Jump to content
chiliwili69

Measuring rig performance: Common Baseline (for IL-2 v3.010)

Recommended Posts

Looking at my system with these settings, I get 7.8 ms frametimes on average and this number is fairly stable, 99% of the time it is within 1 ms. Unfortunately, when multiplied by 2 for VR rendering, that equals 15.6 ms + or - 2 ms.

 

I think that talking about fps or framerates is equivalent. Frequency (fps) is the inverse of the period (framerate) and viceversa. At the end, both measure the number of frames rendered for the test track. So, this test based in fps is equally valid as a pure measure instrument of the power of the CPU/RAM/GPU to make the work. If you divide the number of frames obtained in the tes by 60 (it is 1 min test) you will have your avg (new column V). If you want avg frametime then divide 1000 by avg fps. It doesn´t matter if the frametimes are clustered around 11 and 22, at the end and average is and average, which is FRAMES/TIME.

 

When you say you get 7.8ms you refer to your monitor where you obtained 128 fps avg, since 1000 ms divided by 128 is 7.8ms. So talking about 7.8ms or talking about 128 fps is equivalent.

 

As you said, the reason for what our fps performance in VR is about half of the performance in monitor is because two renders (one per eye) are executed by the graphic card in series. I wonder if one day SLI could work in VR, so we could render one eye in each GPU. Or perhaps even better, future GPUs could be designed specifically just for VR and execute the two renders in paralell (like having two GPU in one).

Edited by chiliwili69

Share this post


Link to post
Share on other sites

Dburne

 

Your card is downclocking as an energy saving feature, because it's not being fully used at those graphics settings. Lot's happening in that track file and your CPU is bottlenecking your card.

 

Normally during single player missions there won't be so many things happening and therefore you will have less of a load on the CPU. Allowing it to feed your Gpu card more.

 

To test my theory you can increase SS while testing that track, this will tax your Gpu card more while not giving your CPU much more to do. Gpu clock will increase as much as is needed. This will not give you any extra frames though. Only better SS.

 

Despite having an overclocked 7700 and a 1070, same thing happens to me in heavy situations.

Share this post


Link to post
Share on other sites

Dburne

 

Your card is downclocking as an energy saving feature, because it's not being fully used at those graphics settings. Lot's happening in that track file and your CPU is bottlenecking your card.

 

Normally during single player missions there won't be so many things happening and therefore you will have less of a load on the CPU. Allowing it to feed your Gpu card more.

 

To test my theory you can increase SS while testing that track, this will tax your Gpu card more while not giving your CPU much more to do. Gpu clock will increase as much as is needed. This will not give you any extra frames though. Only better SS.

 

Despite having an overclocked 7700 and a 1070, same thing happens to me in heavy situations.

 

That certainly makes sense and probably explains it then, thanks for the insight!

 

I think I am not going to focus so much on benchmarking at this point though, if I keep doing so I am going to say screw it and go ahead and do a new build lol. And I really want to hold out until latter part of 2018 to do so, at that point my build will be 5 years old. It has been a rock solid system for me and I am reluctant to replace it just yet. Unless of course something happens unexpectedly beforehand and things go south with this build.

 

And so far I have played the BoS campaign in full, Blazing Steppe , and Ivan's War. With each of those my performance has been quite good, even with ASW set to off. Very smooth and good framerates all around. So for the SP gaming I am doing with this great sim I am still quite pleased with my performance in VR. Just this benchmark had me scratching my head.

Share this post


Link to post
Share on other sites
Just this benchmark had me scratching my head.

 

Yes, it´s more fun to play than do test, but we always want to know the whys of everything.

 

When you finish the campaign, you should know that I have conducted 3 tests leaving the Rift fixed to the chair and  3 tests wearing the rift in my head (so visualizing the test).

 

I obtained a consisten -4fps when It was on my head. I really don´t know the reason for this, but it could be that when I wear the rift and try to look straight forward with the 2 red lights just at the limit of my view at the bottom, I took a screenshoot of the screen monitor (since I can not view the monitor while wearing rift) and saw that the two red ligths were much above than described in procedure. So your test could be affected.

 

Also, if you run the passmark you could get an expectation of the fps for this test.

 

Anyhow, this test is just a measurement instrument but is far below from the fps we normally have in our flights with each one settings.

Edited by chiliwili69

Share this post


Link to post
Share on other sites

Note: I have incorporated to the procedure to press "backspace" to hide in-game fps counter. Since it eats about 1 or 2 fps. Just to be sure everyone do the same.

I have incorporated to the procedure to press "backspace" to hide in-game fps counter. Since it eats about 1 or 2 fps. Just to be sure everyone do the same.

