Benchmark for CPU/RAM performance: Remagen 4.002 to 4.005

[chiliwili69]

7 hours ago, =[TIA]=Stoopy said:

2020-01-27 19:51:28 - Il-2
Frames: 18073 - Time: 180000ms - Avg: 100.406 - Min: 68 - Max: 190

 

Not too shabby for a little BIOS correction, and I really do appreciate you catching that, @chiliwili69!

 

That´s nice. You know what is the Greif expected fps for your PC with RAM at 2133MHz???  it is 100.9!!  so you PC delivers as it should.

Enjoy your Reverb, it has a very nice resolution.

[JG1_Greif]

On 1/27/2020 at 9:21 PM, chiliwili69 said:

Going to that 2400MHz  (check your Mobo can support it) you will gain about 16 fps in monitor.

 

@chiliwili69I know this probably only irks data nerds like me, but because of statistics (blah blah blah, descriptive statistics vs. predictive analytics, small sample size etc. etc.), but more important, because of managing expectations, I would advise to use other wording: "Going to that 2400MHz  (check your Mobo can support it) you should gain about 16 fps in monitor".

 

 

On 1/28/2020 at 9:29 AM, chiliwili69 said:

That´s nice. You know what is the Greif expected fps for your PC with RAM at 2133MHz???  it is 100.9!!  so you PC delivers as it should.

 

 

... nonetheless, this is very cool and satisfying. Really nice to see that the model helps identifying possible bottlenecks in people's configuration!

 

Too bad about the 0.496 deviation between the predicted and observed FPS ? 

Edited January 29, 2020 by JG1_G_Greif

[WallterScott]

has anyone noticed that in version 4.04 fps online has become better? There are still stutters, but they are short and less frequent.

[chiliwili69]

2 hours ago, JG1_G_Greif said:

because of managing expectations, I would advise to use other wording: "Going to that 2400MHz  (check your Mobo can support it) you should gain about 16 fps in monitor".

Yeap, fully agree. I should have used the "should" rather than "will". Failed expectations bring frustration.

[Alonzo]

14 hours ago, WallterScott said:

has anyone noticed that in version 4.04 fps online has become better? There are still stutters, but they are short and less frequent.

 

Yes. The developers fixed something to do with texture caching, which was causing problems in multiplayer. My theory is that in MP there were lots of high resolution aircraft textures required and they were somehow thrashing the texture cache, causing micro stutters as textures were loaded from main RAM into the GPU. They've fixed this, possibly by simply doing a better job caching textures, possibly by reducing texture detail for 'far away' planes or something else.

 

Either way multiplayer is much much better since the patch.

[SvAF/F16_Goblin]

Hello everyone,

I stumbled upon a DCS VR tweaking video that holds some general tweaks that might improve the fps for IL-2 as well.

The tweaks I've applied is the Nvidia graphics settings and the compatibility tweaks for windows. Havn't had the time to redo the benchmark but purely subjectivly I find it to be smoother but it could be placebo effect. I found the info about w10 monitor handling pretty interesting and the option to disable it through the compatibility options found under properties for the executable.

 

 

 

Edited January 31, 2020 by Goblin

[jarg1]

13 hours ago, Goblin said:

I found the info about w10 monitor handling pretty interesting and the option to disable it through the compatibility options found under properties for the executable.

 

Disabling full screen optimization seems like something worth looking into, but I wonder if it would make a difference with VR?

[SvAF/F16_Goblin]

Can't tell ? placebo is good for the feeling in anycase

[jarg1]

Motherboard: ASUSTek ROG MAXIMUS XI HERO (WI-FI)
 CPU:                 Intel Core i9 9900KS
 CPU Freq:        5.5 Ghz
 L3 cache:        16 MB
 Cores:               8  (number of active physical Cores)
 HT:                    Off (HyperThreading. If Cores=Threads then HT is Off)
 RAM type:        DDR4
 RAM size:        32 GB
 NB Freq:           5100 MHz
 RAM Freq:        4133 MHz (this is 2xDRAM freq if Dual channel)
 RAM Latency: 16  (CAS Latency CL)
 GPU:                 2080Ti
 STMark            3330

 

2020-02-03 21:41:49 - Il-2
Frames: 31303 - Time: 180000ms - Avg: 173.906 - Min: 119 - Max: 287

 

I ran this just for fun.  The results aren't as good as you might expect, but I'm getting lower frame rates since the last update.

