Jump to content
chiliwili69

Benchmark for CPU/RAM performance: Remagen 4.002 to 4.004

Recommended Posts

7 hours ago, dburne said:

Yes however you state a gain from 44.8 fps to 67.5 fps in VR.

However does that really mean anything in the final analysis ? Speaking VR here.

 

I don´t use ASW techniques (called Motion Smothness with the Index) since there is still some glitches.

I also tolerate quite well being in the 40 to 80 region.

 

But in a normal mission, 90% of the time I am not shooting (especially in the Spit) or doing low flight over cities. Also other maps are flatter and less demanding than Rheinland. So I am at 80 or very close to it most of the time.

So it is important to gain every single fps but with less eye candy on shadows or having mirrors.

Also Flying Circus is less demanding, so I am most of the time at 80.

That´s why I wanted to know who is eating all my fps.

15 minutes ago, Loki_1982b_rock1080 said:

Biggest impact was on my average FPS, which gained a little over 8 FPS. I'll take it.

 

That´s a very good gain on fps!

I think it clearly shows how important is good RAM in IL-2.

You now lead the table at version 4.003!

Share this post


Link to post
Share on other sites
Posted (edited)

If you are at 80 fps most of the time with an Index and that rig, you are doing quite well. Hang on to it.

Hats off to ya.

 

 

1 hour ago, chiliwili69 said:

 

That´s a very good gain on fps!

I think it clearly shows how important is good RAM in IL-2.

You now lead the table at version 4.003!

 

So now you are keeping track per version? Previous version data collected is what, just trashed?

I would think performance ratios across different rigs would maintain around the same spread between each other ,  regardless of game version as everyone will pretty much be on same version. As long as benchmark settings remained constant as well as rig hardware.

 

What about hotfixes?

New video drivers? Video driver versions being used? GPU boost clocks maintained?

AVX settings?

HT on or off?

:wacko:

 

 

Edited by dburne

Share this post


Link to post
Share on other sites
Posted (edited)
1 hour ago, dburne said:

So now you are keeping track per version? Previous version data collected is what, just trashed?

 

Version 4.003 gave something like a 3-4 FPS loss for high end systems. Data isn't trashed, it's labeled as old and possibly unreliable, if you check the spreadsheet.

 

In answer to your question about other stuff, I expect almost everything you listed could have an impact, yes! Definitely hotfixes and video drivers, 100% will have an impact.

 

My 3600-CL16 B-die ram just arrived. A night of benching for me then!

 

Edit: LOL. Oculus update. Of course! *sigh*

Edited by Alonzo

Share this post


Link to post
Share on other sites
Posted (edited)
33 minutes ago, Alonzo said:

 

Version 4.003 gave something like a 3-4 FPS loss for high end systems. Data isn't trashed, it's labeled as old and possibly unreliable, if you check the spreadsheet.

 

In answer to your question about other stuff, I expect almost everything you listed could have an impact, yes! Definitely hotfixes and video drivers, 100% will have an impact.

 

My 3600-CL16 B-die ram just arrived. A night of benching for me then!

 

Edit: LOL. Oculus update. Of course! *sigh*

 

Version 13 of the PTC. Main change is in the Oculus Mirror.

 

Not sure you are understanding my point however I was trying to make.

Ok so the benchmark was created to share and compare numbers between different rig setups using the same benchmark to get an idea of what computer hardware had more effect

on the results, and was run on version 4.002.  So we got a somewhat decent sampling for with version 4.002 to give a good idea. 

 

Now we have version 4.003. Ok so it could give better performance, same performance, less performance that is true. However the comparisons on  the spread between the different machines running the same benchmark on the same hardware, would also remain the same. So whilst the average could go up, down, or remain the same, they would do this for everyone. Again as long as hardware remained the same. Now if hardware has changed, then yes a new benchmark number would be appropriate and in order. But for the intent of the benchmark, I fail to see any benefit from running all the benchmarking again - it would only reveal better, less, or same performance but would apply to everyone.  At least as long as it is the same group of guys running the benchmark which would be likely.

 

Edit: But then again I am tired and need to go to bed, I may just be confusing myself lol.

It's hell getting old.

 

Edited by dburne

Share this post


Link to post
Share on other sites
9 hours ago, dburne said:

So now you are keeping track per version?

 

Yes, I this Remagen benchmark but also in all previous ones (Balapan, Samuel, Chili) the first column at the left is the version. That´s why I request to report the version of the run. Normally is the latest one. I think it is better to record the version than the date.

9 hours ago, dburne said:

Previous version data collected is what, just trashed?

No, I never trash any data. In this case all data from version 4.002 and 4.003 remains mixed in the same table.

But usually, as done in previous VR benchmarks(Balapan, Samuel, Chili) I group the data by version just to don´t mix them.

Sometimes new versions gives a bump or loss in fps, so usually it is better to group them to compare results from the same version.

In fact, if you run the test again you will most likely see that you will obtain about 3-4 fps less.

When there is more data fro 4.003, I think I will group the data of 4.002 and 4.003 separated.

9 hours ago, dburne said:

What about hotfixes?

Hotfixes were also recorded in previous VR benchmarks (Balapan, Samuel, Chili). You will them with an added letter after version number.

Normally I group them within the same version since usually they don´t affect performance.

So far there is no hotfixes for version 4.002 or 4.003

9 hours ago, dburne said:

New video drivers? Video driver versions being used? GPU boost clocks maintained?

 

This is something which is important for the GPU performance, but in this remagen benchmark I only was keeping focus on CPU/RAM. I wanted to take GPU out of the equation (and the clouds level is set at low)

In fact, the same PC with different high level GPUs (2080Ti, 2080S, 2080, 2070S, 2070, 1080Ti, 1080) should produce the same fps.

That´s why I want someone with a 9900K and a 1080Ti, just to check that GPUs doesn´t matter in this particular test.

 

So, I only keep track of the GPU card. The brand, driver, OC etc is not recorded.

 

 

9 hours ago, dburne said:

AVX settings?

 

In the instructions I explain that the AVX offset should be put at zero. So the CPU freq is maintained constant to the same value during the tests (CPU-Z, Passmark test, IL-2 test).