Share this post


Link to post
Share on other sites

Yes, it´s more fun to play than do test, but we always want to know the whys of everything.

 

When you finish the campaign, you should know that I have conducted 3 tests leaving the Rift fixed to the chair and  3 tests wearing the rift in my head (so visualizing the test).

 

I obtained a consisten -4fps when It was on my head. I really don´t know the reason for this, but it could be that when I wear the rift and try to look straight forward with the 2 red lights just at the limit of my view at the bottom, I took a screenshoot of the screen monitor (since I can not view the monitor while wearing rift) and saw that the two red ligths were much above than described in procedure. So your test could be affected.

 

Also, if you run the passmark you could get an expectation of the fps for this test.

 

Anyhow, this test is just a measurement instrument but is far below from the fps we normally have in our flights with each one settings.

 

Just ran the Passmark , my CPU Mark score was 9942.

Share this post


Link to post
Share on other sites

I think that talking about fps or framerates is equivalent. Frequency (fps) is the inverse of the period (framerate) and viceversa. At the end, both measure the number of frames rendered for the test track. So, this test based in fps is equally valid as a pure measure instrument of the power of the CPU/RAM/GPU to make the work. If you divide the number of frames obtained in the tes by 60 (it is 1 min test) you will have your avg (new column V). If you want avg frametime then divide 1000 by avg fps. It doesn´t matter if the frametimes are clustered around 11 and 22, at the end and average is and average, which is FRAMES/TIME.

 

When you say you get 7.8ms you refer to your monitor where you obtained 128 fps avg, since 1000 ms divided by 128 is 7.8ms. So talking about 7.8ms or talking about 128 fps is equivalent.

 

As you said, the reason for what our fps performance in VR is about half of the performance in monitor is because two renders (one per eye) are executed by the graphic card in series. I wonder if one day SLI could work in VR, so we could render one eye in each GPU. Or perhaps even better, future GPUs could be designed specifically just for VR and execute the two renders in paralell (like having two GPU in one).

 

FPS and frame times are equivalent of course, but fraps gives very detailed frame time results with only minimal fps results. So I think looking at the frame time data is more informative. You'd never notice that step function in frame rates if all you looked at are the two fps files. In any case, the real issue is CPU usage. The GPU isn't particularly stressed. A better solution than what we have would be for one cpu thread to feed one eye and for a second to feed the other. CPU bottlenecking is not an issue at all on a 2d screen, so doing that would theoretically give similar results. Of course, the ideal solution is for the game to use n CPU threads and spread the load equally across them, but I think DX11 only supports dual threaded graphics engines. Obviously physics, sounds and other game logic could be on separate cpu threads, which would also help but the physics and AI are fairly trivial for the CPU compared to the graphics. Some performance can be gleaned by moving them off the rendering thread, but splitting up the rendering thread is the best option.

 

Edit: That 2215 single thread stat might actually have been me. When I first built the new system I tinkered pretty hard with overclocking and did get some 2200+ results in single thread, but it wasn't stable. Last night I did a 4.1 GHz test in BoS for monitor results and gained about 8-9 fps. It didn't crash, but voltage was in a place I wasn't comfortable with. Not something I plan to use, was just curious what it would do.

Edited by BeastyBaiter

Share this post


Link to post
Share on other sites

 

 

ust ran the Passmark , my CPU Mark score was 9942.

 

And the Single Thread Mark?, is subtest of the CPU Mark. See post 1. 

Share this post


Link to post
Share on other sites

And the Single Thread Mark?, is subtest of the CPU Mark. See post 1. 

 

CPU Single Threaded = 2492

Share this post


Link to post
Share on other sites

Just curious to know what change gave you the fps increase, maybe both but not sure really. But you had a great increase for that small change, about 8 fps!

 

Did you check if your CPU_Mark and ST_MArk changed after RAM change?  

 

Yes, I did. The results were the same as before OC.

 

If you have a good cooler you could try more OC. Every 0.1 Ghz increase you could gain between 1.5 to 4 fps. (it depends how close you are to 90 fps, the more closer to 90 the less increase you will have) 

 

It's already overclocked to its max with air cooling.

Share this post


Link to post
Share on other sites

 

 

CPU Single Threaded = 2492

 

OK, thanks, with that STMark value we can make so guess about the expected performance with the test.

A STMark of 2492 should give about 67 fps, but you obtained 52.4 fps, about 15 fps less than expected for you CPU.