[chiliwili69]

1 hour ago, jarg1 said:

 CPU:                 Intel Core i9 9900KS
 CPU Freq:        5.5 Ghz

 

WOOW!!  5.5GHz! that´s really an achievement.  The record for STMark in the Passmark database for 9900KS is 3270, but you exceed that!  You really got a good chip!

Just curious, what Vcore and temps do you have? and what CPU cooling?

 

Your performance is exactly as expected by the Greif correlation, just 0.0 deviation from expected!

[apollon01]

8 hours ago, jarg1 said:

CPU Freq:        5.5 Ghz

???

 

I should stop following this thread ?

[jarg1]

 

6 hours ago, chiliwili69 said:

 

WOOW!!  5.5GHz! that´s really an achievement.  The record for STMark in the Passmark database for 9900KS is 3270, but you exceed that!  You really got a good chip!

Just curious, what Vcore and temps do you have? and what CPU cooling?

 

Your performance is exactly as expected by the Greif correlation, just 0.0 deviation from expected!

 

For this run I had the voltage at 1.31.  I don't see very high temperatures on your benchmark.  Anyway, 5.4 is stable at 1.30 but I can't run 5.5 for long at that voltage.  I'm going to see if I can get it stable at a higher voltage so I can finish a run on something like Time Spy where I am getting to 97C .  Cooling is custom loop using a 560mm radiator with 8 x140mm fans in push/pull.  CPU and GPU are using water blocks though motherboard and memory are not.

 

 

 

[Alonzo]

9 hours ago, jarg1 said:

 

WOW! Nice rig. Do you think I can get anywhere close as a mere mortal, 8086K, and a case that will do at most 360mm AIO / 280mm AIO (and which is better? 360?).

Edited February 5, 2020 by Alonzo

[jarg1]

6 minutes ago, Alonzo said:

 

WOW! Nice rig. Do you think I can get anywhere close as a mere mortal, 8086K, and a case that will do at most 360mm AIO / 280mm AIO (and which is better? 360?).

 

Thanks.  I'm a mere mortal who has an slightly expensive hobby!  I'm using a Thermaltake Core P5 case and really like the open layout though it is pretty loud when the fans are at full speed.  It is designed to hold up to a 480mm radiator, but as you can see I was able to squeeze in the larger size.   It has a tempered glass cover but lately it is rarely mounted as I've been tweaking the components, for example changing the radiator orientation to reduce back pressure on the fans.  I'm contemplating replacing the case with a Core P7 and using 2 water loops but I may wait until the next generation Nvidia cards are released.

 

If you are asking me if a 360mm radiator is better than a 280mm, the answer would be all other things being equal the greater the cooling area the better.  My last build used a 360mm AIO and was fine for moderate overclocking.  I'm guessing you could get to 5 or 5.1 GHz on your 8086K.

[chiliwili69]

17 hours ago, jarg1 said:

Cooling is custom loop using a 560mm radiator with 8 x140mm fans in push/pull

 

OMG!! I didn´t even know this exist.  

if you attach those fans to your seat you don´t even need simshaker!  ? 

 

My future CPU (this year) will be either a 9900KS or the 10900K. And I will try to keep my modest 240mm AIO cooler (Kraken X-52 which is working quite well).

In terms of heat generation for the same overclock I think the 10900K will generate more heat since it has 2 more cores, but who knows. Anyhow, this is discussed in another thread.

Edited February 5, 2020 by chiliwili69

[Alonzo]

10 hours ago, chiliwili69 said:

My future CPU (this year) will be either a 9900KS or the 10900K. And I will try to keep my modest 240mm AIO cooler (Kraken X-52 which is working quite well).

In terms of heat generation for the same overclock I think the 10900K will generate more heat since it has 2 more cores, but who knows. Anyhow, this is discussed in another thread.

 

I'm very annoyed right now that Intel is forcing complete system upgrades while AMD users are happily resocketing their motherboards with the newest chips. I really hope the improved competition forced Intel to stop being such scumbags -- AMD have gotten what, 3 or 4 CPU generations out of a single socket?

