Jump to content
chiliwili69

Benchmark for CPU/RAM performance: Remagen 4.002 to 4.004

Recommended Posts

Posted (edited)
8 hours ago, LLv34_Flanker said:

The GHz is not fixed, just took it from the CPU-Z. It varies between 3.7GHz to 4.6GHz. I have a cooler comparable to Noctua NH15, so temps are no issue. 

 

  @Jaws2002, @ICDP, @LLv34_Flanker, @Gomoto

 

I think all the overclocked test done with Intel CPUs were done at a fixed CPU freq with AVX offset=0. So, the CPU run at the same frequency for the CPU-Z, for the STMark test and for the IL-2 Remagen test.

 

But with the Ryzen chips, it seems that the CPU freq is not fixed to a value, but a kind of OC profile. I have no idea about how Ryzens are overclocked but for this test it would be good to run it at fixed frequency, the maximum the cooler can support.

 

I don´t know if any of you know how to run it at fixed frequency, but I could put these instructions in the post, so we can compare results of different runs and Intel runs.

Edited by chiliwili69

Share this post


Link to post
Share on other sites
Posted (edited)

Hi Chili,

 

The Ryzen 3x00 CPUs will overclock cores based on their thermal/voltage, so an all core overclock for gaming is actually detrimental.  I set my all core OC to 4.3 and the 4.4GHz manually and got the following 2 results

 

4.3GHz

Frames: 24098 - Time: 180000ms - Avg: 133.878 - Min: 85 - Max: 225

 

4.4Ghz

Frames: 24878 - Time: 180000ms - Avg: 138.211 - Min: 87 - Max: 235

 

For lightly threaded loads the CPU will jump from core to core and the cores will be running at 4.4 - 4.6 GHz.  So the overall score in this benchmark will be higher.  I should stress the above scores are atypical for a Ryzen 3x00 CPU and should not be added to your spreadsheet to avoid confusion.

Edited by ICDP

Share this post


Link to post
Share on other sites
Posted (edited)

I did a test with overclocked CPU Cache Ratio settings as it was discussed earlier. Apparently there is no considerable effect on my system:

 

---------------------
IL-2 version:    4.003
Motherboard:    Asus Z97-A
CPU:        4790K
CPU Freq:    4.7 Ghz
NB Freq:    4000 MHz
L3 cache:    8 MB
RAM type:    DDR3
RAM size:    16 GB
RAM Freq:    2666 MHz
RAM Latency:    11
GPU:        GTX 1080 TI
STMark:        2829


RESULTS:
2019-12-26 14:21:59 - Il-2
Frames: 22535 - Time: 180000ms - Avg: 125.194 - Min: 80 - Max: 221


---------------------
IL-2 version:    4.003
Motherboard:    Asus Z97-A
CPU:        4790K
CPU Freq:    4.7 Ghz
NB Freq:    4000 MHz >> 4400 MHz
L3 cache:    8 MB
RAM type:    DDR3
RAM size:    16 GB
RAM Freq:    2666 MHz
RAM Latency:    11
GPU:        GTX 1080 TI
STMark:        2829 >> 2824


RESULTS:
2020-01-04 13:19:18 - Il-2
Frames: 22605 - Time: 180000ms - Avg: 125.583 - Min: 82 - Max: 216

Edited by HunDread

Share this post


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

I set my all core OC to 4.3 and the 4.4GHz manually

 

I see, thanks for the explanation.

 

I understand then that the Auto OC will speed up only the cores that are loaded, so higher freqs are reached (4.6 GHz).

 

I understand as well that 4.4 is the max freq for the 12 cores, but have you tried to disable 6 or 8 cores ( having only 6 or 4 cores active) and reach a higher freq. Technically less heat will be produced and possibly higher freqs could be reached.

2 hours ago, HunDread said:

Apparently there is no considerable effect on my system

 

Thanks for this test. This is something we didn´t test.

My Mobo BIOS doesn´t allow me to modify the ring ratio freq.

Share this post


Link to post
Share on other sites

Ryzen 3900X has 2 chiplets with 6 cores each, one is a good bin and the other is not as good.  So you essentially get a chiplet and a shitlet.  So I used Ryzen Master to turn on game mode which essentially turns off the 6 slower cores.  Normally this makes zero difference or in some cases gives worse results.  Yet with Il-2 I got a mild boost and the cores were running between 4.5 and 4.6GHz, which is up ~100MHz. 

 

Frames: 24703 - Time: 180000ms - Avg: 143.544 - Min: 98 - Max: 238

 

Unfortunately switching from/to game mode requires a reboot so I will leave it at normal mode.

 

  • Upvote 2

Share this post


Link to post
Share on other sites
On 1/3/2020 at 12:53 PM, Jaws2002 said:

I tried SMT off, and it made no difference. I got worse performance running 16 cores/16 threads than 16 cores 32 threads. I got better performance disabling one chiplet and running the CPU at 8 cores and 16 threads. 