The reason for that could be:

- Your RAM speed (2133MHz is not low but not top either)

- The way you did the test (I think you used your head, looking the two red indicators at the bottom through the rift, but in the monitor screen they appear higher than required in procedure)

- Other issues with Hardware/software/settings

 

I would say that RAM for sure contribute to that -15 fps. Why?... Well take a look of the 71st_AH_Statuskuo tests with an i5-6600K.

He gained about +6fps (54.7 to 61) when going from 2133 to 3200 memory upgrade.

And later, he OC 0.2Ghz more and he gained 6-7 fps more, reaching 67.7 (the expected fps for him is 67.3!)

 

The interesting thing is that his CPU (@ 4.4) has a STMark of 2498, almost the same than your CPU. But he is running with DDR4 3200MHz RAM.

 

You run DDR3, so it wold be difficult to find DDR3 at those speeds. I have 2933MHz RAM but it was not easy to find it (neither cheap!).

 

So, it you are planning a new rig for next year it would pearhaps no worth to invest in RAM for a few extra fps, but I least I wanted to let you know the most likely reason of your low performance with the test (which again, it is just a stress benchmark, not the fps we normally have with playing)

Share this post


Link to post
Share on other sites

 

 

Yes, I did. The results were the same as before OC.

 

Thanks, as expected. RAM doesn´t affects STMark, just measure CPU one core power.

 

STMark is a combination of the Floating Point Test, Sorting Test and Compression test and they might not access the RAM, but cache.

Share this post


Link to post
Share on other sites

 

 

So, it you are planning a new rig for next year it would pearhaps no worth to invest in RAM for a few extra fps, but I least I wanted to let you know the most likely reason of your low performance with the test (which again, it is just a stress benchmark, not the fps we normally have with playing)

 

Yeah as I will be doing a new build in likely less than a year, and I am getting very smooth performance playing in SP, I am not going to invest more into this current rig. 

 

I still think in addition, the fact that my GPU was only running around 1500 MHz rather than full boost speed of right at 2 GHz as it normally does  played a factor in it as well. 

 

Thanks,

Share this post


Link to post
Share on other sites

Chiliwili, please consider adding to the initial test instruction that people make sure to disable their AVX offset. All newer boards as well as older ones with bios upgraded import AVX throttling by default, which must be turned off in order to not have the system throttle to a fixed (lower) clock when the track starts.

Share this post


Link to post
Share on other sites

 

 

So I think looking at the frame time data is more informative. You'd never notice that step function in frame rates if all you looked at are the two fps files.

 

For the purposes of this benchmark the frametime data is really not need. So, don´t activate the Frametime checkbox in Fraps it when running the test.

If someone has strange results or lower than expected results, then they can activate it in Fraps and take a look of the frametimes, just to check how they are produced.

Share this post


Link to post
Share on other sites
The Rift can display 22.5, 45 and 90 fps only.

 

Well, just to be more precise, the Rift physical display has a 90Hz constant refresh frequency that can not be changed via software. So, it will always display 90 images per second.

 

Another matter is that some of those images (or frames) can be repeated or not generated by a real render but synthetically generated (ASW).

All displays (monitor and VR devices) has a small RAM called "buffer". The Rift integrated circuit is constantly dumping at 90Hz whatever is in the buffer´s display, it doesn´t matter if has been refreshed by the GPU or not. So, if it is not refresehed, it will repeat the same image for two cycles, and this is noticeable in VR if you move your head. (it depends on how frequent it is)

 

1.- RENDER TIME CONTINUOSLY BELOW 11ms: Then everything is fine, your frame period is 11 ms and you have 90fps (ie 90 rendered frames per second).

2.- RENDER TIME AROUND 11ms (sometimes below, sometimes above): Then a frame is skipped (or dropped) and you wait until next cycle (11+11=22ms) If this only happen once in a second your avg fps will be 89. If happen twice in a second your average will be 88, and so on.

3.- RENDER TIME CONTINUOUSLY ABOVE 11ms: Then you will skip then two cycles (11+11=22ms), so you render period is 22ms in a continuous way. You 45 drops per second, so 45 fps.

 

The three casuistic mentioned above are shown in the below graph (I put frameperiod and fps in the same scale):

post-18865-0-65519300-1504453316_thumb.jpg

When frame period is 11 ms you have solid 90fps,

when you have dropped frames you have a mix 11ms and 22ms frame period, then fps is a value between 90 and 45

When frame period is constantly 22 ms you have 45fps.

 

Here it is shown a zoom for a period of time where you have a mix of 11ms and 22ms frame period:

post-18865-0-93317700-1504453335_thumb.jpg

 

Someone was asking before why 44-45 fps is very frequent as min fps reported by Fraps for medium range rigs.(mine for example)

The explanation is easy.