So, the overclocking should be done manually to keep constant the CPU freq and compare apples with apples in the results table.

9 hours ago, dburne said:

HT on or off?

 

I recently added some extra info (number of active cores and HT) to the data requested from the test. So I can also keep track of that detail.

Normally it is a bit better HT off to achieve better overclocking. But this is something up to everyone. I just record that data.

Share this post


Link to post
Share on other sites
8 hours ago, Alonzo said:

My 3600-CL16 B-die ram just arrived. A night of benching for me then

 

Based in Loki he got 8fps from 3000 CL16 to 3600 CL16.

So if you go from 3333 CL16 to 3600 CL16 you should get about 4 fps more in the same version.

Let´s see how it goes.

7 hours ago, dburne said:

So whilst the average could go up, down, or remain the same, they would do this for everyone. Again as long as hardware remained the same.

 

You are right here. A new version should affect to everyone on the same way (more, less or equal fps).

That´s why it is not strictly needed that all of us run the benchmark again for every new version, but we should take into account that results from different versions (from different PCs) are not comparable.

 

But it doesn´t harm to run the benchmark again with new versions. I will keep all data.

Share this post


Link to post
Share on other sites

Just got round to running the benchmark,

MB: MSI MPG Pro Carbon z390

CPU: i7 9700K

Freq: 4.8Ghz

L3 Cache: 12Mb

NB: 4300

active cores: 8  - no HT

RAM Type: DDR4

RAM Size: 16Gb

RAM Freq: 3600

Latency: CL 17

GPU: MSI RTX2070 Armour Stock clocks

GPU passmark: 15763

ST Mark: 2951

BoX version: 4003b

 

2020-01-09 17:06:11 - Il-2
Frames: 27224 - Time: 180000ms - Avg: 151.244 - Min: 104 - Max: 239

Share this post


Link to post
Share on other sites

So my 3600-CL16 Samsung B-Die RAM arrived. It's a G.Skill Ripjaws kit and was about $180 Canadian. I ran a bunch of benchmarks including some VR ones at the end.

 

There's a bunch of data here. Some summary info:

  • Upgrading from 3200-CL16 to 3700-CL15 gained me ~4 FPS in 2D mode and ~2.5 FPS in VR (Rift S).
  • Overclocking from 5.0 to 5.2ghz (+200Mhz) gained me ~2.5 FPS in 2D mode.
  • Increasing ring ratio from 4.7 to 5.0ghz (+300Mhz) gained me ~2.1 FPS in 2D mode.

Note that my 5.2ghz CPU and 5.0ghz ring ratio results are unstable, I can run IL2 in benchmark mode but they fail a Prime95 test, so I would expect to get a game crash eventually when using those settings.

 

My conclusion is that for me, the $180 RAM upgrade is an improvement, but I might do better putting that money into better overclocking for the CPU. My motherboard had decent power delivery and so I thought it was a good one for overclocking, sadly it turns out many features just don't work correctly and it was painful to even get the RAM to run at 3600. In my situation if I had a do-over and could purchase again, I would be better off taking that $180 and spending a little bit to get 3200-CL14, a little bit for a beefy AIO liquid cooler instead of my air cooler, and a little bit on a motherboard with better overclocking features.

 

If you are building from scratch for VR, my recipe would be:

  • Intel 9700K / actually good overclocking motherboard / 280mm or better AIO cooling
  • 16GB 3200-CL14 memory, or 3600-CL16 that is motherboard compatible, don't worry about Samsung b-die
  • Non-NVMe SSD as your main drive to save some cash
  • 2070 Super or better GPU, match GPU to your headset (Reverb or Pimax = 2080ti if you can afford it, otherwise 2070S or 2080)

Wall of text / benchmarks incoming!

 

Motherboard:    MSI Z370 SLI Plus
CPU        8086K
CPU freq:    5.0 Ghz        (hyper-threading disabled, 6-cores only)
L3 Cache:    12 MB
RAM Type:    DDR4
Ram size:    16 GB
Ram freq:    Varies, see individual results
Ram latency:    Varies
GPU:        2080  (EVGA 2080 XC Ultra, 'stock' pre-OC by manufacturer)
STMark:        Varies slightly with RAM OC, see individual results.

This is game version 4.003. Start at old RAM DDR4-3200-CL16:

HT off, 5.0ghz, 4.7ghz ring ratio.
DDR4-3200, 16-18-18-38

Passmark: 3029 / 3026 / 3030

Benchmarks:    Frames / Avg / Min / Max
Run 1:        26214 / 145.794 / 100 / 239
Run 2:        26050 / 145.144 / 102 / 239
Run 3:        26090 / 144.739 / 101 / 238

New RAM installed. Initially I couldn't get it to work at 3600Mhz, so here's a 3500-CL16 bench.

HT off, 5.0ghz, 4.7ghz ring ratio.
DDR4-3500, 16-16-16-36 2T  (not 3600)

Passmark: 3024 / 3022 / 3029

Benchmarks:    Frames / Avg / Min / Max
Run 1:        26743 / 148.572 / 104 / 240
Run 2:        26540 / 147.444 / 103 / 238

I increased voltage on the RAM to 1.4v, now I can get 3600 out of it.

DDR4-3600, 16-16-16-36 2T

Passmark: 3025

Benchmarks:    Frames / Avg / Min / Max
Run 1:        26794 / 148.856 / 104 / 240
Run 2:        26690 / 148.278 / 104 / 240

Headroom available to hit 3700. Auto system agent and IO voltage is too high, I don't like it, but here's a benchmark.

DDR4-3700, 16-16-16-36 2T

Passmark: 3033 / 3028 / 3031

Benchmarks:    Frames / Avg / Min / Max
Run 1:        26973 / 149.850 / 105 / 241
Run 2:        26727 / 148.483 / 104 / 240

Tighten the timings on the memory to CL15.

DDR4-3600, 15-15-15-36 2T

Passmark: 3034 / 3029 / 3031

Benchmarks:    Frames / Avg / Min / Max
Run 1:        26895 / 149.417 / 105 / 240
Run 2:        26815 / 148.972 / 105 / 239

Overclock the CPU beyond stability on my rig.