[ZachariasX]

31 minutes ago, Alonzo said:

I really hope the improved competition forced Intel to stop being such scumbags

If they stopped being that, I mean what would be left of them? All you could do is turn off the light.

 

See, if your CPUs are not competitive, you can make someone making your competition less competitive... And if you think "so what, I bought Intel anyway, they won't further hurt *me*", fear not, they do. They have the next downgrade for your CPU performance ready. It will not be the last.

 

All of this would be somewhat bearable, but since it is largely industry standard (Intel is just better at all that) we're stuck with that.

[JG1_Greif]

On 1/7/2020 at 11:54 PM, chiliwili69 said:

Regarding the RAM, I would go for just 16Gb, you don´t need more for IL-2 VR.

But look also for the lowest true latency. 

@chiliwili69 as mentioned, I was somewhat surprised that the model favours raw RAM MHz vs. true latency to predict average fps as it seems counter intuitive. However, when I had a look at the spectacular 5.5 GHz data from jarg1, I couldn't help but notice an observation that supports this result even further:

 

Obviously, this is just one example and I would feel comfortable considering that the variation in outcome falls well in the error margin. That having said, when you compare WallterScott's configuration and the resulting average fps with those from jarg1, you can see that all Walterscott's values are lower or equal to jarg1's, save RAM MHz (I will consider jarg1's true latency of 7.74 ns virtually equal to Walterscott's 7.73 ns) except his 4400 MHz RAM vs. jarg1's 4100 MHz. Nonetheless, Walterscott's average fps is > 5 fps higher (179.2 vs. 173.9).

 

I found this result quite striking, especially given the fact that jarg1 actually has a higher CPU speed, which does correlate positively with average fps (it is one of the other predictors in the model) as well as a (somewhat) higher STMark. I'd be curious to see if additional data can actually disprove this counter intuitive result where RAM MHz has a stronger correlation with average fps than true latency.

[chiliwili69]

indeed!  I was also intrigued by being RAM MHz more important than true latency. It would be a nice discovery if more data support it.

 

But I found something weird in the data reported by the WalterScott runs.

You will see that I keep track of Frames, min, max and avg values. But the frames and avg fps is redundant since avgfps should be equal to frames/180 (since it is a 3 minutes track).

You will see that almost all test results follow that rules except the Walterscott run. It reports more frames than the corresponding to the avgfps*180. I wonder how this can be reported by Fraps.

[NiiranenVR]

Well its not .....but ......still ....:

 

[WallterScott]

06.02.2020 в 21:23, chiliwili69 сказал:

But I found something weird in the data reported by the WalterScott runs.

I did some more tests on different frequencies.

**************************************************************************************

Motherboard: asus rog maximus xi gene
 CPU:                 i9 9900k
 CPU Freq:        5.3 - 5.5 Ghz  AVX=0
 L3 cache:        16 MB

 NB Freq:          5.0
 RAM type:        DDR4
 RAM size:        16 GB
 RAM Freq:        4400 MHz
 RAM Latency:   17
 GPU:                 2080Ti MSI, max. core 2130Mhz
 STMark:           3216

 
5.3 HT On

1# 2020-02-11 22:27:40 - Il-2
Frames: 31586 - Time: 180000ms - Avg: 175.478 - Min: 119 - Max: 287

2# 2020-02-11 22:33:00 - Il-2
Frames: 30960 - Time: 180000ms - Avg: 172.000 - Min: 119 - Max: 287

 

5.3 HT Off

2020-02-11 22:41:07 - Il-2
Frames: 31220 - Time: 180000ms - Avg: 173.444 - Min: 119 - Max: 287

 

5.4 HT Off

2020-02-11 22:49:11 - Il-2
Frames: 31671 - Time: 180000ms - Avg: 175.950 - Min: 122 - Max: 286

 

5.5 HT Off

1# 2020-02-11 22:58:10 - Il-2
Frames: 31883 - Time: 180000ms - Avg: 177.128 - Min: 121 - Max: 287

2# 2020-02-11 23:08:12 - Il-2
Frames: 31755 - Time: 180000ms - Avg: 176.417 - Min: 121 - Max: 287

 

BOX  4.004

**************************************************************************************

I think up to 5 fps is the test error. And after 5.3 GHz, I don't see any increase in frame rate. 5.5 is achievable for IL2, but it is not possible for other applications. At this frequency, you can only work 24/7 with waterchiller.