If your render time is continuously equal to 11.5 ms your reported fps is 45 fps.

If your render time is continuously equal to 12 ms your reported fps is 45 fps.

If your render time is continuously equal to 13 ms your reported fps is 45 fps.

.....

If your render time is continuously equal to 20 ms your reported fps is 45 fps.

If your render time is continuously equal to 21 ms your reported fps is 45 fps.

If your render time starts to be above 22.22 ms for some frames, then you will drop 2 frames per cycle and your reported fps will be below 45 fps.

 

So, that´s why 45fps appear as a wall (or discontinuity region) when you look at the fps profile along that 60 second test. This is clearly shown when I was making the SS experiment:

post-18865-0-07983600-1504453616_thumb.png

Going below 45fps should be avoided always in VR.

Edited by chiliwili69

Share this post


Link to post
Share on other sites

 

 

Chiliwili, please consider adding to the initial test instruction that people make sure to disable their AVX offset. All newer boards as well as older ones with bios upgraded import AVX throttling by default, which must be turned off in order to not have the system throttle to a fixed (lower) clock when the track starts.

 

Yes, I have added that. I have to verify this in my PC. I don´t know how much this affects to temperatures.

Share this post


Link to post
Share on other sites

 

Going below 45fps should be avoided always in VR.

If you render time is continuously equal to 12 ms your reported fps is 45 fps.

 

Exactly, that's the point I was getting at. You went into much more detail than I did though. On the topic of RAM, I am wondering if I should swap RAM kits. It would cost $180-190 USD to go from my 2800 MHz to 3200+ but BoS and Ryzen absolutely care about RAM speed.

Share this post


Link to post
Share on other sites

 

 

On the topic of RAM, I am wondering if I should swap RAM kits. It would cost $180-190 USD to go from my 2800 MHz to 3200+ but BoS and Ryzen absolutely care about RAM speed.

 

Uhm.. very difficult question.

As you know, I upgraded my RAM from 1866 to 2933 and had an increase of 5 fps.

In your case you are going from 2800 to 3200, it is a moderate upgrade and it perhaps would not deliver the extra fps you expect. May be not worth 190$.

You are already at good RAM speed.

 

Ideally you could borrow from someone and try before buying that.

 

I also don´t know if IL-2 is your prime game or you spread your game time with other or use the rig for other application which really benefit of the nice power of your CPU in multithread.

 

If IL-2 is your loved baby and play in VR a considerable amount of time, I would consider sell your Ryzen (you currently will get a good deal) and take a new/second hand CPU/Mobo with high STMark scores and good OCing capabilities.

  • Upvote 1

Share this post


Link to post
Share on other sites

I went with Ryzen over the 7700k because I wanted programming experience on a relatively high thread count computer rather than the same old quad cores we've been using for the last 10 years. So selling it is pretty much out of the question. I do still have this partially cannibalized I5-4690 + Z87 + RX 480 8GB system sitting here but an I7-4790k is almost as expensive as a 7700k for some reason. The particular motherboard also only supports around 800 MHz ram or whatever the absolute lowest speed is. It has a 2400MHz kit in it and the board theoretically supports it, but I never got anywhere with it. It's a crappy board tbh and was one of the main reasons I went ahead and built a new PC a few months ago instead of waiting for Intel's response to Ryzen before deciding.

Edited by BeastyBaiter

Share this post


Link to post
Share on other sites

 

 

I do still have this partially cannibalized I5-4690 + Z87 + RX 480 8GB system

 

In that case you could try to make it work the RAM kit to 2400 and it successful you can enable the Turbo  to 3.9 and put the 1080Ti card on it just to test it.

 

If not, a 2nd hand (or new )newer Mobo for the i5-4690 could work and it should not cost too much.

 

If the above run decently, then you might consider 2nd hand 4790K (from time to time there are bargains in the web and you are not in a hurry) and a good cooler.

 

Yes, the old 4790K is still surprisingly pretty expensive.

Share this post


Link to post
Share on other sites

I went with Ryzen over the 7700k because I wanted programming experience on a relatively high thread count computer rather than the same old quad cores we've been using for the last 10 years. So selling it is pretty much out of the question. I do still have this partially cannibalized I5-4690 + Z87 + RX 480 8GB system sitting here but an I7-4790k is almost as expensive as a 7700k for some reason. The particular motherboard also only supports around 800 MHz ram or whatever the absolute lowest speed is. It has a 2400MHz kit in it and the board theoretically supports it, but I never got anywhere with it. It's a crappy board tbh and was one of the main reasons I went ahead and built a new PC a few months ago instead of waiting for Intel's response to Ryzen before deciding.

