chiliwili69 Posted March 9, 2023 Author Posted March 9, 2023 We have been testing new servers in the office with the 13900KS with Aircooling! (AK620). Our application is mainly a singlethreaded appl, so no thermal limits. We achieved 5333 on the Passmark CPU Single threaded, just out of the box.
thermoregulator Posted March 9, 2023 Posted March 9, 2023 55 minutes ago, chiliwili69 said: Is your monitor a 120Hz monitor? (Remember that Fraps always shows the monitor freq, either 60 or 120 or whatever, before you launch the mission. While you run the mission is uncaped). Yes, it is LG OLED TV. I will try it once more, but he last time, the result was capped to 120 Hz, at which it sat all the time during the benchmark. G-sync was turned off, v-sync too.
MilitantPotato Posted March 13, 2023 Posted March 13, 2023 (edited) On 3/5/2023 at 11:16 PM, dgiatr said: So after doing all tests 1080 cpu, vrtest1, vrtest2 and vrtest3 with my new cpu and some basic cpu overclocking here are my results : Motherboard: Asus z690 P D4 CPU: i9 13900kf CPU Freq: 5900 Mhz, OCTVB to boost1 , (P cores 2x59, 4x58, 6x57, 8x56 and E cores to: 16x44) Cores: 8P, 16E Threads: 32 RAM: 16 GB DDR4 RAM Freq: 3200 Mhz XMPI NB Freq: 4600 Mhz RAM timings: 14 14 14 34 GPU: 3090 Ti with default clocks Basically, at the beginning i have done some cpu undervolting in order to achieve lower temps from my stock cpu BIOS settings using the following BIOS settings : MCE=Enforce All Limits LLC#4 DC_LL = 0.70 ( only with that setting vid core matches to v core....) AC_LL = 0.12 IA VR Voltage limit = AUTO and after that i created frequency groups for P-Cores and done some "thermal opportunity" overclocking by setting OCTVB to : +1Boost, so my P cores work to the following combinations: 2x59, 4x58, 6x57, 8x56 and E cores to: 16x44. Generally i dont run any other heavy multi core applications in my P.C. and with that settings I dont suffer from any instability issues playing heavy il2 scenarios with G2 So with that settings i get only about 3 more fps during Synvander 1080 cpu test comparing to 1080 cpu test with stock cpu bios settings BUT my cpu runs WAY MORE cooler than before with no thermal throttling at all during dense il2 playing with G2 headset. So : 1080 CPU TEST WITH OCTVB TO BOOST1 2023-03-04 08:15:41 - Il-2 Frames: 9475 - Time: 60000ms - Avg: 157.917 - Min: 133 - Max: 204 2023-03-04 08:19:48 - Il-2 Frames: 9537 - Time: 60000ms - Avg: 158.950 - Min: 139 - Max: 201 2023-03-04 08:22:00 - Il-2 Frames: 9466 - Time: 60000ms - Avg: 157.767 - Min: 137 - Max: 207 VR TEST1 WITH STEAM VR 2023-03-04 09:27:32 - Il-2 Frames: 4966 - Time: 60000ms - Avg: 82.767 - Min: 66 - Max: 91 2023-03-04 09:29:25 - Il-2 Frames: 4977 - Time: 60000ms - Avg: 82.950 - Min: 65 - Max: 91 2023-03-04 09:31:19 - Il-2 Frames: 4944 - Time: 60000ms - Avg: 82.400 - Min: 66 - Max: 91 VR TEST1 WITH OPENCOMPOSITE+OPENXR TOOLKIT 2023-03-04 12:00:35 - Il-2 Frames: 5344 - Time: 60000ms - Avg: 89.067 - Min: 84 - Max: 92 2023-03-04 12:03:04 - Il-2 Frames: 5336 - Time: 60000ms - Avg: 88.933 - Min: 84 - Max: 92 2023-03-04 12:04:49 - Il-2 Frames: 5255 - Time: 60000ms - Avg: 87.583 - Min: 80 - Max: 92 VR TEST2 WITH STEAM VR 2023-03-04 10:04:20 - Il-2 Frames: 4474 - Time: 60000ms - Avg: 74.567 - Min: 61 - Max: 91 2023-03-04 10:06:16 - Il-2 Frames: 4498 - Time: 60000ms - Avg: 74.967 - Min: 60 - Max: 91 2023-03-04 10:08:15 - Il-2 Frames: 4461 - Time: 60000ms - Avg: 74.350 - Min: 62 - Max: 91 VR TEST2 WITH OPENCOMPOSITE+OPENXR TOOLKIT ( i see a gain of about 12 fps comparing to steamVR!!) 