[chiliwili69]

10 minutes ago, WallterScott said:

I did some more tests on different frequencies.

 

Thank you very very for reporting this additional tests!! You have a nice nuclear submarine.

Now the framerates are consistent with the avg fps.

I also have seen you changed Mobo from Apex to Gene, not sure if this should affect performance. Probably not.

 

Regarding the fps error test, normally it is up to 2 fps up and down. I think it is influenced by how the GPU is overclocked, but not really sure.

Perhaps at very high freq the error is larger.

 

If I could ask you something is to run the test at 5.0Ghz  three times, with HT off and giving STMArk. So we could compare with other chips (9700K, 9600K) which run at 5.0Ghz.

And also see if the error at lower freq is smaller.

[WallterScott]

13 минут назад, chiliwili69 сказал:

I also have seen you changed Mobo from Apex to Gene, not sure if this should affect performance. Probably not.

Apex is dead)

14 минут назад, chiliwili69 сказал:

If I could ask you something is to run the test at 5.0Ghz  three times, with HT off and giving STMArk. So we could compare with other chips (9700K, 9600K) which run at 5.0Ghz.

And also see if the error at lower freq is smaller.

I will do it tomorrow

[WallterScott]

CPU Mark 5.5 HToff - 3336

CPU Mark 5.0 HToff - 3030

 

5.0 Ghz

2020-02-12 23:26:56 - Il-2
Frames: 30012 - Time: 180000ms - Avg: 166.733 - Min: 117 - Max: 284

2020-02-12 23:32:20 - Il-2
Frames: 29935 - Time: 180000ms - Avg: 166.306 - Min: 116 - Max: 283

2020-02-12 23:49:02 - Il-2
Frames: 29944 - Time: 180000ms - Avg: 166.356 - Min: 116 - Max: 284

Edited February 12, 2020 by WallterScott

[chiliwili69]

Many thanks!

 

We see that the test error has been reduced to just 0.5 fps. Maybe when a PC is pushed to the limit the test error is higher.

 

And we also see that a 9900K at 5.0GHz (with your 4400MHz RAM) delivers around 15fps more than any other 9700K or 8086K at 5.0GHz (and lower RAM speed 3600Mhz).

 

If we discount the 4400 to 3600MHz, it will deliver 800Mhz*0.01667 less fps, so this is 13fps.

 

So maybe the effect of larger L3cache is not really important. If this is true a 9700K or 8086K or 9600K with RAM at 4400K could also deliver 166 fps at 5.0GHz.

 

My doubt is then what CPU to buy: 9700K, 9900K or 9900KS.  And for sure 16GB of RAM at 4400Mhz (or higher if exist)

[dburne]

And for VR headset you are talking running 90 /45 fps, or 80/40 fps...

[Stoopy]

After experiencing lockups, hangs glitches and many error messages having to do with "unable to execute instruction at memory location XXXX", etc. to no end with IL2 and my new Reverb when running with any settings above 1280x768, I decided to replace my original Corsair 2133Mhz DDR3 RAM and go with the 2400Mhz DDR3 RAM recommended.  I bought 32Gb's worth this time just because I was seeing a lot of memory utilization with everything going, and wanted to have no questions about headroom.  Plus, this stuff is getting in short supply (and the RAM I purchased went completely out of stock nationally a few days after mine arrived)

 

I am absolutely STOKED to report that so far, things are now completely different and QUITE stable - I'm running with settings that previously caused immediate system hangs, and having a blast in extended dogfights with no glitches or hangs at all.  It is WONDERFUL and now I can finally get serious about fine-tuning.