Hi there. I had a 1700 at 3.8GHz and 3ghz ran and I had to have graphics on low and aa at 2 for it to be playable. It's not bad but to get full performance you need a high end CPU for vr in il2. I switched to 7700k and even without overclock I have solid 90 on high graphics and aa4.

 

I wouldn't spend anything on ram unless you think you can get really high speed, like 4 GHz which might possibly give you a boost. Who knows how much though.

Share this post


Link to post
Share on other sites

 

 

I switched to 7700k and even without overclock I have solid 90 on high graphics and aa4.

 

Glad to hear you switched to 7700K!, Did you keep the 1070? Let us know your test performance with your new rig. 

Share this post


Link to post
Share on other sites

Glad to hear you switched to 7700K!, Did you keep the 1070? Let us know your test performance with your new rig. 

Managed to exchange the Ryzen so wasn't to bad a financial hit. I figured my results would be same as the others with 7700k, 1070 and 3ghz RAM but will do another test at some stage to see.

Share this post


Link to post
Share on other sites

I threw the GTX 1080 TI into my old system and gave it a look as suggested. Full system specs here:

CPU: I5-4690 (3.5-3.9GHz)

GPU: GTX 1080 TI

RAM: 16GB 1333 MHz DDR3

OS: Win 10

 

1080p result:

Min: 90
Max: 156
AVG: 116
 
Rift result:
DNF (BoS consistently crashed halfway through)
 
Aside from it having 10-20 fps less than the R5 1600x for all 3 fps results, the passmark scores are also interesting. Below is a comparison between the two systems. I've bolded the better result for each. The 3d mark results are very unexpected but also highly reproducible. I ran it over and over and kept getting the same result on the I5 system. I have previous benches to compare to for on the Ryzen system, and although that newest one is on the higher side, it's consistent with the rest. Both systems have fully up to date drivers.
 
                 I5-4690           R5 1600x
                      CPU RESULTS
Overall:      8489              14278
Int:              9873               35496
FP:             8886                13011
Prime:        34                    38
SSE:          330                  626
Comp:       9637                21938
Enc:          1483                3462
Phys:         468                 666
Sort:          6035               13096
Single:       2341               2135
 
                       3d mark
Overall:    12969              15801
DX9:         196                   222
DX10:       329                   319
DX11:       339                   343
DX12:       47                     73
Comp:      10168           10572
 
                      Memory
Overall:     2529             2076
DB:            97                88
cached:     28271           29124
!cached:   15238            16382
write:         8902             8314
Amount:    14649           14243
Latency:    27                 64
Threaded: 19002           37978

 

R5 system specs:

CPU: R5 1600x at 4.0 GHz

GPU: same physical card

RAM: 16GB DDR4 @ 3200 MHz

OS: Win 10

 

R5 results for BoS already posted.

Edited by BeastyBaiter

Share this post


Link to post
Share on other sites
I threw the GTX 1080 TI into my old system and gave it a look as suggested.

 

I think your old system has 3 weak points for IL-2:

 

1.- No OC capability (beyond Turbo 3.9GHZ). Many other intel K or X processors benefit from that.

2.- RAM speed. 1333 is too low. A 3000MHz will help a lot in monitor and VR

3.- MoBo. Don´t know the model but you said it was 800MHZ for RAM. There could be other bottlenecks in the Mobo that we don´t see.

 

So, I agree with you, the old system, as it is, is a non-go.

 

Apart from this, what it surprise me more, is that you have better performance in the three sub-test which is based the STMark( Floating Point Test, Sorting Test and Compression test) but you have worse STMark!!

 

                          i5-4690           R5 1600x

 
FloatingP.:         8886                13011
Compress.:       9637                21938
Sort:                  6035                13096
SingleThread:   2341                  2135
 
Edited 7-Sept-17: I asked to Passmark and they reply me: "All tests except Single threaded test run on all available cores (by default). So the scores are naturally higher. You can't really compare the result from the single threaded test to the numbers from the other test."
So, for the STMark, they run a combination of the FP, Comp. and sort test using only one core. Bottom line: STMark is what matter for BOS.
Edited by chiliwili69

Share this post


Link to post
Share on other sites

I knew it was a low RAM speed, thought it was 800 MHz but I verified it was running at 1333. The motherboard is a Biostar Hi-Fi Z87X 3d. The big point is that despite having a lower ST rating, Ryzen is clearly faster in game. It could be the RAM speed of course, but there are also the GPU results too. I looked into it a bit before taking it back apart and what I found is a heavy CPU bottleneck for the 3d part that didn't exist on the Ryzen system. It's also interesting to see the difference in RAM performance. It's well known that Ryzen has some serious latency issues with RAM but the overall RAM performance is very similar between the two on most points.