DDR4-3600, 15-15-15-36 2T, CPU 5.2ghz (not Prime95 stable)

Passmark: 3149 / 3144 / 3152

Benchmarks:    Frames / Avg / Min / Max
Run 1:        27387 / 152.150 / 105 / 242
Run 2:        27232 / 151.289 / 105 / 242

Overclock the ring ratio (uncore / north bridge).

DDR4-3600, 15-15-15-36 2T, CPU 5.2ghz (+200), ring ratio 5.0 (+300) (not Prime95 stable)

Passmark: 3143 / 3147

Benchmarks:    Frames / Avg / Min / Max
Run 1:        27764 / 154.244 / 107 / 244
Run 2:        27605 / 153.361 / 107 / 244

Set things back to a more 'sane' setting, this one is actually stable for Prime95 large FFTs.

DDR4-3700, 15-15-15-36 2T, CPU 5.0ghz

Passmark: 3033

Benchmarks:    Frames / Avg / Min / Max
Run 1:        26969 / 149.828 / 104 / 239
Run 2:        26874 / 149.300 / 104 / 240

Overclock ring ratio beyond stability, for more numbers.

DDR4-3700, 15-15-15-36 2T, CPU 5.0ghz, ring ratio 4.9 (+200) (not Prime95 stable)

Passmark: 3022 / 3021 / 3022

Benchmarks:    Frames / Avg / Min / Max
Run 1:        26991 / 149.950 / 104 / 241
Run 2:        26789 / 148.828 / 103 / 240

Now some VR benchmarks using the above settings and recorded track. Let's compare the old memory to the new. VR / Rift S / RTX 2080 stock / 1.0 pixel density / 1280x720 resolution / fullscreen.

HT off, 5.0ghz, 4.7ghz ring ratio.
DDR4-3200, 16-18-18-38

Passmark: 3029 / 3026 / 3030

Benchmarks:    Frames / Avg / Min / Max
Run 1:        12534 / 69.633 / 34 / 81
Run 2:        12450 / 69.167 / 39 / 81

The frame graph is important here too, gives you an idea how many stutters are on the track: https://www.dropbox.com/s/l1tiihm4brhwyjj/Il-2 2020-01-08 17-55-45-68-Time.png?dl=0

 

And same benchmark but with increased settings. I need to do more VR tuning like Chili has done, to come up with finalized settings.

DDR4-3700, 15-15-15-36 2T, CPU 5.0ghz, ring ratio 4.9 (+200) (not Prime95 stable)

3022 / 3021 / 3022

Benchmarks:    Frames / Avg / Min / Max
Run 1:        12972 / 72.067 / 39 / 81    
Run 2:        12958 / 71.989 / 37 / 81

Frame time graph showing improvement: https://www.dropbox.com/s/imw2odnukub3qjj/Il-2 2020-01-09 15-21-55-92-Time.png?dl=0 

 

 

 

 

 

  • Like 1
  • Thanks 1
  • Upvote 2

Share this post


Link to post
Share on other sites

Just went a bit further in OC my 7900X.

 

As 4.7 GHz on 10 cores (HT off, defalut uncore speed) clearly exceeds the cooling power of my be quiet! Silent Loop 280mm AIO, I them went and reduced the number of active cores to 5.

 

At 4.7 GHz my results were, for expample:

As2020-01-10 15:09:47 - Il-2
Frames: 24346 - Time: 180000ms - Avg: 135.256 - Min: 88 - Max: 240   5C no HT

 

That is is identical to 10 active cores. +-2 FPS is the variation I get between runs. At least with 5 active cores, my system could take Prime95, but only barely so.

 

I then decided against upping uncore frequency, as I would certainly need to up voltage for that. At least that is what most did when OC the Skylake-X. Also the consesus seems to be that while you can get great benchmark numbers like that, there is no real world gain. It seems that at 2.4 GHz, the cache runs at a sweet spot in terms of cost/benefit.

 

I went then to 4.9 GHz on 5 cores, HT off.

2020-01-10 15:31:32 - Il-2
Frames: 24764 - Time: 180000ms - Avg: 137.578 - Min: 92 - Max: 245  5C no HT

 

No change. Here, it is certain that therma throtteling keeps the CPU from doing 4.9 GHz, although this frequency can be reached at light loads. In terms of cooling the package, it is never an issue, even at 4.9 GHz, "CPU temp" does not go above 80°C. But I suspect there be spikes in core temps that cause throtteling down.

 

I think this is also visible on the graphs:

2.thumb.jpg.06e4c7958bcae9b8ede58ffe7620b1b0.jpg

You can see that at 4.9 GHz 5C can only be faster when there is less work/heat. Then it can clock up. On higher workloads, it is even contraproductive.

 

 

It seems that it is indeed my cooling solution that sets the limits me at ~137 FPS.

 

 

Share this post


Link to post
Share on other sites
23 hours ago, otherhunmiketeams said:

Avg: 151.244

 

Thanks for your test.

You got a nice fps for being only at 4.8GHz. 

Share this post


Link to post
Share on other sites

There seems to be good news for VR users, AMD getting competitive again.

 

There, some "engineering sample" (EDIT) supposedly beat the 2080 Ti by ~15% in the OpenVR benchmark.

 

Edited by ZachariasX
  • Upvote 1

Share this post


Link to post
Share on other sites

I have done this test on a Ryzen 2700X with the stock wraith Prism cooler.

 

Motherboard:    Asus Crosshair VI Hero
CPU:    2700X
CPU Freq:    4.3GHz
L3 Cache:    2x8MB
Cores:    8
HT:    On
RAM Type:    DDR4
RAM Size:    16GB
NB Freq:    1700
RAM Freq:    3400
RAM Latency:    16
GPU:    1060S
STMark:    2254

 

Frames: 22277 - Time: 180000ms - Avg: 112.650 - Min: 75 - Max: 191

Share this post


Link to post
Share on other sites

Hi Chilli,

yes FPS surprised me too.