I think this is quite interesting. We know from Intel CPUs that IL2 likes fast, low latency memory as well as single threaded performance. Now we also know from AMD that IL2 does not like cross-chiplet stuff (maybe latency, maybe coordination).

 

My suspicion is there’s something in the IL2 code that is doing some sort of memory/variable locking, and that making the lock process faster (fast ram, fast north bridge, only one chiplet) allows more FPS. Of course since the game is multi threaded it’s not a surprise there’s locking and thread synchronization going on, but if that part could be optimized we might see a big jump in performance.

Share this post


Link to post
Share on other sites
Posted (edited)

This happens in more games, not only Il-2. No idea what's the reason, but from what I've seen, turning SMT off on Zen 2 CPUs drops the low 1% framerate in most games. Some games do have a bit of boost in average frames, with SMT off, but most of them have lower minimum 1% frame rate. SMT ON will give you a smoother experience.

 I left mine in 16x32 mode, since, droping it to 8x16 gives me ove 100MHz boost increase, but only gets me three to four extra frames and for everything outside gaming, 16x32  is a heck of a lot faster.

  

 

Edited by Jaws2002
  • Thanks 1

Share this post


Link to post
Share on other sites

Results as per OP instructions

 

Motherboard: MSI Z390 Gaming Edge
 CPU:                 9600K
 CPU Freq:        5.0 Ghz
 L3 cache:        9 MB

 Cores:              6 

 HT:                    -/Off 
 RAM type:        DDR4
 RAM size:        16 GB

 NB Freq:           4.7 GHz
 RAM Freq:        3226 MHz 
 RAM Latency:  CL14
 GPU:                 MSI  GTX 1080Ti Seahawk
 STMark:           3053  (averaged)

 

2020-01-04 17:31:37 - Il-2
Frames: 26741 - Time: 180000ms - Avg: 148.561 - Min: 102 - Max: 239

2020-01-05 11:39:14 - Il-2
Frames: 26734 - Time: 180000ms - Avg: 148.522 - Min: 103 - Max: 239

2020-01-05 11:43:07 - Il-2
Frames: 26586 - Time: 180000ms - Avg: 147.700 - Min: 104 - Max: 240

2020-01-05 11:48:01 - Il-2
Frames: 26566 - Time: 180000ms - Avg: 147.589 - Min: 104 - Max: 239

 

 

Some additional Info

 

this is at 5.0ghz OC , for 9600k and my lower/mid budget MB, 5.0ghz is almost a one click easily achieved result, 9600k has A bin percentage of 37% vs 9700k A bin of 9% and 9900k A bin of 20%

 

All 9 series CPU's seem to have very similar overall A+B bin %

 

Chance of C bin 9600k 27% 9700k 29% and 9900k 16%

 

(Figures from MSI website from their testing, (maybe a bit out of date with latest F and S chips)

 

basically with Intel 9 series, silicon lottery will almost always give you a 5.0ghz possibility with adequate cooling

 

Some observations/thoughts

 

I did a quick overclock of 5.2ghz (no long stress tests but it seemed to run fine in game. My time is currently very restricted)  Vcore was 1.43 (auto) which is deemed fine by intel and MSI engineers on 9600k, if temps are fine, however a lot of people would consider this high!

 

this gave an ST mark of 3180 

 

5.3 ghz was unattainable at auto settings

 

I don't use/have VR so 5.0ghz seems acceptable for me, RAM has been left at XMP standard, am reasonably sure i could get a stable 24/7 5.2ghz with manual lower voltages with more testing and small RAM O/C if needed.

 

I have not studied the spreadsheet (yet) but from my results and what i remember from the posted/test results are very similar to 9900k results? IL-2 not needing the extra two cores and easier cooling giving easier O/C? Perhaps 9900k gives advantage in actual VR when combined with 2080ti?

 

Cheers, Dakpilot

 

 

  • Upvote 1

Share this post


Link to post
Share on other sites

5 GHz would be quite acceptable for VR as well.

My avg fps at 5.2  GHz was just under 161 FPS which would make sense.

 

I am able to run my 9900k at 5.2 GHz with voltage being applied under 1.4, so am happy with that and temps are good. I started to try to get 5.3 however

was not comfortable with the voltage I was having to give it , so figured not worth it for an extra 100 Mhz. I like having the extra headroom on the core temp side of things.

 

  • Upvote 1

Share this post


Link to post
Share on other sites

Thanks for the info Dakapilot and dburne.  It seems that 5GHz on the newer i7 or i9 chips will get ~5%-10% better performance than Ryzen 3x00.  Though if you can get 5.2 then that performance lead goes to 15%+

 

Ryzen is getting closer but for single threaded Il-2 you are better with Intel i7 or i9.

Share this post


Link to post
Share on other sites
Posted (edited)
2 hours ago, Dakpilot said:

Or i5? 

 

Cheers, Dakpilot 

 

Indeed an i5 will also hit 5GHz and give better results than a 3600X.