Share this post


Link to post
Share on other sites
With the benchmark settings@1440p :

 

Pre-rendered frames 1

2017-09-04 00:22:22 - Il-2 

Frames: 9329 - Time: 63344ms - Avg: 147.275 - Min: 116 - Max: 193

 

pre-rendered frames, default.

2017-09-04 00:24:58 - Il-2

Frames: 9076 - Time: 61922ms - Avg: 146.571 - Min: 118 - Max: 194

Edited by icecream

Share this post


Link to post
Share on other sites

Dropping shadows down to medium almost brings the FPS back inline. High and Ultra lead to fps issues almost like the 1080ti was running out of memory perhaps. No data, just guessing.

That's a negative.

 

The only game that draws my 11GB out of my vram is joinsqaud at ultra with 2.0 times supersampling , on a few maps.

Haven't seen any other game's use my 11GB of vram out.

Edited by icecream
  • Upvote 1

Share this post


Link to post
Share on other sites

Thanks for checking. I still hadn't got around to cranking it back up to ultra/ultra and looking.

 

I wonder what was going so badly then. It was a constant stutter, even just moving my view around the cockpit. Performance felt worse than just being stuck in re-projection. Hopefully I'll get some time to test eventually but right now I try to fly online with my limited free time.

Share this post


Link to post
Share on other sites
2017-09-03 22:18:26 - Il-2

Frames: 4278 - Time: 60000ms - Avg: 71.300 - Min: 44 - Max: 91  (High, Shadows 2, ss 3.2)

 

2017-09-04 22:22:35 - Il-2

Frames: 4017 - Time: 60000ms - Avg: 66.950 - Min: 44 - Max: 91  (High, Shadows 3, ss 3.2)

 

2017-09-04 22:20:24 - Il-2

Frames: 3286 - Time: 60000ms - Avg: 54.767 - Min: 44 - Max: 78  (High, Shadows 4, ss 3.2)

 

Will redo these tests with ss 1.0 later. To see how much fps are lost then.

Share this post


Link to post
Share on other sites
2017-09-03 22:18:26 - Il-2 Frames: 4278 - Time: 60000ms - Avg: 71.300 - Min: 44 - Max: 91  (High, Shadows 2, ss 3.2)

 

Interesting.Thanks. I have updated them in the Graph Settings tab tests.

You tested the same than pantercules above but with 3.2 SS instead of 1.7. So, you started to test shadows from a heavy loaded GPU and the fps drops earlier then.

 

Did you run those tests @4.9 OC? If not let me know the OC and also the CPU Mark and STMark for your OC freq.

 

If you use other stable OC you can also run the test with just the VR settings of the test with new version 2.012. Just for reference in main page..

Edited by chiliwili69

Share this post


Link to post
Share on other sites

I ran them at 4.8GHz, LLC 2, Cache 4.4GHz, AVX offset 0, 1.330Vcore.

 

The GTX 1080 OCed to 1840MHz core / 5039MHz VRAM,  boosts to around 2040MHz core if I remember correctly, power target 120%.

 

 

 

Since it seems impossible to run the game 2 hours continuously under heavy load at 4.9GHz or 5.0GHz without an AVX offset. Sometimes it runs 20 minutes, sometimes 60 minutes, but there is always a freeze within 2 hours - which we want to avoid.

 

 

To Panther: I believe you got slower results because the AVX offset might be active, so it actually does not run at 5.0GHz as soon as the balapan test track starts, but at 4.6GHz (Intel's clearance) instead. You might want to check on that, and whether you have deactivated the AVX offset manually. Also, what is your cache OC and LLC? If you found out that to be true and deactivate the AVX offset, please make sure to rise the normal clock slowly from 4.6 or 4.7 again, don't start out at 5.0 by all means. 

 

 

I think my initial test, rated at 4.9GHz that topped in the 2.011 summary, was actually just a sneaky 4.7GHz because of the AVX offset! The GPU was not overclocked btw. I do not remember the AVX offset precisely, but will run new tests for 2.012. I assume more than half of the test results were not run at the clock speeds that people reported, unknowingly to them.

Edited by 2./JG51_Fenris_Wolf

Share this post


Link to post
Share on other sites

To Panther: I believe you got slower results because the AVX offset might be active, so it actually does not run at 5.0GHz as soon as the balapan test track starts, but at 4.6GHz (Intel's clearance) instead. You might want to check on that, and whether you have deactivated the AVX offset manually. Also, what is your cache OC and LLC? If you found out that to be true and deactivate the AVX offset, please make sure to rise the normal clock slowly from 4.6 or 4.7 again, don't start out at 5.0 by all means. 