2023-03-04 11:34:33 - Il-2 Frames: 5190 - Time: 60000ms - Avg: 86.500 - Min: 79 - Max: 92 2023-03-04 11:36:24 - Il-2 Frames: 5101 - Time: 60000ms - Avg: 85.017 - Min: 79 - Max: 91 2023-03-04 11:41:39 - Il-2 Frames: 5202 - Time: 60000ms - Avg: 86.700 - Min: 80 - Max: 91 2023-03-04 11:43:40 - Il-2 Frames: 5162 - Time: 60000ms - Avg: 86.033 - Min: 80 - Max: 91 2023-03-04 11:45:58 - Il-2 Frames: 5101 - Time: 60000ms - Avg: 85.017 - Min: 78 - Max: 91 VR TEST 3 WITH STEAM VR 2023-03-04 10:23:09 - Il-2 Frames: 2621 - Time: 60000ms - Avg: 43.683 - Min: 38 - Max: 46 2023-03-04 10:25:28 - Il-2 Frames: 2551 - Time: 60000ms - Avg: 42.517 - Min: 37 - Max: 46 2023-03-04 10:27:21 - Il-2 Frames: 2607 - Time: 60000ms - Avg: 43.450 - Min: 37 - Max: 46 VR TEST 3 WITH OPENCOMPOSITE+OPENXR TOOLKIT( i see a gain of about 15 fps comparing to steamVR!!!! ) 2023-03-04 10:56:26 - Il-2 Frames: 3537 - Time: 60000ms - Avg: 58.950 - Min: 48 - Max: 67 2023-03-04 10:58:20 - Il-2 Frames: 3527 - Time: 60000ms - Avg: 58.783 - Min: 48 - Max: 67 2023-03-04 11:00:30 - Il-2 Frames: 3501 - Time: 60000ms - Avg: 58.350 - Min: 48 - Max: 66 Finally i set OCTVB to : +2Boost, so my cpu will run to the following combinations : P cores 2x60, 4x59, 6x58, 8x57 and E cores to: 16x44) and my results are the following : 1080 CPU TEST WITH OCTVB TO BOOST2 2023-03-05 09:33:09 - Il-2 Frames: 9425 - Time: 60000ms - Avg: 157.083 - Min: 137 - Max: 201 2023-03-05 09:35:57 - Il-2 Frames: 9321 - Time: 60000ms - Avg: 155.350 - Min: 137 - Max: 200 2023-03-05 09:38:26 - Il-2 Frames: 9348 - Time: 60000ms - Avg: 155.800 - Min: 136 - Max: 199 as you can see i dont get any more fps during 1080 cpu test , i also suffer from thermal throttling during heavy il2 playing. I also suspect that i cannot get more fps than 157 due to DDR4 memory plus mobo d4 combination? ...your comments are welcomed... I don't want to be a kill joy, but there's zero chance openxr, which reduces cpu load by a marginal percentage gives 30% gains to a 4090s abilities. Prior to this result my 5800x3d on steamvr scored the same as a heavily tuned 13600k on openxr with vr3. There's either a configuration or settings issue with these results. Even my personal testing shows zero benefit to gpu bottle necked results with open xr, and at best a few percentage points with openxr in a locally run replay heavy in ai Edited March 13, 2023 by MilitantPotato
dgiatr Posted March 13, 2023 Posted March 13, 2023 (edited) 1 hour ago, MilitantPotato said: I don't want to be a kill joy, but there's zero chance openxr, which reduces cpu load by a marginal percentage gives 30% gains to a 4090s abilities. Prior to this result my 5800x3d on steamvr scored the same as a heavily tuned 13600k on openxr with vr3. There's either a configuration or settings issue with these results. Even my personal testing shows zero benefit to gpu bottle necked results with open xr, and at best a few percentage points with openxr in a locally run replay heavy in ai I don't think i have any different settings in opencomposite while running Synvander Benchmark with VR3 test , i didn't enable any upscaling via openxr toolkit , I checked to keep the same resolution between Openxr and Steamvr. Vr3 test loads gpu and cpu at the same time, so my i9 13900 kf cpu after overclocking to 5.9 ghz gets 158 fps during 1080cpu test which i think affects the number of 58,8 FPS i get during VR3 test with opencomposite . Edited March 13, 2023 by dgiatr
MilitantPotato Posted March 13, 2023 Posted March 13, 2023 (edited) Your results are so extreme that some sanity check should be done, double check proper resolution and ingame settings. This benchmark is at best a check of how well a cpu handles being choked by ai. The 8x msaa exteme clouds is either valid as a gpu metric or dependent on gpu vram bandwidth. Neither explains a 30% increase in gpu throughput when previous results show zero benefit from higher teir cpus. 4080s getting similar results makes me think vr3 is fundamentally flawed. Not trying to be a jerk but something is wrong Edited March 13, 2023 by MilitantPotato