 

Though bear in mind that in order to keep that 9600K or 9700K at ~5GHz requires fairly expensive cooling such as a decent AIO or very good air cooler. at $80 ~ $200.  If you play all your games at 1080p and low/mid settings then yeah, a good CPU is king and you can use this benchmark to determine you next CPU for Il-2.  But realistically once you start increasing resolution, or graphical settings then the GPU becomes the bottleneck as long as you have any half decent CPU.  The same applies when you start using the newer higher resolution HMDs such as Reverb, Index or Pimax.  So the question then becomes "should I use that extra ~$80 - $200 towards a better GPU"?

 

When making a new build I would always ask people what their target resolution is, or what HMD (if any) they plan to use.  For example if playing at 4K, a 3600X with a 2080Ti is going to run Il-2 faster than a 9900K @5GHz with a 1080Ti and the difference in the cost for the CPUs (and expensive cooling) will pay for the 2080Ti.

 

The thing is when building a PC it is all about finding the balance between CPU and GPU.  I have a 3900X with a 2080 and if I were to change my CPU to a 9900K at 5.2GHz my gaming experience at 4K would be identical, and my VR experience with Pimax would be marginally faster at best because that 2080 is my main bottleneck, not my CPU.  The trade off would be I would lose out in productivity with 4 less cores.

 

As you can see here at 4K an i9 9900K at 5GHz has zero advantage over even cheaper CPUs.

relative-performance-games-38410-2160.pn

 

 

Now change that 2080Ti for a 1080Ti and at 4K you will see a 20-25% performance drop.  So at 4K GPU is the main factor for increasing performance and my experience with Pimax 5K indicates that my 2080 GPU is the bottleneck.

relative-performance_3840-2160.png

Edited by ICDP
  • Upvote 1

Share this post


Link to post
Share on other sites

TL;DR: Lesser systems do better in IL-2.

 

I redid the Remagen V1 track with various settings on my Socket 2066 system. Here's what I've got:

 

Motherboard: ASUS Rampage VI Extreme
 CPU:                 i9-7900X
 CPU Freq:        4.6/4.2 Ghz (4.5, 4.7)
 L3 cache:        13,75 MB L3 Cache

 Cores:              10 cores /20 threads

 HT:                    yes
 RAM type:        DDR4
 RAM size:        32 GB

 NB Freq:           2.7 GHz
 RAM Freq:        3600 MHz 
 RAM Latency:  CL16
 GPU:                 Gigabyte GTX 1080 G1 Gaming

 STMark:           2611 (4.6 GHz up to 4 cores, 4.2 GHz up to 10 cores/20 threads; HT: ON)

Spoiler

Per_core_speed_HT_4.6-4.2.jpg.5f32648bb3e3a8f74a59de96088c5ef0.jpg

 STMark:           2521 (4.5 GHz all cores; HT: OFF)

Spoiler

sync_all_cores_noHT_4.5.jpg.e6c9b208830849570a27ca57765ee338.jpg

 STMark:           2627 (4.7 GHz all cores; HT: OFF)

Spoiler

sync_all_cores_noHT_4.7.jpg.3ec5e12cd8e1252c30d003b9c1504b55.jpg

 

It is of note that I use no AVX offset in all conditions. For both 4.5 Ghz on all cores and 4.7 GHz on all cores I use AVX512 offset 6. It would just toast my system if I didn't do so, AVX as such can be handled borderline, but 4.7 GHz would require a HUGE cooler to let it run at 100% load for a long time. It is of note that IL-2 hardly gives load on the CPU besides when missions are loading. Then there is a real power surge. But while flying, temps drop from spiking in the high 90's (Celsius) back to the 60's, even at 4.7 GHz, while the coolerfan goes almost off. The CPU however remains set at the set highest frequency.

 

I had to redo my last track. I just saw that i omitted the landscape drawing back to 20 km. This of course had an impact on my previous results.

 

 

Now the results:

 

4.6 GHz up to 4 cores, 4.2 GHz up to 10 cores/ 20 threads; HT: ON

2020-01-05 13:55:52 - Il-2
Frames: 20596 - Time: 180000ms - Avg: 114.422 - Min: 70 - Max: 195


4.5 GHz all cores, HT off (10 cores):

2020-01-05 13:00:36 - Il-2
Frames: 22085 - Time: 180000ms - Avg: 122.694 - Min: 54 - Max: 226

 

4.7 GHz all cores, HT off (10 cores):

2020-01-05 13:40:15 - Il-2
Frames: 24963 - Time: 180000ms - Avg: 138.683 - Min: 79 - Max: 229

 

 

 

Comparing the frequencies and FPS over time gives this:

chart.thumb.jpg.94be45c8101baf9666f908ff6f50923f.jpg

 

You can see that if the CPU cores are not synchronized but are given varying top frequencies, for the game they will work at their slowest top speed. In my case, the grey chart basically reflects a "4.2 GHz" system, even though individual cores are allowed to exceed that frequency (They go sometimes even slightly above 4.6 GHz!) but in total, we make bad use of the system.