I have had an issue with running the benchmark today - not sure why but on one occasion it’s ran at 10 FPS for half the test.

 I had turned the mirrors to simple and the clouds to extreme to see how it would run for my VR set up but everything else was the same.

strange

Share this post


Link to post
Share on other sites
On 1/9/2020 at 11:55 PM, Alonzo said:

This is game version 4.003. Start at old RAM DDR4-3200-CL16:

 

First at all, thanks you very much for all these multiple tests. Your methodic process and report is so nice. It confirms and give lights in more areas.

 

Your first test confirms that the version 4.003b eats about 3 fps over 4.002.

On 1/9/2020 at 11:55 PM, Alonzo said:

New RAM installed. Initially I couldn't get it to work at 3600Mhz, so here's a 3500-CL16 bench.


HT off, 5.0ghz, 4.7ghz ring ratio.
DDR4-3500, 16-16-16-36 2T  (not 3600)

Passmark: 3024 / 3022 / 3029

Benchmarks:    Frames / Avg / Min / Max
Run 1:        26743 / 148.572 / 104 / 240
Run 2:        26540 / 147.444 / 103 / 238

 

In this test we see two things:

 

- The S mark is insensitive to RAM upgrade. It should be based in some kind of simple but intensive calculation which should work mostly with cache rather than RAM.

- You improve your fps by 3, this is very much aligned with the expected results based on extrapolation from Loki_1982b results. Good to know.

 

On 1/9/2020 at 11:55 PM, Alonzo said:

Tighten the timings on the memory to CL15.

 

This test is interesting.

 

You initial test with the old memory has a true latency of 10 ns.

Your test at 3600-C16 has a true latency of 8.9 ns and you got 3.8 fps more. So 3.8/1.1 is 3.5, so we can say we get 3.5 extra for every ns in true latency.

 

When you tested the CL15, you went from true latency from 8.9 to 8.3. So 0.6 ns less. And 3.5*0.6 is +2.1fps.  But you only got 0.6 extra fps. Maybe frequency is more important than CL.

Share this post


Link to post
Share on other sites
On 1/9/2020 at 11:55 PM, Alonzo said:

Overclock the CPU beyond stability on my rig.


DDR4-3600, 15-15-15-36 2T, CPU 5.2ghz (not Prime95 stable)

Passmark: 3149 / 3144 / 3152

Benchmarks:    Frames / Avg / Min / Max
Run 1:        27387 / 152.150 / 105 / 242
Run 2:        27232 / 151.289 / 105 / 242

 

Thanks for going up to 5.2. This test is also quite interesting.

You run to the same freq than dburne test, so almost same STMArk. A bit less true latency. Same NB freq. But still 9 fps less, well 6 fps if we take into account version.

I wonder if the extra cache (12 to 16) plays a role here. You also only run 6 cores, but I think there is no difference in running 6 to 8 cores for IL-2.

You fps is pretty much aligned with test of Dakpilot at 5.2.

On 1/9/2020 at 11:55 PM, Alonzo said:

Overclock the ring ratio (uncore / north bridge).


DDR4-3600, 15-15-15-36 2T, CPU 5.2ghz (+200), ring ratio 5.0 (+300)

 

Thank you so much by this test. It clearly shows that NB freq is important to IL-2. You got +2fps by +300MHz.

You say it is not prime95 stable, but the important thing here is if it is IL-2 VR stable. At the end of the day you play IL-2 VR not Prime95.... 😉

I will continuedigestingg your tons of data tomorrow. Dinner time!

  • Like 1

Share this post


Link to post
Share on other sites
1 hour ago, chiliwili69 said:

 

Thank you so much by this test. It clearly shows that NB freq is important to IL-2. You got +2fps by +300MHz.

You say it is not prime95 stable, but the important thing here is if it is IL-2 VR stable. At the end of the day you play IL-2 VR not Prime95.... 😉

I will continuedigestingg your tons of data tomorrow. Dinner time!

 

Hmm I would never recommend this.

When overclocking I would be cautious and stress test with a decent stress test program for a bit of time, keeping an eye on both core voltage being applied as well as core temps. Make changes in small increments stress testing in between.

Of course everyone is free to decide how to overclock and run their own systems.  I would certainly urge any newcomers to overclocking to study up on overclocking their CPU's and Motherboard settings that can have an affect along with good stress testing in between each adjustment whilst watching temps/voltages.

 

Then when satisfied with the overclock along with the voltage/temps and passes stress tests, fly IL-2 and have a lot of fun.

 

Edited by dburne
  • Upvote 1

Share this post


Link to post
Share on other sites
6 hours ago, chiliwili69 said:

Thanks for going up to 5.2. This test is also quite interesting.

You run to the same freq than dburne test, so almost same STMArk. A bit less true latency. Same NB freq. But still 9 fps less, well 6 fps if we take into account version.

I wonder if the extra cache (12 to 16) plays a role here. You also only run 6 cores, but I think there is no difference in running 6 to 8 cores for IL-2.

You fps is pretty much aligned with test of Dakpilot at 5.2.

 

It's interesting, isn't it? The other difference is dburne has a 2080ti, I only have a 2080, but 6 fps is quite a difference when the GPU isn't heavily loaded.
 

Quote

 

You say it is not prime95 stable, but the important thing here is if it is IL-2 VR stable. At the end of the day you play IL-2 VR not Prime95....

 

 

Unfortunately I have found that if the system is not Prime95 stable (for large FFTs, so it's not quite the heat virus that small FFTs do) then it's not VR multiplayer stable. I run the Combat Box server and my rig crashed twice in one night using 3700-CL15, which was frustrating (in both cases I had flown for 10+ minutes towards a bombing target, then game crashed before I could deliver bombs). 3700-CL15 was "Prime95 stable" for 25 minutes, but crashed in IL2. I'm now experimenting with both 3600-CL15 and 3200-CL13, as they are Prime95 stable for 8+ hours.

 

5 hours ago, dburne said:

When overclocking I would be cautious and stress test with a decent stress test program for a bit of time, keeping an eye on both core voltage being applied as well as core temps. Make changes in small increments stress testing in between.