MilitantPotato Posted March 13, 2023 Posted March 13, 2023 (edited) As an alternative there could be a serious issue with steamvr and core parking on new ryzen chips, and zen5 is just 30% better at utilizing 4090s. A quick check for that would be disabling the 2nd ccd and running the benchmark with steamvr. If that's the case 7800x3ds will absolutely run sim gaming until the scheduler is fixed Edited March 13, 2023 by MilitantPotato
dgiatr Posted March 13, 2023 Posted March 13, 2023 13 minutes ago, MilitantPotato said: Not trying to be a jerk but something is wrong No not at all! your remarks are welcomed. As you say, since i double checked ingame settings and resolution ( i activated override resolution in openxr toolkit ) to be the same between Steamvr and openxr ( i also disabled any upscaling method in openxrtoolkit, enabled only CAS sharpening to 100%) what else to check that might affects Steamvr or openxr performance?
MilitantPotato Posted March 13, 2023 Posted March 13, 2023 (edited) Cas is an upscaling technology, that alone boosts fps by about 30% (same as nis/fsr at .70) as far as I'm aware Edited March 13, 2023 by MilitantPotato
TG-55Panthercules Posted March 13, 2023 Posted March 13, 2023 1 hour ago, MilitantPotato said: Cas is an upscaling technology, that alone boosts fps by about 30% (same as nis/fsr at .70) as far as I'm aware Maybe I'm doing something wrong, but I've tried it twice (running OpenXR/OpenComposite instead of Steam VR, and with NIS at .70) a few weeks apart, after watching a couple of different videos to make sure I was setting things up right, and my results on the benchmarks (options 1 and 2 - didn't try 3) and with my usual QMB test missions were essentially the same as with Steam VR (maybe 2 or 3 FPS better, but negligible) and visual quality was way worse. I've gone back to using SteamVR with the settings you sent me in PM and things are looking much better again.
MilitantPotato Posted March 13, 2023 Posted March 13, 2023 This benchmark is cpu limited in all but vr3, with vr3 relying on maxed msaa to increase gpu load which may or may not be a valid setup. If 4080s post similar numbers to a 4090 in vr3 it's just testing vram bandwidth
dgiatr Posted March 14, 2023 Posted March 14, 2023 (edited) 16 hours ago, MilitantPotato said: Cas is an upscaling technology, that alone boosts fps by about 30% (same as nis/fsr at .70) as far as I'm aware Results without CAS and with CAS for VR1, VR2 and VR3 tests as follows : VR TEST1 STEAM VR 2023-03-04 09:27:32 - Il-2 Frames: 4966 - Time: 60000ms - Avg: 82.767 - Min: 66 - Max: 91 2023-03-04 09:29:25 - Il-2 Frames: 4977 - Time: 60000ms - Avg: 82.950 - Min: 65 - Max: 91 2023-03-04 09:31:19 - Il-2 Frames: 4944 - Time: 60000ms - Avg: 82.400 - Min: 66 - Max: 91 VR TEST1 OPENCOMPOSITE WITHOUT CAS 2023-03-04 13:41:12 - Il-2 Frames: 5314 - Time: 60000ms - Avg: 88.567 - Min: 83 - Max: 92 2023-03-04 13:43:02 - Il-2 Frames: 5312 - Time: 60000ms - Avg: 88.533 - Min: 82 - Max: 92 2023-03-04 13:44:50 - Il-2 Frames: 5257 - Time: 60000ms - Avg: 87.617 - Min: 81 - Max: 92 VR TEST1 OPENCOMPOSITE WITH CAS 2023-03-04 12:00:35 - Il-2 Frames: 5344 - Time: 60000ms - Avg: 89.067 - Min: 84 - Max: 92 2023-03-04 12:03:04 - Il-2 Frames: 5336 - Time: 60000ms - Avg: 88.933 - Min: 84 - Max: 92 2023-03-04 12:04:49 - Il-2 Frames: 5255 - Time: 60000ms - Avg: 87.583 - Min: 80 - Max: 92 VR TEST2 STEAM VR 2023-03-04 10:04:20 - Il-2 Frames: 4474 - Time: 60000ms - Avg: 74.567 - Min: 61 - Max: 91 2023-03-04 10:06:16 - Il-2 Frames: 4498 - Time: 60000ms - Avg: 74.967 - Min: 60 - Max: 91 2023-03-04 10:08:15 - Il-2 Frames: 4461 - Time: 60000ms - Avg: 74.350 - Min: 62 - Max: 91 VR TEST2 OPENCOMPOSITE WITHOUT CAS 2023-03-04 11:20:08 - Il-2 Frames: 5280 - Time: 60000ms - Avg: 88.000 - Min: 80 - Max: 91 2023-03-04 11:22:02 - Il-2 Frames: 5242 - Time: 60000ms - Avg: 87.367 - Min: 81 - Max: 91 2023-03-04 11:23:51 - Il-2 Frames: 5303 - Time: 60000ms - Avg: 88.383 - Min: 82 - Max: 92 VR TEST2 