As soon as I force the cores to collectively scale frequencies, then I get results matching the "top core", hence the 4.5 GHz system that is actually slower than the "4.2 GHz system" in single thread (!) Passmark is faster running IL-2. Forcing 4.7 GHz is just about the edge where cooling becomes enough of a pain to not bother further. While IL-2 runs cool on 4.7 GHz and in fact 4.9 GHz could borderline be achieved, as soon as a thread puts constant (real!) load on a core (like passmark or any productivity app), then meltdown is imminent.

 

As a conclusion we can say that one of the prime problems of the IL-2 code is that it doesn't a constantly running thread (such as FSX or P3D have; or Passmark) that stays locked on a core, but instead seems to be a thread that is spawned and terminated constantly. This makes Windows distribute that thread in pieces all over the cores. This way, idle (and thus low clocked) cores always get a piece of that, THEN have to clock up, deliver and terminate the task. This way of distributing a single task over the cores can effectively lock all cores in a "used" state and are therefore in a lower boost scenario, despite the CPU drawing very little electricity.

 

A problem I see with IL-2 code is not that it is "single core", but that it is limited by a single core frequency while acting such that the system assumes severe load on all cores! The effort for "all core OC" is just the way to beat this problem, as there is no way to force Windows to assign a certain thread to a specific CPU. If we could do that, all we needed is a CPU that allows you to OC one core to >5 GHz with the main thread locked to that core. All the other cores old run at 3 GHz or less, you'd still get the same FPS as an "all core OC".

 

If Windows had a proper scheduler that programmers could use, it would make the silicon lottery so much easier. How easy is it to find a fantastic core among 12? Certainly much much easier then finding 12 such cores in a single CPU die. Now we have to OC all cores for the sake of one.

 

If we compare the Passmark scores, we can see that the system doing the worst in IL-2 is actually the best in Passmark. And certainly the best in real life work scenarios. My "4.2 GHz system" matches the 4.7 GHz OC'd system in single thread while beating it clearly in multithread. At less heat. AND I can run Handbrake overnight with no problems, AVX512 offset 0!

 

 

  • Like 1

Share this post


Link to post
Share on other sites
18 hours ago, ICDP said:

game mode which essentially turns off the 6 slower cores.  Normally this makes zero difference or in some cases gives worse results.  Yet with Il-2 I got a mild boost and the cores were running between 4.5 and 4.6GHz, which is up ~100MHz.

 

Thank you for this test. It gives a small boost. Good to know it.

Share this post


Link to post
Share on other sites
59 minutes ago, ICDP said:

Indeed an i5 will also hit 5GHz and give better results than a 3600X.

Only if all core OC'd, which I think you implied.

 

To compare CPU's always compare them at reaspective clockspeeds of the slowest core at the ALL CORE usage scenario.

 

The fewer cores you have, the higher the chance getting all cores to a similar high clock. Then again, the silicon lottery is not fair, as there is no random binning. Everything that really clocks well goes to the 9900 series, giving other bins less margin.

Share this post


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

i could get a stable 24/7 5.2ghz with manual lower voltages with more testing and small RAM O/C if needed.

 

Thank you very much for this test!

 

You got a quite good result (3rd in table). I wonder what would be your achievement if you could run the test at 5.2 with a reasonable manual voltage. Just to see if you achieve comparable results to the 9900K at 5.2.

The 2080Ti should not influence in theory. We need someone with a 9900K at 5.2 and a 1080Ti.

Share this post


Link to post
Share on other sites
18 hours ago, ICDP said:

So I used Ryzen Master to turn on game mode which essentially turns off the 6 slower cores.  Normally this makes zero difference or in some cases gives worse results.  Yet with Il-2 I got a mild boost and the cores were running between 4.5 and 4.6GHz, which is up ~100MHz. 

THIS! What I see in the Intel chips as well.

 

Best is to selectively OC each core to the max, individually. Then out of 6 or 8, you can select the 4 best and use just them for running IL2 at their max frequency.

Share this post


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

THIS! What I see in the Intel chips as well.

 

Best is to selectively OC each core to the max, individually. Then out of 6 or 8, you can select the 4 best and use just them for running IL2 at their max frequency.

 

As you stated previously in your excellent post, Il-2 core bounces and this can cause problems with the chiplet/shitlet design AMD went with on Ryzen 3900X.  If Il-2 jumps from a well binned core at 4.5-4.6 GHz to one running at 4.3 it results in an FPS drop.  Turning off those slower cores helps Il-2 performance.  I have tried to set CPU affinity for Il-2 but it doesn't seem to work.

15 minutes ago, chiliwili69 said:

 

Thank you very much for this test!

 

You got a quite good result (3rd in table). I wonder what would be your achievement if you could run the test at 5.2 with a reasonable manual voltage. Just to see if you achieve comparable results to the 9900K at 5.2.

The 2080Ti should not influence in theory. We need someone with a 9900K at 5.2 and a 1080Ti.