Of course everyone is free to decide how to overclock and run their own systems.  I would certainly urge any newcomers to overclocking to study up on overclocking their CPU's and Motherboard settings that can have an affect along with good stress testing in between each adjustment whilst watching temps/voltages.

 

Then when satisfied with the overclock along with the voltage/temps and passes stress tests, fly IL-2 and have a lot of fun.

 

I would agree with this advice. I've only just noticed that my motherboard, using auto system agent and system IO voltage, puts a lot of volts through those components. And the board won't boot unless it's set to auto. So even for me, someone who has been tinkering with overclocking in the name of VR and thinks he knows what he's doing, can easily miss important stuff. 

 

 

 

Comparing 3600-CL15 (real latency 8.33ns) vs 3200-CL13 (real latency 8.13ns):
 

DDR4-3600, 15-15-15-36 2T

Passmark: 3034 / 3029 / 3031

Benchmarks:    Frames / Avg / Min / Max
Run 1:        26895 / 149.417 / 105 / 240
Run 2:        26815 / 148.972 / 105 / 239
DDR4-3200, 13-13-13-36 2T, CPU 5.0ghz, ring ratio 4.7

Benchmarks:    Frames / Avg / Min / Max
Run 1:        26750 / 148.611 / 103 / 247
Run 2:        26507 / 147.261 / 103 / 245

So even though in theory the 3200-CL13 setting has lower memory latency, IL2 slightly prefers the 3600-CL15 setting. This is in line with the poster above who did the interesting multi-variable regression, which found that "RAM megahertz" was a stronger correlator to IL2 performance than "RAM real latency ns".

 

Next test, 2 vs 4 sticks of RAM...

Share this post


Link to post
Share on other sites
3 hours ago, Alonzo said:

Next test, 2 vs 4 sticks of RAM...

 

I'm going to have to disappoint on this one -- my CPU fan fractionally blocks the slot, so I can't get the extra memory sticks in there.

Share this post


Link to post
Share on other sites
On 1/9/2020 at 11:55 PM, Alonzo said:

Overclock ring ratio beyond stability, for more numbers.


DDR4-3700, 15-15-15-36 2T, CPU 5.0ghz, ring ratio 4.9 (+200) (not Prime95 stable)

Passmark: 3022 / 3021 / 3022

Benchmarks:    Frames / Avg / Min / Max
Run 1:        26991 / 149.950 / 104 / 241
Run 2:        26789 / 148.828 / 103 / 240

 

In this test you didn´t get almost any gain for the +200NB freq. Maybe too close to actual CPU freq. Really don´t know.

On 1/9/2020 at 11:55 PM, Alonzo said:

Benchmarks:    Frames / Avg / Min / Max Run 1:        12972 / 72.067 / 39 / 81     Run 2:        12958 / 71.989 / 37 / 81

 

Wow, having a 72 fps avg with those settings is quite an achievement!!

 

If you just relax the settings (Mirror off or simple, Shadow to medium, distan building off) you will probably will be almost at 80.

 

Then you can adjust the clouds and PD to load your GPU below 80% and that´s all.

  • Like 1

Share this post


Link to post
Share on other sites
On 1/10/2020 at 4:42 PM, ZachariasX said:

Min: 88 - Max: 240   5C no HT

 

That is is identical to 10 active cores. +-2 FPS is the variation I get between runs

 

Thank you for making the test with less cores. The avg is about 3.4 fps less, but the min and max is +10. That´s strange.

On 1/10/2020 at 4:42 PM, ZachariasX said:

Here, it is certain that therma throtteling keeps the CPU from doing 4.9 GHz,

 

When you went from 4.5 to 4.7 GHz you gained +16fps.

Now, with 5 cores, you went from 4.7 to 4.9 but only gained +2fps. As you said, there is some thermal throttling during the test.

On 1/10/2020 at 4:42 PM, ZachariasX said:

You can see that at 4.9 GHz 5C can only be faster when there is less work/heat. Then it can clock up. On higher workloads, it is even contraproductive

 

Yes, exactly that. 

 

By the way, your NB frequency (ring ratio) is 2700, which is quite low comparing with other intel CPUs on the table. Is there any way you can increase it?

On 1/10/2020 at 10:48 PM, ICDP said:

GPU:    1060S

Thanks for reporting.

 

The GPU is a 1060 o the 2060Super?

On 1/10/2020 at 11:05 PM, otherhunmiketeams said:

it’s ran at 10 FPS for half the test.

If you run it without wearing the headset it detect your head is not there. You can put something soft inside to activate the Rift.

Share this post


Link to post
Share on other sites
14 hours ago, dburne said:

Hmm I would never recommend this.

When overclocking I would be cautious and stress test with a decent stress test program for a bit of time, keeping an eye on both core voltage being applied as well as core temps. Make changes in small increments stress testing in between.

Of course everyone is free to decide how to overclock and run their own systems.  I would certainly urge any newcomers to overclocking to study up on overclocking their CPU's and Motherboard settings that can have an affect along with good stress testing in between each adjustment whilst watching temps/voltages.

 

Then when satisfied with the overclock along with the voltage/temps and passes stress tests, fly IL-2 and have a lot of fun

 

Everything you said is quite right.

 

I didn´t wrote well what I wanted to say. 

When people uses stress tool software sometimes they go to far in stressing the tool (as aLonzo said it is a virus heat more than a stress tool).

 

When I studied my overclock I use Prime95 (in fact you recommended me that tool) using the Torture Test. You can launch that stress test in 3 modes: 1 thread, 2 threads and 4 threads.

I was comparing the heat generation of all those with IL-2 VR:

 

https://forum.il2sturmovik.com/topic/29322-measuring-rig-performance-common-baseline-for-il-2-v3010/?do=findComment&comment=514455

 

So, I think that if one is going to use the PC for mainly play IL-2 VR, I think that the Prime95 with Torture Test with 2 threads is enough as a stress.

 

Otherwise the CPU is unrealistically stressed and you end up in a higher voltage (or lower clock) than you could use for playing IL-2 VR.