 

Most importantly, since the system is running much better I thought I'd run another benchmark and post results.  Here is the complete history from the first benchmark to the latest:

 

Jan 26 - Original Benchmark:

System Specs:

Motherboard:    Gigabyte Z97X-UD5H
CPU:            i7-4790K  
CPU Freq:    4.38 Ghz
L3 cache    8 MBytes
Cores:        4 
HT:        On (8 threads)
RAM Type    DDR3
RAM Size    16 GBytes
NB Frequency    3999.5 Mhz
RAM Freq:    1330 Mhz (2x Dual Channel 665.1)
RAM Latency     10.0
GPU:         1660Ti
STMark:        2630

 

IL2 BoX Release Version: 4.004

 

Results with Graphic Settings as specified:

2020-01-26 11:38:33 - Il-2
Frames: 16687 - Time: 180000ms - Avg: 92.706 - Min: 61 - Max: 175

 

 

Jan 27:  Second benchmark, after configuring RAM correctly for 2133Mhz setting:

 

2020-01-27 19:51:28 - Il-2
Frames: 18073 - Time: 180000ms - Avg: 100.406 - Min: 68 - Max: 190

 

 

Feb 16:  Removed 16Gb 2133Mhz RAm and installed 32Gb 2400Mhz RAM, disabled HyperThreading. Updated system specs:

 

Motherboard:    Gigabyte Z97X-UD5H
CPU:            i7-4790K  
CPU Freq:    4.38 Ghz
L3 cache    8 MBytes
Cores:        4 
HT:        Off (4 threads)
RAM Type    DDR3
RAM Size    32 GBytes
NB Frequency    3990.5 Mhz
RAM Freq:    2400 Mhz 
RAM Latency     13.0
GPU:         1660Ti
STMark:        2639

 

2020-02-16 13:45:32 - Il-2
Frames: 20876 - Time: 180000ms - Avg: 115.978 - Min: 72 - Max: 211

 

The most impressive thing?  Here's what chiliwili69 originally said about my first benchmark:

  

On 1/27/2020 at 3:21 PM, chiliwili69 said:

 

Going to that 2400MHz  (check your Mobo can support it) you will gain about 16 fps in monitor.

 

...and it seems I've gained about 15.5 fps from the previous benchmark with correctly configured 2133Mhz RAM, running this benchmark on my monitor.   Pretty impressive call you made there, chiliwili!    And a 23fps gain from the original. 

 

So this has gotta be, like, the best thread EVARRR.  Thanks so much for the great benchmark tool and fantastic advice!!!

 

 

Edited February 16, 2020 by =[TIA]=Stoopy

[dburne]

1 minute ago, =[TIA]=Stoopy said:

After experiencing lockups, hangs glitches and many error messages having to do with "unable to execute instruction at memory location XXXX", etc. to no end with IL2 and my new Reverb when running with any settings above 1280x768, I decided to replace my original Corsair 2133Mhz DDR3 RAM and go with the 2400Mhz DDR3 RAM recommended.  I bought 32Gb's worth this time just because I was seeing a lot of memory utilization with everything going, and wanted to have no questions about headroom.  Plus, this stuff is getting in short supply (and the RAM I purchased went completely out of stock nationally a few days after mine arrived)

 

I am absolutely STOKED to report that so far, things are now completely different and QUITE stable - I'm running with settings that previously caused immediate system hangs, and having a blast in extended dogfights with no glitches or hangs at all.  It is WONDERFUL and now I can finally get serious about fine-tuning.

 

Most importantly, the system is running much better so I thought I'd run another benchmark and post results.  Here is the complete history form the first benchmark to the latest:

 

 

Nice !

Congrats and have fun!!

[chiliwili69]

1 hour ago, =[TIA]=Stoopy said:

The most impressive thing?  Here's what chiliwili69 originally said about my first benchmark:

  

On 1/27/2020 at 9:21 PM, chiliwili69 said:

 

Going to that 2400MHz  (check your Mobo can support it) you will gain about 16 fps in monitor.

 

...and it seems I've gained about 15.5 fps from the previous benchmark with correctly configured 2133Mhz RAM, running this benchmark on my monitor.   Pretty impressive call you made there, chiliwili!    And a 23fps gain from the original.

 

Thanks for reporting again your test with your new memory. This is telling us more things.

 

But When I said you will gain 16fps, I said with respect your initial memory configuration, that was 1330. I just calculated based in the gain of greif correlation (2400-1330)*0.0167=17.8.

I said 16 because I was calculating it mentally 2400-1330 is about 1000 and 0.0167 is about 16. So the real gain should have been 17.8 but you got +23fps. Congrats!