 

I would be interested as well but don't break your CPU trying.

Edited by ICDP

Share this post


Link to post
Share on other sites
21 minutes ago, ICDP said:

I have tried to set CPU affinity for Il-2 but it doesn't seem to work.

I've tried the same today while doing the tests, similar result.

 

I might try do deactivate cores, but on the Skylake it is not straightforward finding cores with diversity. There is a ranking of the CPU cores in "Turbo Boost max 3" software once you bring up the UI. But I'm not sure what makes the odd succession of cores in the list. Is that diversity, top to bottom? I mean, Intel must know that, as it becomes a datum while binning the chips.

 

So I might disable 5 of my 10 cores and gun for 4.9 GHz, HT off. The heat should still be within cooler capabiliies and 4.9 is about the edge where BS appear. I might have to up volt.

 

The sole reason I'm doing this is that I'm considering getting a Rift S.

Share this post


Link to post
Share on other sites

Having read more into core jumping I believe it is done to help with thermals.  One core constantly loaded is harder to cool than a single thread jumping between cores.  Hopefully Windows/MS can learn how to improve their core scheduler to account for best reported cores.

  • Thanks 1

Share this post


Link to post
Share on other sites
Posted (edited)

@chiliwili69, I run the benchmark in VR with the results below:

 

VR HMD: Reverb

Freq HZ: 90

SteamVR SS: 120% (I know the others are at 100% but with my GPU I believe the difference is negligible)

frames: 9448

min: 37

max: 91

avg: 52.489

 

 

In about two weeks time I will have a new PC - built around i9 9900k and 2080ti. I will then run the tests again (both on screen and in VR) to contribute to your table.

 

A request to the wise: Shall I go for ddr4 16gb @ 4000mhz or rather ddr4 32gb @ 3200mhz? (similar expenditure; the guy building the rig is advising the latter option to make it future proof but I feel the former one suits better the current VR simming needs...?).

 

Thanks!

 

Milan

 

Edited by apollon01

Share this post


Link to post
Share on other sites
Posted (edited)
33 minutes ago, apollon01 said:

@chiliwili69, I run the benchmark in VR with the results below:

 

VR HMD: Reverb

Freq HZ: 90

SteamVR SS: 120% (I know the others are at 100% but with my GPU I believe the difference is negligible)

frames: 9448

min: 37

max: 91

avg: 52.489

 

 

In about two weeks time I will have a new PC - built around i9 9900k and 2080ti. I will then run the tests again (both on screen and in VR) to contribute to your table.

 

A request to the wise: Shall I go for ddr4 16bg @ 4000mhz or rather ddr4 32bg @ 3200mhz? (the guy building the rig is advising the latter option to make it future proof but I feel the former one suits better the current VR simming needs...?).

 

Thanks!

 

Milan

 

 

I would go with 32GB 3200 and overclock it, but getting the DDR 4000 at 2x8GB means you can still upgrade in the future.  Have a look here for some DDR comparison tests on a 9900K.

 

 

Edited by ICDP

Share this post


Link to post
Share on other sites
Posted (edited)
2 hours ago, ZachariasX said:

If we could do that, all we needed is a CPU that allows you to OC one core to >5 GHz with the main thread locked to that core. All the other cores old run at 3 GHz or less, you'd still get the same FPS as an "all core OC".

 

 

  The reason windows sheduler makes processes to jump cores is heat management and CCX/CCD management. Now Intel mainstream CPUs don't have this problem because all cores sit on the same CCX and share cache, so when Windows scheduler switches cores, the data is already in the cache and the next core can pickup from there. On Ryzen, where cores can be on separate CCX or even different dies, the cache is built on the CCX, so when windows scheduler pairs up cores from different CCX, it has to send the data from cache to Ram and then to the next core.

There was a very good explanation from AMD, few weeks ago, when they talked about best cores in Ryzen Master Overclocking application. 

 

Here's an article about it:

 

https://www.techpowerup.com/261377/amd-admits-stars-in-ryzen-master-dont-correspond-to-cppc2-preferred-cores

 

 

 

Edited by Jaws2002
  • Thanks 2

Share this post


Link to post
Share on other sites
32 minutes ago, ICDP said:

Hi Milan, what is the PC spec you run this on please?

Look here

 

I have already run the test on screen and my specs are listed in the post above and chili's table linked in the first post.

 

Thanks for a quick reply. The "upgrade later" option makes sense to me. I will check the video now 👍

Share this post


Link to post
Share on other sites
Posted (edited)
On 1/5/2020 at 6:52 PM, apollon01 said:

 Shall I go for ddr4 16gb @ 4000mhz or rather ddr4 32gb @ 3200mhz? (similar expenditure; the guy building the rig is advising the latter option to make it future proof but I feel the former one suits better the current VR simming needs...?).

 

Thanks!

 

I don't have a clue about the impact memory has on VR, but 9900k can make good use of fast memory.  If you only build the PC for this game, the 16 GB of fast memory will make a difference. If you plan to play DCS i'd take the 32GB option and try to overclock it.