 

Also, about the time running the stress. I think 4-5 hours it is enough. There is people running that for 24 hours which I think it is not needed. (if your room is properly refrigerated)

Share this post


Link to post
Share on other sites

For info on anyone wondering why the NB frequency on AMD Ryzen always looks low.  RAM controllers for AMD Ryzen are integrated on the CPU die and typically run half the RAM speed.

 

 

Edited by ICDP

Share this post


Link to post
Share on other sites
50 minutes ago, chiliwili69 said:

By the way, your NB frequency (ring ratio) is 2700, which is quite low comparing with other intel CPUs on the table. Is there any way you can increase it?

Yes, I could do that by upping uncore cache frequency. By pushing voltage, I could certainly get 3.6 GHz instead of 2.4 GHz, giving me a NB frequency of ~4 GHz. It does produce more heat though, hence I would expect it to cap me earlier through thermal protection. Also, it seems common wisdom that games hardly profit from incresing uncore frequency.

 

My problem is here that I am obviously constrained through my cooling solution. The cetaceum (or whatever Intel used) in the CPU package is certainly of ne help either to dissipate heat. The case, a Fractal Define XL R2, only allows 280 mm radiators to be installed. I can put in two of those, a slim one in the top (as I have now) or a thicker one in the front. Right now I have the be quiet! Silent Loop 280 mm installed (top). It cools ok (~70°C Prime95, 20 threads @4.2 GHz), is quiet, and runs hassle free for 2 years now (I have a new spare one just "in case") and it doesn't require crapware to operate it. But now, would there be a better option? Is a Corsair H115i Pro that much better? I wouldn't want to scratch build a solution myself by combining two coolers and do all the piping etc.

 

But just for the hell of it, I might try to up uncore/NB as detailed above and run the cores only at 4.3 GHz and compare then.

Share this post


Link to post
Share on other sites
5 hours ago, chiliwili69 said:

Wow, having a 72 fps avg with those settings is quite an achievement!!

 

If you just relax the settings (Mirror off or simple, Shadow to medium, distan building off) you will probably will be almost at 80.

 

Then you can adjust the clouds and PD to load your GPU below 80% and that´s all.

 

In fact the best I have achieved in overall tuning is 75.1 fps. Detailed tuning spreadsheet. This is the graph:

 

1290679672_Il-22020-01-0918-27-12-87-Time.thumb.png.33ecf81de631272afb58fac5284f9047.png

 

The recipe for me is as follows, I use the config file to set options rather than the in-game control usually:

  • 3D Migoto Mod installed (needed for multiplayer)
  • 1.2 pixel density
  • 1280x720 full screen mode
  • Bloom enable 0 (yuck)
  • Draw distance 0.54 (100km)
  • far_blocks 0 (distant buildings off)
  • Grass distance 100 (normal)
  • HDR enable 1
  • Land anisotropy 1 (landscape filter sharp)
  • Max clouds quality 2 (high)
  • Mirrors 0 (off, fairly useless in multiplayer)
  • Multisampling 0 (pixel density does the job here, but I can also enable 2x AA for a 0.4 fps average drop)
  • Post sharpen 1 (sharpening on, helps spotting)
  • Preset 2 (High)
  • Shadows Quality 2 (High, in VR I feel like I need the definition in shadow or they look really bad on my canopy)
  • SSAO enable 0

Unfortunately everything falls apart when I run a real multiplayer game with these settings. Having many aircraft in one spot on the map is very CPU intensive and I drop down into 40 ASW almost all the time. I find I have a smoother overall experience simply forcing 40 ASW and then dealing with the artifacts for fast moving stuff. Today I think I will experiment with how much headroom I need for a multiplayer situation.

  • Like 1

Share this post


Link to post
Share on other sites

thank you for let me in the crowd 👍   , Im not so experienced in testing pc/game/graphich, but its great one have the power to control it 

I thought I have Ram 3000 MHz , but as I see in my CPU-Z it is 2x1067 = 2134 - But I have ordered one 2x8 3600 MHz cas 15 and maybe Ill get it to morrow 

But heres the test without VR and whats written first page 

 

Motherboard:  ROG MAXIMUS CODE
 CPU:                 i5 9600k
 CPU Freq:       5.2 Ghz
 L3 cache:        9 MB

 Cores:             6  

 HT:                    Off ( not possible to on/off in Bios ) 
 RAM type:        DDR4
 RAM size:        16 GB ( single )

 NB Freq:          4297,9 MHz
 RAM Freq:        1067,2  MHz (this is 2xDRAM freq if Dual channel) <<<<< but this is single ...so ??? 
 RAM Latency:  15  
 GPU:                 2080super
 STMark:          3128

 

Frames: 21002 - Time: 180000ms - Avg: 116.678 - Min: 77 - Max: 196

Share this post


Link to post
Share on other sites
11 hours ago, Niiranen-VR said:

RAM Freq:        1067,2  MHz (this is 2xDRAM freq if Dual channel) <<<<< but this is single ...so ???

Thank you for your report and welcome to the remagen results table.

You have a nice CPU and nice OC to 5.2 GHz, but your fps is really low (116 fps), you should really achieve much more.

The reason for that could be that you don´t run in Dual channel since you only have one stick of 16Gb.

 

I think you will have a very nice surprise when your new pair of memory stick arrive. Please, let us know how it goes.