OPENCOMPOSITE WITH CAS 2023-03-04 11:34:33 - Il-2 Frames: 5190 - Time: 60000ms - Avg: 86.500 - Min: 79 - Max: 92 2023-03-04 11:36:24 - Il-2 Frames: 5101 - Time: 60000ms - Avg: 85.017 - Min: 79 - Max: 91 2023-03-04 11:41:39 - Il-2 Frames: 5202 - Time: 60000ms - Avg: 86.700 - Min: 80 - Max: 91 2023-03-04 11:43:40 - Il-2 Frames: 5162 - Time: 60000ms - Avg: 86.033 - Min: 80 - Max: 91 2023-03-04 11:45:58 - Il-2 Frames: 5101 - Time: 60000ms - Avg: 85.017 - Min: 78 - Max: 91 VR TEST 3 STEAM VR 2023-03-04 10:23:09 - Il-2 Frames: 2621 - Time: 60000ms - Avg: 43.683 - Min: 38 - Max: 46 2023-03-04 10:25:28 - Il-2 Frames: 2551 - Time: 60000ms - Avg: 42.517 - Min: 37 - Max: 46 2023-03-04 10:27:21 - Il-2 Frames: 2607 - Time: 60000ms - Avg: 43.450 - Min: 37 - Max: 46 VR TEST 3 OPENCOMPOSITE WITHOUT CAS 2023-03-04 11:08:16 - Il-2 Frames: 2834 - Time: 60000ms - Avg: 47.233 - Min: 38 - Max: 54 2023-03-04 11:10:16 - Il-2 Frames: 2833 - Time: 60000ms - Avg: 47.217 - Min: 38 - Max: 54 2023-03-04 11:12:21 - Il-2 Frames: 2811 - Time: 60000ms - Avg: 46.850 - Min: 38 - Max: 54 VR TEST 3 OPENCOMPOSITE WITH CAS 2023-03-04 10:56:26 - Il-2 Frames: 3537 - Time: 60000ms - Avg: 58.950 - Min: 48 - Max: 67 2023-03-04 10:58:20 - Il-2 Frames: 3527 - Time: 60000ms - Avg: 58.783 - Min: 48 - Max: 67 2023-03-04 11:00:30 - Il-2 Frames: 3501 - Time: 60000ms - Avg: 58.350 - Min: 48 - Max: 66 as one can see : 1. for VR1 and VR2 tests FPS are almost the same between CAS and NO CAS sharpening option in OPENXR TOOLKIT, 2. for VR3 test i see a significant FPS gain using CAS sharpening method comparing to NO CAS method or Steam Vr, 3. for VR1 test there is a small gain of 5-6 FPS using opencomposite ( with CAS or not) versus steamvr , 4. for VR2 test there is a larger gain of about 11 FPS using opencomposite ( with CAS or not) versus steamvr , 5. for VR3 test there is a significant gain of about 15 FPS using opencomposite with CAS versus steamvr and a small gain of about 4 FPS using opencomposite without CAS versus steamvr. Edited March 14, 2023 by dgiatr 1 2
chiliwili69 Posted March 14, 2023 Author Posted March 14, 2023 Thank you @MilitantPotato for raising the doubt and @dgiatr for making such exhaustive tests which clearly shows the influence of CAS in the vrtest3. I have updated the table and also instructions to make it clear to not use any upscaling or other features of OpenXR toolkit when running the benchmark
MilitantPotato Posted March 14, 2023 Posted March 14, 2023 (edited) Thanks for taking the time and hassle to re run those tests, glad to see it was such a simple to diagnose oddity! Sorry if i came across as rude at all, was not my intent. I think requiring a screen shot of steamvr and/or openxr resolution used for the passes, while annoying, would help reduce the noise in vr3. Steamvr also does it's stupid "advanced super sample antialiasing" by default, which reduces framerates a small amount, and openxr doesn't do. I'd highly recommend having people turn that off (kills spotting anyway.) Edited March 14, 2023 by MilitantPotato Words
firdimigdi Posted March 15, 2023 Posted March 15, 2023 (edited) On 3/13/2023 at 6:10 PM, MilitantPotato said: Cas is an upscaling technology, that alone boosts fps by about 30% (same as nis/fsr at .70) as far as I'm aware 11 hours ago, chiliwili69 said: Thank you @MilitantPotato for raising the doubt and @dgiatr for making such exhaustive tests which clearly shows the influence of CAS in the vrtest3. I have updated the table and also instructions to make it clear to not use any upscaling or other features of OpenXR toolkit when running the benchmark Probably worth mentioning that CAS only has optional upscaling, which is not used in OpenXR Toolkit's implementation. Relevant line of code in OpenXR Toolkit: https://github.com/mbucchia/OpenXR-Toolkit/blob/69234c6be9fd77318130664e47d00215c4ad1512/XR_APILAYER_MBUCCHIA_toolkit/cas.cpp#L112 