 

 

[edited]

Edited by SYN_Haashashin
One more politics from you and it will be the last time.
  • Upvote 1

Share this post


Link to post
Share on other sites
On 1/5/2020 at 5:25 PM, ICDP said:

my experience with Pimax 5K

 

Quite right in all your explanation.

These charts show the relative performance for 4K, so in that case, (and probably for games less dependent on CPU than IL-2) it is better to go for a 2080Ti than for a 9900K.

 

But IL-2 VR is another matter. It depends a lot of the VR device. In the case of Pimax5K it requires a lot of GPU power, but others like Rift-S or even Index require less GPU power, so a 1080Ti could be still quite OK.

I was experimenting a lot with the Pîmax5K+ and the SS values (you can see that in the "SS per device" tab of the online sheet). For the Pimax5K+ definetely it is worth to invest more in GPU, but for Rift-S, O+ and Index I am not sure.

 

For example, see my case with 4790K and 1080Ti and Index.  What upgrade wold give more fps in VR?

This year I am waiting for the intel 10th for PC and also new GPUs. Most likely I will upgrade first Mobo+CPU+RAM but will keep 1080Ti. We will see.

Share this post


Link to post
Share on other sites

Some runs at 5.2 

 

Motherboard: MSI Z390 Gaming Edge
 CPU:                 9600K
 CPU Freq:        5.2 Ghz
 L3 cache:        9 MB

 Cores:              6 

 HT:                    -/Off 
 RAM type:        DDR4
 RAM size:        16 GB

 NB Freq:          - Auto
 RAM Freq:        3226 MHz 
 RAM Latency:  CL14
 GPU:                 MSI  GTX 1080Ti Seahawk
 STMark:           3180  (averaged)

 

 

2020-01-07 16:38:02 - Il-2
Frames: 27172 - Time: 180000ms - Avg: 150.956 - Min: 105 - Max: 240

2020-01-07 16:41:48 - Il-2
Frames: 27052 - Time: 180000ms - Avg: 150.289 - Min: 105 - Max: 239

2020-01-07 16:45:59 - Il-2
Frames: 27040 - Time: 180000ms - Avg: 150.222 - Min: 105 - Max: 239

 

Cheers, Dakpilot

Share this post


Link to post
Share on other sites

I should be getting some new RAM today, and I notice my results are on the 4.002 version of the game. @chiliwili69 if I'm going to spend time testing, could you confirm that these steps might be sensible? Can you think of any other testing that would be useful?

  • Re-benchmark using yesterday's NVidia driver, low-latency off, 5.0ghz 8086K, old 3200 CL16 RAM.
  • Install new RAM, benchmark using 3600 CL16 RAM, don't change anything else.
  • Overclock RAM as high as I can, maybe tighten timings, bench again.
  • Increase north bridge / ring ratio / uncore as much as I can, bench again.
  • Do a benchmark run at 5.2ghz if I can get the chip stable (my cooling isn't good enough for this really, but I could probably squeeze a benchmark run out of it).

I can't guarantee I'll have loads of time, I will definitely do the things in bold though.

Share this post


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

if I'm going to spend time testing, could you confirm that these steps might be sensible? Can you think of any other testing that would be useful?

  • Re-benchmark using yesterday's NVidia driver, low-latency off, 5.0ghz 8086K, old 3200 CL16 RAM.
  • Install new RAM, benchmark using 3600 CL16 RAM, don't change anything else.
  • Overclock RAM as high as I can, maybe tighten timings, bench again.
  • Increase north bridge / ring ratio / uncore as much as I can, bench again.
  • Do a benchmark run at 5.2ghz if I can get the chip stable (my cooling isn't good enough for this really, but I could probably squeeze a benchmark run out of it).

 

OK, the three first steps are quite logical. But you can also do the first three points in VR, just to see if the gain is also in VR.

 

The last two would be also nice to know it. You should know that IL-2 in monitor (even VR) is really much much less demanding than Prime95 stress tests. I was comparing the load and temps of Prime95 vs IL-2 in the old benchmark thread. So you only need your CPU to be stable for IL-2 for the test not for Prime95.

Did you consider to put a AIO liquid cooling?

On 1/5/2020 at 6:11 PM, ZachariasX said:

A problem I see with IL-2 code is not that it is "single core", but that it is limited by a single core frequency while acting such that the system assumes severe load on all cores! The effort for "all core OC" is just the way to beat this problem, as there is no way to force Windows to assign a certain thread to a specific CPU. If we could do that, all we needed is a CPU that allows you to OC one core to >5 GHz with the main thread locked to that core. All the other cores old run at 3 GHz or less, you'd still get the same FPS as an "all core OC".

 

Thank very much you for your test.

Your 138 fps is a nice surprise for a freq of 4.7GHZ. I think your four-channel memory is helping here.