Share this post


Link to post
Share on other sites

Just for do something - better than do nothing 

 

The test with Lefuneste mod on :        Frames: 19700 - Time: 180000ms - Avg: 109.444 - Min: 75 - Max: 186              ( forgot to uncheck it yesterday ) 

The test used without mod          :        Frames: 21002 - Time: 180000ms - Avg: 116.678 - Min: 77 - Max: 196                allready sent and written 

The Test with VR Reverb              :         Frames: 8869 - Time: 180000ms - Avg: 49.272 - Min: 26 - Max: 91

Share this post


Link to post
Share on other sites

Im retired in about 10 years from now , but I still remember - in the past -  the feeling about doin my homework ...... thats the feeling I have now 😁  ...but ...so .... no more well-worded rubbish  😉

 

Today I got my new ram 2x8 Trident  G Skill 3600 cas 15 ......first in CPU-Z the timing was around  1200 /2400 MHz but in Bios set to auto and was the same yesterday , so .....now the new test 

 RAM type:        DDR4
 RAM size:        16 GB 

 NB Freq:          4300 MHz
 RAM Freq:      3600  MHz (this is 2xDRAM freq if Dual channel) <<<<< Now dual 
 RAM Latency:  15  

 

test       :     Frames: 23817 - Time: 180000ms - Avg: 132.317 - Min: 83 - Max: 214

VR         :     Frames: 9991 -   Time: 180000ms - Avg: 55.506 - Min: 44 - Max: 91

 

its great you do it Chilli , and I like the schedule sheet on net - and if you want me to make more test - tell me 👌

Share this post


Link to post
Share on other sites
4 hours ago, Niiranen-VR said:

the feeling about doin my homework

 

good guys always do homework before having fun... 😉

 

With you PC (CPU at 5.2, 3600-CL15 RAM, good Mobo, 2080) you should achieve around 150 fps avg.

 

I wonder why you don´t reach that. Let´s review:

 

CPU:

Your STMark is 3128. It is a bit below the 3180 that Dakpilot achieved with the same CPU at 5.2GHz. So maybe your CPU freq is not mantained at 5.2 during the test and it is changing depending on the load/temperature/amps of the CPU.

What CPU cooler do you have? (To maintain 5.2 fixed for all cores you need a decent CPU cooler)

To check your CPU freq you can trend it with MSI Afterburner. You can configure trends for CPU freq, core freq, core temps, etc.

 

RAM:

In your BIOS you will several XMP profiles available for your new memory. You can try them to see which one delivers you best performance (typically the highest if stable).

 

GENERAL:

You can also verify this old post, but still  could useful:

https://forum.il2sturmovik.com/topic/34107-items-to-review-if-you-have-low-performance-in-il-2-vr-test/

  • Thanks 1

Share this post


Link to post
Share on other sites

Follow-up 

My watercooler is a 3' - fractal design celcius s 36 

 

I have now XMP  settings my RAM  to the highest of 2 possible 

I can OC my ram more ( as I see 'google' ) but don't think it will give so much benefit as I understand it's mostly the CPU and GPU 

 

I'll not make a test right now - my setting are back to normal and eyecandy 😁

 

( And I donno why my pic' not stand correct 🥴

 

 

16867BAE-CC3A-4494-8292-DD5ABBF9D899.jpeg

8F824243-BC9D-4C24-B44F-94E3BA8A348D.jpeg

1E3CFEC4-00E5-4F4F-8DBF-F6DDFA61DC29.jpeg

Share this post


Link to post
Share on other sites

I am not familiar with Asus BIOS, but in general it is better to do manual OC rather than Auto.

In Auto mode the BIOS decide the freq at which it should be run and usually put higher voltages than needed, producing more heat, so higher temps/power, so lower CPU frequencies.

Maybe someone here with Asus BIOS or Mr google can tell you how to setup your BIOS for manual overclocking (all cores to same freq, AVX=0).

I am not an expert in OC, I have just asked advice to people in this forum and they gave me good insight:

https://forum.il2sturmovik.com/topic/29881-overclocking-a-4790k-for-better-bos-performance/

 

Share this post


Link to post
Share on other sites

Short - time for work .....

Yes about the auto and Ram , I change that ( as written above ) 

My AVX = 0 .... Set on auto but still 0 

 

I'm not an expert to pc either ( I'm a bricklayer )  I'm more superficial profound Than opposite 🥴 

 

I think my game is running good ..... But - ( read a lot ) my CPU have 6 cores , I donno wether it would be a benefit to roll on 4 

Share this post


Link to post
Share on other sites

The first and easiest step to do now is to check your actual CPU frequency. For this you download something like HWMonitor and run that (as Admin). make it show you both teperatures of the cores as well as core frequencies. Next is you (download and) run Prime95. You select torture test with suggested number of 6 threads. Now you check HWInfo, look at the frequencies your CPU cores are running at as well as the temperatures. In your setting you should have 5.2 GHz on all 6 cores, temperatures high but they really shouldn't pass 90°C. Even 80°C is not a temperature you want to bake your CPU for extended time. But if it is above 90°C, you can expect thermal protection to kick in (unless you have set that at a higher temp.) and you will have slower running cores, maybe 4.8 GHz.

 

IF you have lower clocks but still permissible temperatures, then go and enter your BIOS. You should scroll down a bit on your BIOS screen. Where it says "CPU Core Ratio", there you can select either [Auto] (what I think you have), [Sync All Cores], [By Core usage], and [By Specific Core]. It would now be interesting what you selected. For running IL-2, you potentially want to select [Syn All Cores], but this can put severe strain on your system should you run a program that REALLY puts load on all your cores. If it hasn't been selected like that and temperarutes were ok when running Prime95, set that to [Sync All Cores].

 

If you boot Windows, run HWInfo and Prime95, you should see now all cores running at the same frequency and at your set 5.2 GHz. Then run the IL2 track.

 

If you have 5.2 GHz on all cores, permissible temps in Prime95 but when you run IL2 and you get something markedly short of 150 FPS, then something is really at odds. Why don't you run Passmark performance test and there you select the Memory test. My memory test looks like this, but on a 4 channel system @3.6GHz, CL16:

Spoiler

mem.jpg.447f308c8d33f95353d4dd41415387a9.jpg

 

What is your score?

 

Going to 4 cores will do nothing to your scores. The only gain would be that you in principle could save 50% heat by turning off two cores, making up for an insufficient cooler. But for IL2, even that would be of no consequence as the same load (heat) gets shuttled around 4 cores instead of 6. Thus, you are essentially shooting yourself in the foot as you keep heat production in more concentrated spots on the die.

 

Share this post


Link to post
Share on other sites

Well , .....I stand at the top of the  World , looking down to the mob 😆

But ....please don't look at my 'stats' 😖🥴

 

And'eeh , as I understanding it , no benefit for me to only 4 cores 👍

D04B829B-B822-46CD-9D55-A67D50001F6E.jpeg

182C3481-1265-4B8F-BF1C-4597CD1D2D20.jpeg

Edited by Niiranen-VR

Share this post


Link to post
Share on other sites

If these are HWInfo readings while Prime95 is running, you should be more than fine with your rig. It looks good, also memory speed looks what can be expected.