 

I've been wondering about that myself recently.  I've always had the AVX offset at 0 (didn't do it manually - it was just that way normally in my bios), but because of some fan spiking I was getting at 5.0 and your comments about the 7700k not liking 5.0 in AVX for extended periods I backed off to 4.6 a few days ago to see if it made a difference. It did seem to help with the fan spiking issue. Then in running a few tests regarding OVRDrop yesterday I started seeing results that were as good or better at 4.6 than I had been seeing with 5.0 last week.  Unfortunately, due to some RL issues I won't be able to run any more tests for a few days, but I am beginning to think that my earlier 5.0 test results may have been somewhat suspect.

 

Is there some way to see a report/chart or something about what the CPU has been doing while these tests were running?  The only things I've found in CPU-Z or the ASUS utility that came with my MB seem to be real time, while you're watching kinds of things that don't seem to generate a retrievable record of the past few minutes, but it may just be that I don't know where to look for it.

Share this post


Link to post
Share on other sites

If you are looking for things such as thermal throttling, you might want to try H/W monitor, start it up, clear min/max, then after the testrun watch the average. If average clock != maximum clock, you had some thermal throttling going on in the background. You can also log your findings for more extensive details. AIDA64 Extreme (trial) is also a good program, more powerful, but more complex to get into

Share this post


Link to post
Share on other sites
New results from balapan test track, in order to compare the effect of Load Line Calibration:


2017-09-06 14:38:44 - Il-2

Frames: 4643 - Time: 60000ms - Avg: 77.383 - Min: 46 - Max: 91  (4.8, avx 0, auto voltage, llc 2)

 

2017-09-06 14:50:26 - Il-2

Frames: 4622 - Time: 60000ms - Avg: 77.033 - Min: 46 - Max: 91  (4.8, avx 0, auto voltage, llc 2)

 

2017-09-06 14:54:09 - Il-2

Frames: 4627 - Time: 60000ms - Avg: 77.117 - Min: 45 - Max: 91  (4.8, avx 0, auto voltage, llc 2)

 

2017-09-06 15:02:00 - Il-2

Frames: 4758 - Time: 60000ms - Avg: 79.300 - Min: 47 - Max: 91  (4.8, avx 0, voltage 1.330, llc 1)

 

2017-09-06 15:03:18 - Il-2

Frames: 4750 - Time: 60000ms - Avg: 79.167 - Min: 46 - Max: 91  (4.8, avx 0, voltage 1.330, llc 1)

 

2017-09-06 15:15:03 - Il-2

Frames: 4819 - Time: 60000ms - Avg: 80.317 - Min: 44 - Max: 91  (4.9, avx 0, voltage 1.330, llc 2)

 

2017-09-06 15:16:26 - Il-2

Frames: 4840 - Time: 60000ms - Avg: 80.667 - Min: 47 - Max: 91  (4.9, avx 0, voltage 1.330, llc 2)

 

2017-09-06 15:27:17 - Il-2

Frames: 4851 - Time: 60000ms - Avg: 80.850 - Min: 49 - Max: 91  (4.9, avx 0, voltage 1.330, llc 1, cache 4.4)

 

2017-09-06 15:28:42 - Il-2

Frames: 4949 - Time: 60000ms - Avg: 82.483 - Min: 48 - Max: 91  (4.9, avx 0, voltage 1.330, llc 1, cache 4.4)

 

2017-09-06 15:44:33 - Il-2

Frames: 4906 - Time: 60000ms - Avg: 81.767 - Min: 50 - Max: 91  (4.9, avx 0, voltage 1.330, llc 1, cache 4.5)

 

2017-09-06 15:45:59 - Il-2

Frames: 4852 - Time: 60000ms - Avg: 80.867 - Min: 48 - Max: 91  (4.9, avx 0, voltage 1.330, llc 1, cache 4.5)

 

 

 

 

 

 

Patch 2.012 sure worsened some performance? No!! I found the (beautiful but sadly performance cutting!) culprit. 

 

I removed "3DMigoto mod for VR", which is a injected shader to increase contrast.

 

Now, low and behold, see the comparison:

 


2017-09-06 16:28:26 - Il-2

Frames: 4372 - Time: 60000ms - Avg: 72.867 - Min: 44 - Max: 91  (4.6, avx 0, voltage 1.260, llc 1, cache 4.4)

 

2017-09-06 16:29:56 - Il-2

Frames: 4473 - Time: 60000ms - Avg: 74.550 - Min: 45 - Max: 91  (4.6, avx 0, voltage 1.260, llc 1, cache 4.4)

 

2017-09-06 16:33:12 - Il-2

Frames: 5011 - Time: 60000ms - Avg: 83.517 - Min: 51 - Max: 91  (4.6, avx 0, voltage 1.260, llc 1, cache 4.4) 3DMigoto shader disabled!