And In the AMD provided header: https://github.com/GPUOpen-Effects/FidelityFX-CAS/blob/9fabcc9a2c45f958aff55ddfda337e74ef894b7f/ffx-cas/ffx_cas.h#L408 So, if anything, using this version should theoretically drop performance (by a tiny amount - likely undetectable) rather than enhance it - so for the performance discrepancy above I suspect a difference in rendered resolution which stems from a different reason. Perhaps looking in startup.cfg after a benchmark run to validate the or_width/height numbers written there is a good idea. Edited March 15, 2023 by firdimigdi
Alonzo Posted March 15, 2023 Posted March 15, 2023 9 hours ago, firdimigdi said: Probably worth mentioning that CAS only has optional upscaling, which is not used in OpenXR Toolkit's implementation. You're right, at least looking at the code you linked to I agree with your assessment. And the OpenXR Toolkit in-game overlay doesn't show a render scale if you select CAS, which lends further evidence that the developer didn't intend to use upscaling mode. But then something else is changing the resolution. I'm fairly sure the folks are swapping between modes just using the OpenXR Toolkit UI and not messing with anything else, so that would imply there's something wonky elsewhere in the toolkit code.
firdimigdi Posted March 15, 2023 Posted March 15, 2023 (edited) 2 hours ago, Alonzo said: I'm fairly sure the folks are swapping between modes just using the OpenXR Toolkit UI and not messing with anything else, so that would imply there's something wonky elsewhere in the toolkit code. In the last tests pasted above the only one that really stands out to me is the VR Test 3 - the rest of the results both with and without CAS are within error margin. If there was something to it the same difference would manifest in all tests I'd imagine; much like the difference between SteamVR and OpenComposite does in all cases. Unless CAS caps the render size, something I do not see being done in the relevant code provided by AMD. Not sure if that'd be something that would likely show up in the startup.cfg values but it would be worth checking. EDIT: Actually, if that was the case then it would also apply to VR Test 2 as according to instructions it's "same IL-2 settings that VRTest2, but adding Clouds to Extreme and MSAAx8". And I cannot imagine how it would affect either MSAA or an in-game setting like the cloud quality. Edited March 15, 2023 by firdimigdi
MilitantPotato Posted March 16, 2023 Posted March 16, 2023 6 hours ago, firdimigdi said: In the last tests pasted above the only one that really stands out to me is the VR Test 3 - the rest of the results both with and without CAS are within error margin. If there was something to it the same difference would manifest in all tests I'd imagine; much like the difference between SteamVR and OpenComposite does in all cases. Unless CAS caps the render size, something I do not see being done in the relevant code provided by AMD. Not sure if that'd be something that would likely show up in the startup.cfg values but it would be worth checking. EDIT: Actually, if that was the case then it would also apply to VR Test 2 as according to instructions it's "same IL-2 settings that VRTest2, but adding Clouds to Extreme and MSAAx8". And I cannot imagine how it would affect either MSAA or an in-game setting like the cloud quality. The only VR test that isn't mostly CPU bound in this benchmark is VR3, any improvements in GPU rendering performance would be hidden by the fact that the game runs at 55-60FPS most of the test, with super low GPU usage. MSAA 8 and extreme clouds has a 4090 at 100% usage and 10fps under bottleneck of a 5800x3d. I posted some usage and frametime plots a page or two back. Once video cards are able to push 50-60FPS, if CPUs haven't improved by quite a bit, the test will need modified again.