 

Regarding your comments, you can always de-activate 6 or 4 cores (the worse ones). In this way you are forcing the system to run the threads in certain cores. It will produce less heat, so you would be able to even increase the freq further.

 

Regarding your NB frequency, it is "only" 2700MHz. have you tried to increase it as well?

Share this post


Link to post
Share on other sites
On 1/5/2020 at 7:08 PM, ZachariasX said:

So I might disable 5 of my 10 cores and gun for 4.9 GHz, HT off. The heat should still be within cooler capabiliies and 4.9 is about the edge where BS appear. I might have to up volt.

 

The sole reason I'm doing this is that I'm considering getting a Rift S

 

OK, I see you were already planning to do that. Let´s see how it goes.

I think with your actual numbers you will be quite OK with the Rift-S, it only need to run at 80Hz and this helps.

And if you don´t reach that the ASW 2.0 tech could help in complex scenarios.

On 1/5/2020 at 7:52 PM, apollon01 said:

SteamVR SS: 120% (I know the others are at 100% but with my GPU I believe the difference is negligible)

 

Thanks for this test, but could you let me know what resolution per eye is showing SteamVR for 120%SS?

 

Good to know you go for the 9900K. But perhaps the 9900KS could be a bit better for OC?

 

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. 

 

I think the best true latency available at "good" price is 4226MGHz with CL16 :

https://pcpartpicker.com/product/VmgQzy/corsair-vengeance-lpx-16-gb-2-x-8-gb-ddr4-4266-memory-cmk16gx4m2k4266c16

 

If not you can get 3200 with CL14 which is also quite good.

8 hours ago, Dakpilot said:

 CPU:                 9600K
 CPU Freq:        5.2 Ghz

Many thanks for this extra test at 5.2GHz!.

It is the first Intel non-9900K at 5.2. But it gives about 10-20 less fps than the 9900K (but it was with v 4.002).

You currently has the record in v4.003 tests.

For some reason your system doesn´t gives much more fps beyond 5.0GHz.

 

Perhaps the L3 cache size could make a difference here. (the other NB freq and RAM true lat at quite similar)

  • Thanks 2

Share this post


Link to post
Share on other sites
Posted (edited)

Today I have been running the Remagen benchmark about 70 times! Just two runs for each single option of the graphics settings.

Something that I always wanted to do in all these years, just to know the impact of every single option of the graphics settings in 1080p monitor (not constrained by GPU)

I have compared it with the avg fps obtained with the settings indicated in the instructions (the base case) and I have changed just one option at a time, and calculated the fps gain (positive or negative).

30850517_Gainfps.thumb.png.45dfabee6213234985e088c0f415db90.png


All the results are compiled in a tab called "GainMap" of the online sheet of the test. I think the table is self-explanatory.

 

Some conclusions:

- The presets have not the biggest impact. The difference between High and Ultra is only 4 fps.
- Shadows have a fair impact. Based on this I think I will choose "Medium" for VR.
- Mirrors have the biggest impact. I can live without mirrors in VR. (BTW, one nice thing of mirror in Complex mode is that I can see my own smoke after being hitted)
- As we know, Clouds just load GPU. So there is very small impact on CPU.
- Distant building Off gives me 6.7 fps more. I can live without that in VR.
- All the SSAO, HDR and Sharpen gave me some small fps bump. I can´t explain that. Perhaps they speed-up GPU, but it is very close to the noise level (about 1.5 fps)

 

Here I show some graphs for the impact of Shadows and Mirror in monitor:

shadows.thumb.png.e59a5bfd03376626c197d90073fad521.png

mirrors.thumb.png.97d08ebdf5370cc3347e451d943abcfb.png

 

With all this info I run the benchmark again with OP settings but with Shadows=Medium (+9.1), Mirror=Off (+14.7), HorDrawDist=100Km (+2.5) and Distant Building Off (+6.7).
So, 9.1+14.7+2.5+6.7 is 33 fps more if this behave in perfect linearity (ie total effect=sum of individual effects). But I obtained 26.3 more, which is not bad.

 

I run again the benchmark twice in VR (Index at 80Hz, 100%SS): with the OP settings (44.8 fps) and with my new VR Settings (67.5 fps), so I got  a gain of 22.7 fps in VR!
In the below graph I compare fps in monitor (left axis) with VR (right axis) with my VR settings. I can see that I don´t reach the 80fps when in monitor I am under 130fps (shooting bombers and low flight over Remagen)
So, checking the performance in monitor is a valid way to know what should I expect in VR (if GPU is not a constraint).

 

MonvsVR.thumb.png.1328a827dc5feaef84abbcce0a9fdc3f.png

Edited by chiliwili69
  • Like 3
  • Thanks 7
  • Upvote 1

Share this post


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

 

 

Many thanks for this extra test at 5.2GHz!.

It is the first Intel non-9900K at 5.2. But it gives about 10-20 less fps than the 9900K (but it was with v 4.002).

You currently has the record in v4.003 tests.