 

2017-09-06 16:34:30 - Il-2

Frames: 5010 - Time: 60000ms - Avg: 83.500 - Min: 51 - Max: 91  (4.6, avx 0, voltage 1.260, llc 1, cache 4.4) 3DMigoto shader disabled!


Edited by 2./JG51_Fenris_Wolf

Share this post


Link to post
Share on other sites

My final test results for the chart comparison 2.012:

 

CPU Passmark 13617
CPU Passmark ST 2863
 
2017-09-06 16:49:54 - Il-2
Frames: 5213 - Time: 60000ms - Avg: 86.883 - Min: 60 - Max: 91
 
 
i7 7700K
4.9GHz core, 4.4GHz cache, AVX offset 0, voltage 1.330, LLC 1
ASRock Z270 Fatal1ty K6
16GB GSkill DDR4   3200MHz
Gainward Phoenix GS GTX1080 @boost 2050MHz Core,  5039MHz VRAM
Edited by 2./JG51_Fenris_Wolf

Share this post


Link to post
Share on other sites

Ok - decided to run a new batch of tests after the new hotfix.  Here are results so far (will keep adding results to this post as I complete more tests this afternoon).  

 

 

Intel Core i7-7700K @ 4.6 GHz (Corsair liquid cooler); other settings stock/auto
16GB RAM (Corsair Vengeance DDR4 3200MHz) @ 3200MHz
EVGA GTX 1080 TI SC2, 11 GB (all settings stock); ASUS ROG Maximus IX Code MB; Realtek ROG SupremeFX audio

Windows 10 Pro 64-bit; Oculus Rift (CV1); Logitech Force 3D Pro; Saitek Rudder Pedals and Throttle Quadrants

Passmark CPU mark = 13118; ST = 2688

IL-2 v.2012c; 1st post VR test settings (SS=1.7 via SteamVR):  Frames: 5040 - Time: 60000ms - Avg: 84.000 - Min: 55 - Max: 91

Edited by TG-55Panthercules

Share this post


Link to post
Share on other sites

I made a few tests this afternoon, as you've seen. Many more than I posted.

 

I think your configuration will profit the most from going 4.7 or 4.8GHz (maximum, don't recommend going higher, it gives maybe 1/2 fps more, aim for stability), auto vcore, LLC1, AVX offset 0, then play on Ultra / terrain 4x / AA 0 / sharpen 0 / terrain_sharpen 1 / shadows 2 (medium) / SteamVR SS 2.2. 5.0GHz was damn high, my Computer doesn't even boot at that range :blink:

 

Average fps will hit around 67-70 on balapan test track. In less complex scenes, it should stay at 90.

Edited by 2./JG51_Fenris_Wolf

Share this post


Link to post
Share on other sites

I made a few tests this afternoon, as you've seen. Many more than I posted.

 

I think your configuration will profit the most from going 4.7 or 4.8GHz (maximum, don't recommend going higher, it gives maybe 1/2 fps more, aim for stability), auto vcore, LLC1, AVX offset 0, then play on Ultra / terrain 4x / AA 0 / sharpen 0 / terrain_sharpen 1 / shadows 2 (medium) / SteamVR SS 2.2. 5.0GHz was damn high, my Computer doesn't even boot at that range :blink:

 

Average fps will hit around 67-70 on balapan test track. In less complex scenes, it should stay at 90.

 

Thanks - my normal, mostly maxed settings now are close to what you stated, except I use SS=2.3 via SteamVR (something of a cliff somewhere between 2.3 and 2.6 on my rig) and I use shadows = 3 (high) for the eye candy of the new shadows (and everything else pretty much maxed out because none of the other settings seemed to have much impact on FPS).  At my current 4.6 OC that gives me the following on the benchmark track (with 2.012c):

 

Frames: 3394 - Time: 60000ms - Avg: 56.567 - Min: 44 - Max: 91  

 

Dropping shadows to 2 (medium) with those settings yields better results but then you lose some of the benefits of the new shadows, that seem to only appear at high or ultra settings) - I'm still debating (as I run various tests) about whether those shadow eye candy benefits are worth the price in FPS, but so far I think they are.  If I start running into action-heavy situations where the FPS seems to be an issue, that's likely to be the first thing I jettison since there's a pretty big hit going from medium to high shadows.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×
×
  • Create New...