 

Two things come to mind now. First is whether you have some „protection“ program running in the background that actually protects you from a good part of the added FPS you should have. The other is that you should really populate both DIMM banks. Usually, you should buy paired kits that match best. But you‘re not too far off the mark with your timings. If you buy another DIMM as you put in now, I‘d expect it to work.

Share this post


Link to post
Share on other sites

OK, with all these new entries, I thought I'd make an update for the multivariate regression.

 

The best model I got used CPU GHz, (RAM)MHz, GPU Passmark and  STMark and got an R^2 of 0.938. However, because I considered the collinearity between CPU Ghz and STMark to severe, I think the better model is the one only using CPU GHz, (RAM)MHz and GPU Passmark, with a still very comfortable R^2 of 0.929.

 

image.png.1fcef03b5f7e98c56eaa116976df5c05.png

 

This is the updated model:

Parameter Estimates
Variable Label DF Parameter Standard t Value Pr > |t| Tolerance Variance 99% Confidence Limits
Estimate Error Inflation
Intercept Intercept 1 -58.1902 10.26161 -5.67 <.0001 . 0 -86.09648 -30.28392
CPU GHz CPU GHz 1 18.84354 2.12835 8.85 <.0001 0.86813 1.1519 13.05553 24.63155
MHz MHz 1 0.01669 0.00126 13.24 <.0001 0.96012 1.04154 0.01326 0.02011
GPU Passmark GPU Passmark 1 0.00356 0.0004605 7.73 <.0001 0.8432 1.18596 0.00231 0.00481

 

In other words: avg FPS = -58.1902 + (18.84354 * CPU GHz) + (0.01669 * (RAM) Mhz) + (0.00356 * GPU Passmark)

 

Obviously there are caveats to the model, for example the number of observations is somewhat limited combined with the fact that some people have contributed multiple lines concerning the same computer configuration with some changes. More data is (generally) better, but this can mean that particularities of their setup will end up being over represented in the model.

 

Normalizing the data so that the predictors all have the same scale gives the same model may give insight in the variables that have the most influence on the result. Again, the relatively high factor for Mhz  (as well as the higher t-value) suggests that if improving RAM speed and upgrading were the same cost, improving your RAM speed would give you the most bang for your buck. Given the limitations of the data and because of the counter intuitive results of raw RAM Mhz being a better predictor that true latency, I remain skeptical, though obviously faster RAM will never hurt.

 

Parameter DF Estimate Standard t Value Pr > |t|
Error
Intercept 1 136.643 0.72123 189.46 <.0001
CPU GHz 1 6.94063 0.78393 8.85 <.0001
MHz 1 9.87021 0.74544 13.24 <.0001
GPU Passmark 1 6.1469 0.79544 7.73 <.0001
  • Like 1

Share this post


Link to post
Share on other sites
1 hour ago, JG1_G_Greif said:

In other words: avg FPS = -58.1902 + (18.84354 * CPU GHz) + (0.01669 * (RAM) Mhz) + (0.00356 * GPU Passmark)

 

Normalizing the data so that the predictors all have the same scale gives the same model may give insight in the variables that have the most influence on the result. 

 

Can I just multiply up? So the above line is the same as 

 

avg FPS = -58.1902 + (18.84354 * CPU GHz) + (16.69 * (RAM) Ghz) + (0.00356 * GPU Passmark)

 

Which means that adding 100Mhz on the CPU is worth slightly more FPS than 100Mhz on the RAM, but doesn't compare the prices of the two.

 

Let's look at my rig before I did any upgrades. I have 5.0ghz CPU and 3200mhz RAM. I think I can spend $180 to get to 5.2ghz, or $190 to get 3600mhz RAM. Which should I do?

  • $180 would give 200mhz CPU improvement = 18.84 * 0.2 = 3.768 fps expected
  • $190 would give 400mhz RAM improvement = 16.69 * 0.4 = 6.676 fps expected

Therefore I should spend the money on the RAM, in this model.

 

I agree that peculiarity of setup doesn't help us. For example, my actual FPS upgrades are:

  • RAM from 3200 to 3600 (both CAS-16) gave +3.06 fps
  • CPU from 5.0ghz to 5.2ghz (unfortunately with already 3600-CL15 RAM, so not "base case" test) gave +2.73 fps

And we have margin of error on all the measurements and probably some effects as we approach the limits of the improvements.

  • Like 1

Share this post


Link to post
Share on other sites
3 hours ago, JG1_G_Greif said:

More data is (generally) better, but this can mean that particularities of their setup will end up being over represented in the model.

 

Thank again for making the regression!, it is interesting what we obtain.

I agree that particularities of one PC could weight more the regression to that PC, but in the other hand that PC will also show the clear gains of that PC.

 

There are several things here that we can do to improve this:

 

- Separate versions in groups (In this case the 4.003 gets about 3 fps less)

- Separate the Ryzens in other group since they have a different IPC than Intel and very different NB freq. (unless we take NB freq)

- I think it is better to take STMark rather than CPU freq, since the STMark accounts for both CPU freq and IPC

- We have seen that NB freq has an impact as well, so it should show some correlation.

- The GPU Passmark should not affect. I think that the best CPU freq has also good GPUs, so there should be some collinearity in GPUMark and CPU freq or STMark

 

For the time being I am going to group by version.

 

With your correlation all of us will know the fps we will achieve before buying the PC, so we can do a wiser purchase.

  • Like 1

Share this post


Link to post
Share on other sites
30 minutes ago, chiliwili69 said:

- We have seen that NB freq has an impact as well, so it should show some correlation.

Maybe one could add that it looks like the game uses only one channel to read/write to the RAM. My system performs much more like a 3600 MHz dual channel system in IL2 that a "7200" MHz system. On synthetic benchmarks (and my real life experience) this shows rather well that you require threaded applications to fully use a quad channel memory system. Hence, memory performance comes down to plain latency. 

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...