For some reason your system doesn´t gives much more fps beyond 5.0GHz.

 

Perhaps the L3 cache size could make a difference here. (the other NB freq and RAM true lat at quite similar)

 

For the 5.2 I just upped the block to 52 and left everything else at auto, so not a carefully optimized O/C, I did not monitor temps during the runs but they had dropped to 40/50s just after the run, auto sets the vcore to 1.4+ V so I may have got isolated thermal throttling (guessing) Am sure with more careful set-up I could drop the voltage quite a bit. Just do not have time

 

The ST Mark got quite a boost, so I left it at that. 

 

My MB is at the bottom end of "enthusiast" level so may have an effect, considering in my country 9600k is less than half of  9900K and  cheaper MB I am quite happy with value for money/performance ratio and I need good cooling anyway, was nearly 30 degrees ambient when I was testing. 

 

Cheers, Dakpilot 

Share this post


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

Regarding your NB frequency, it is "only" 2700MHz. have you tried to increase it as well?

So far I only went with the proposed "XMP" settings, and this is what came along with it. I then just upped turbo multiplier synch on all cores.

 

Initially, I played with it, but it gave me little real world gain. (Haven't tested on IL-2 specifically.) Then again, that was several BIOS versions ago. I'll definitely have a look at it when I crank the CPU up beyond 4.7+ GHz with maybe 6 cores. I might hit 4.9, but Skylake doesn't like much more than that. Maybe upvolting, but so far I didn't want to do that. The CPU was too pricey for such abuse.

Share this post


Link to post
Share on other sites
Posted (edited)
11 hours ago, chiliwili69 said:

could you let me know what resolution per eye is showing SteamVR for 120%SS?

Of course, sorry for forgetting that:

 

120% SteamVR SS gives me 2408x2361

 

Thanks for the HW tips. The build is already ongoing so it is going to be i9 9900k and ddr4 32gb @ 3600mhz CL16.

 

I will update you with tests run on the new machine 👍 

Edited by apollon01

Share this post


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

Today I have been running the Remagen benchmark about 70 times! Just two runs for each single option of the graphics settings.

Something that I always wanted to do in all these years, just to know the impact of every single option of the graphics settings in 1080p monitor (not constrained by GPU)

I have compared it with the avg fps obtained with the settings indicated in the instructions (the base case) and I have changed just one option at a time, and calculated the fps gain (positive or negative).

 

<><> SNIP <><>

 

With all this info I run the benchmark again with OP settings but with Shadows=Medium (+9.1), Mirror=Off (+14.7), HorDrawDist=100Km (+2.5) and Distant Building Off (+6.7).
So, 9.1+14.7+2.5+6.7 is 33 fps more if this behave in perfect linearity (ie total effect=sum of individual effects). But I obtained 26.3 more, which is not bad.

 

I run again the benchmark twice in VR (Index at 80Hz, 100%SS): with the OP settings (44.8 fps) and with my new VR Settings (67.5 fps), so I got  a gain of 22.7 fps in VR!
In the below graph I compare fps in monitor (left axis) with VR (right axis) with my VR settings. I can see that I don´t reach the 80fps when in monitor I am under 130fps (shooting bombers and low flight over Remagen)
So, checking the performance in monitor is a valid way to know what should I expect in VR (if GPU is not a constraint).

 

 

 

Excellent work Chillwili69, BIG thanks from another VR user.

Share this post


Link to post
Share on other sites
Posted (edited)

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.

Isn't the headset ( if 90 Hz or 80 Hz) still going to cut it down to 40 or 45 fps interpolating every other frame to equal 80/90 fps?

So is it even worth cutting some graphics settings down to get this gain if the headset is going to basically throw those fps away and run with reprojection or ASW anyway? In that case why not run with some higher graphics...

 

I guess if one was close to maintain a constant 90 fps , that gain might put them over the top where they never drop below, but not sure there is a system yet that 

can do so.

 

 

Edited by dburne

Share this post


Link to post
Share on other sites

But if you can maintain 60+ fps you can put Reverb in 60Hz mode and not have reprojection

Share this post


Link to post
Share on other sites
12 minutes ago, Goblin said:

But if you can maintain 60+ fps you can put Reverb in 60Hz mode and not have reprojection

Currently 60Hz is distorted in HP Reverb.

Yesterday was released the newest version of WMR hope is fixed

Share this post


Link to post
Share on other sites
Posted (edited)

I re-ran the benchmark after receiving my new ram. Gained a few fps but not as much as I was hoping though. It has seemed to get rid of some of the stuttering though which is nice.

 

I first ran it again with my original ram (3000mhz CL16) to eliminate any affect driver updates etc might have since last time:

 

Frames: 25747 - Time: 180000ms - Avg: 143.039 - Min: 98 - Max: 238

 

Then with my new 3600mhz CL16:

 

Frames: 27282 - Time: 180000ms - Avg: 151.567 - Min: 104 - Max: 242

 

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

 

 

Edited by Loki_1982b_rock1080

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...