43

 

It seems that P3D scales nicely in line with the specs for the GPU. It also appears that setting the frames to unlimited is worthwhile for the 780 but not the 760 (probably because the slider settings were too high to meet even the 33 target for the 760).  In short, P3D makes nice use of the GPU's power, which is a welcome change from FSX. In addition to having active development from LM, we can look forward to increased performance in the future as GPU technology continues to advance. 

 

[ricka47 611]

That's good to hear.  I'll my new 780Ti running tomorrow (replacing 570HD) and will be able to see what difference  it makes - thanks!

[BMagann 11]

Ricka47, I made exactly that upgrade on Monday. I think you will be pleased with the result!

[FlyIce 354]

Just couple observations to add. P3D2 uses multiple cores but it always tends to max out on one core under demanding situations. Let's call this the "pirmary core" of P3D.

 

In my cases I have most of my settings maxed or nearly maxed. Overall I'm getting around 25-35 fps in autogen heavy sceneries like KSFO and KSEA, and well over 35 in autogen-light sceneries over mountains of Iceland. What I have observed are the following:

 

If primary core is not maxed out, GPU always runs on 99%. This mostly happens when there are not many autogens like flying over the bare iceland. In this scenario, limiting fps works very well, if the fps fluctuates say 25-35, limiting it to 30 gives butter smooth performance and there is not net loss of fps. Limiting fps also makes primary core works harder.

 

If primary core is maxed out at 99%, then GPU usage drops, often goes down to like 80%-90% or even lower. In this scenario limiting fps give much worse results. If the fps fluctuates between 25-35, limiting it to 30 you barely get sth higher than 25 and the primary core keeps maxed out.

 

Overall, it seems to me that when primary core is saturated, the CPU couldn't feed GPU enough to make GPU run 100%. When GPU is saturated, then CPU may become relaxed as it has to wait for GPU.

[Jimmy RFR 406]

FlyIce, those are very astute observations.  I agree with much of what you said.  I think there's a pretty complex relationship going on between the CPU and GPU; certain conditions seem to tax one or the other more.  I'm not sure though that I could pin it down as closely, as repetition of the same type of flying would often have slightly different results. 

 

I think the load balancing that exists between the CPU and GPU is highly dependent on both pieces of equipment.  Two users might make the same upgrade from one video card to another, but have different CPU's and thus have quite a bit of difference in how the upgrade might affect them.  I think the load balancing is also going to make definitive benchmarking a pain in the butt.  ^_^
 

[okupton 308]

Great work. Thanks a lot for doing this. I am gonna get a 760 because I cant afford the 780.

[SteveW 1,792]

The quoted specs show the GTX 760 is two thirds the texture fill rate of the 780, which compares well to the fps results above. Setting the internal frame rate limiter introduces look ahead frames, which has a cost in ultimate fps, but pre-rendered frames improve the smoothness overall.

[MarkSC 122]

The quoted specs show the GTX 760 is two thirds the texture fill rate of the 780, which compares well to the fps results above. Setting the internal frame rate limiter introduces look ahead frames, which has a cost in ultimate fps, but pre-rendered frames improve the smoothness overall.

 

 

Thanks, Steve.  Can we look forward to the new version of Ideal Flight for P3D?

[okupton 308]

Yea I mean the 780 would be ideal, but its way out of my

*budget was the last word to that last post

[blueshark747 1,843]

I'm running P3D w/FTX Global with high shadows with all features ticked and dense autogen with everything else maxxed.

 

40-65 FPS with unlimited Vsync and Triple Buff fullscreen

 

I don't even bother with an FPS counter anymore because I simply don't need it I'm too busy enjoying the flights and fluid movement.

 

Specs are in the sig

[BMagann 11]

To some extent I think it comes down to where you are flying and in what aircraft. I can get over 50+ fps in some areas with everything cranked up, but slow down to 25 or so in areas where dense cities are present. The choice of plane matters too. The A36 that a lot of people like as the default is not one of the best planes in frame rate terms.

 

But blueshark747 also makes a good point. Turning of the frame counter is often a good thing, because it allows you to just enjoy yourself rather than testing and fretting all the time!

[ricka47 611]

Ricka47, I made exactly that upgrade on Monday. I think you will be pleased with the result!

That's good to hear - thanks!

[SteveW 1,792]

Thanks, Steve.  Can we look forward to the new version of Ideal Flight for P3D?

Yes, thank you Marc! IF10 version[sf] is out shortly (before Christmas) then the proper P3D version comes as soon as possible after.

[Arwen 289]

I have a thread in the screenshot forums that might be of some interest: http://forum.avsim.net/topic/428296-prepar3d-v2-with-ftx-global/

 

My GPU is EGVA's GeForce GTX 780 SC w/ ACX Cooler, which is about 9% faster than the standard GTX 780 (and is quieter and runs cooler), and actually scores higher than a standard Titan in some benchmark tests.  I've also posted screenshots of for all 5 P3D display pages, so you can see exactly what my settings are .

[MarkSC 122]

I have a thread in the screenshot forums that might be of some interest: http://forum.avsim.net/topic/428296-prepar3d-v2-with-ftx-global/

 

My GPU is EGVA's GeForce GTX 780 SC w/ ACX Cooler, which is about 9% faster than the standard GTX 780 (and is quieter and runs cooler), and actually scores higher than a standard Titan in some benchmark tests. I've also posted screenshots of for all 5 P3D display pages, so you can see exactly what my settings are .

I should have mentioned that that is the same card I bought for the benchmarks above.

Yes, thank you Marc! IF10 version[sf] is out shortly (before Christmas) then the proper P3D version comes as soon as possible after.

Great! I think Ideal Flight would be a nice front end for P3D. Can't wait to see it.