 

As you know there could many other factors no taking into account in this tests. So this correlation is only a guidance.

 

[SCG_Fenris_Wolf]

On 2/4/2020 at 7:05 AM, jarg1 said:

 CPU:                 Intel Core i9 9900KS
 CPU Freq:        5.5 Ghz
 L3 cache:        16 MB
 Cores:               8  (number of active physical Cores)
 HT:                    Off (HyperThreading. If Cores=Threads then HT is Off)

Wow!! Is that with AVX offset = 0? You said it does not perform as expected? It's already a bomb though.

[jarg1]

2 hours ago, SCG_Fenris_Wolf said:

Wow!! Is that with AVX offset = 0? You said it does not perform as expected? It's already a bomb though.

 

No AVX offset - as far as I know IL-2 doesn't use AVX instructions so it wouldn't be a factor.  I feel like I lost a couple of fps after the latest update.  Anyway, this isn't a sustainable overclock with water cooling and in fact for daily use I run at 5.2 Ghz with no AVX offset.

[ZachariasX]

1 hour ago, jarg1 said:

as far as I know IL-2 doesn't use AVX instructions

It does and it runs on your rig as fast as your AVX offset goes.

[dburne]

19 minutes ago, ZachariasX said:

It does and it runs on your rig as fast as your AVX offset goes.

 

Yes agreed.

[jarg1]

1 hour ago, ZachariasX said:

It does and it runs on your rig as fast as your AVX offset goes.

 

Interesting.  Anyway I've never used an offset.  I'm going to do some reading on the subject to see if there is a scenario where I would see a benefit.

[E69_Qpassa_VR]

1 hour ago, jarg1 said:

 

Interesting.  Anyway I've never used an offset.  I'm going to do some reading on the subject to see if there is a scenario where I would see a benefit.

The offset decreases 0.1 for each 1 you assign in the BIOS. You should check the default value.

[ZachariasX]

2 hours ago, jarg1 said:

Anyway I've never used an offset.

Easier to get "great numbers" in benchmarks, as AVX (especially AVX512) produces more heat. If heat becomes an issue, you sometimes can clock the CPU higher when keeping AVX frequencies lower.

[jarg1]

51 minutes ago, E69_Qpassa_VR said:

The offset decreases 0.1 for each 1 you assign in the BIOS. You should check the default value.

 

I'm not using default.  It is set to 0.

 

9 minutes ago, ZachariasX said:

Easier to get "great numbers" in benchmarks, as AVX (especially AVX512) produces more heat. If heat becomes an issue, you sometimes can clock the CPU higher when keeping AVX frequencies lower.

 

Makes sense but I'm more curious whether there are any real world situations where I would see a benefit.

[dburne]

5 hours ago, jarg1 said:

 

No AVX offset - as far as I know IL-2 doesn't use AVX instructions so it wouldn't be a factor.  I feel like I lost a couple of fps after the latest update.  Anyway, this isn't a sustainable overclock with water cooling and in fact for daily use I run at 5.2 Ghz with no AVX offset.

 

Probably should have just posted the benchmark results from 5.2  then to help keep the data as accurate and realistic as possible.

 

Edited February 23, 2020 by dburne

[jarg1]

14 minutes ago, dburne said:

 

Probably should have just posted the benchmark results from 5.2  then to help keep the data as accurate and realistic as possible.

 

 

The data is accurate and realistic.  I can run 5.4 GHz without crashes but I think 5.2 GHz is better for the longevity of the CPU and the performance difference is minimal.  And I noted the 5.5 run was just for fun.  Someone with a chilled cooler could probably get a stable overclock at the speed so it is an option.

[SCG_Fenris_Wolf]

Yes, it seems that, depending on settings, 5.0 to 5.1, thoroughly clears the bottleneck posed on CPU by IL-2's engine on full details. At least, that's indicated by the VR performance tests. As I complained about earlier, this cannot be checked by correlation in and to 2D-monitor tests. The remaining problematic situations are mostly furballs bringing the frametime up. 

 

That said, the game could use some cheap performance optimizations. Could start by not showing surface movements of other planes once airborne / if gear is pulled in.

Edited February 24, 2020 by SCG_Fenris_Wolf