RossMarBow Posted March 16, 2023 Posted March 16, 2023 (edited) I'm taking a break from IL-2. But keeping an eye on things. What I would like to see on the benchmark is if the CPU is running stock or not. Maybe colour the mhz column on the table. Not saying overclocking is invalid. But its pretty obvious theirs some wild differences in what people define as stable and usable. And theirs not much you can do about that without requiring a range of stability tests be undertaken. I think it should be more clear if someone is running on the bleeding edge of stability with a 0.1% CPU/RAM setup that's challenging for overclocking world records. Considering most people don't even test their setup at stock properly, not highlighting results that are most likely unobtainable without a good amount of luck, or a lot of work, but probably both seems a little bit dishonest. A lot of dream oc results that would easily be global top 100s in timespy, on this spreadsheet. Borderline unbelievable the concentration of bleeding edge results here, compared with the results I see in actual overclocking communities. I don't believe in passing p95 either or any of the other multicore stress tests/benchmarks as they are just not relevant to gaming workloads on modern cpus, but a minimum test for me personally would be 24hrs of corecycler testing single core with the lightest workload, an hour of linpack, overnight memtest, and an hour of 7zip all core which is about 25% lighter than r20 and actually reflects what load a gaming cpu will take multicore as it decompresses assets. And I say all this as someone who dailies settings that most hobby overlockers wouldn't. Again any results valid and stability is a state of mind. I think this issue needs to be highlighted so that the average person isn't misled though. It's a bit like following a skier with a parachute on. At first it's a minor difference barely noticeable, then all of a sudden it's not. Also I still don't understand the methodology used here. If you ran 1080p or even 720p on minimum graphics settings, that would give use a far better measure of CPUs. Then ran 4k max settings to test the graphics cards. Then just run vr and something sensible. At the moment its too hard to figure out of IL-2 benefits from 3d cache. If the new amd cards perform better, than the 6000s did. If the 4070 ti is underperforming cause of its gimped ram bus, compared to green 3000s. The only conclusion I can draw currently is il-2 runs like csgo, wanting max frequency. Which is pretty unusual, but also believable. But this doesn't make a lot of sense when cache is king for the other simulators I looked at i.e. msfs, f1 2020 Edited March 16, 2023 by RossMarBow
firdimigdi Posted March 16, 2023 Posted March 16, 2023 6 hours ago, MilitantPotato said: The only VR test that isn't mostly CPU bound in this benchmark is VR3, any improvements in GPU rendering performance would be hidden by the fact that the game runs at 55-60FPS most of the test, with super low GPU usage. MSAA 8 and extreme clouds has a 4090 at 100% usage and 10fps under bottleneck of a 5800x3d. I posted some usage and frametime plots a page or two back. Once video cards are able to push 50-60FPS, if CPUs haven't improved by quite a bit, the test will need modified again. Indeed, but that doesn't explain why enabling CAS which by all indications in the code should be costing performance, as in this current instance is not upscaling the image, instead increases performance by such a substantial amount.
MilitantPotato Posted March 16, 2023 Posted March 16, 2023 9 hours ago, firdimigdi said: Indeed, but that doesn't explain why enabling CAS which by all indications in the code should be costing performance, as in this current instance is not upscaling the image, instead increases performance by such a substantial amount. I'll ask the dev for clarification. There's zero chance CAS just magics performance so I'm expecting it's just missing a GUI element or something.
firdimigdi Posted March 16, 2023 Posted March 16, 2023 44 minutes ago, MilitantPotato said: I'll ask the dev for clarification. There's zero chance CAS just magics performance so I'm expecting it's just missing a GUI element or something. For what it's worth, even though it's quite obvious in the code I linked above, I did a quick check and using CAS via OpenXR Toolkit does not upscale the image, rendered resolution remains the same (it is also outputted in startup.cfg, so if you use upscaling like FSR or NIS you'll see it recorded there; not to mention that upscaled image quality is quite apparent even without looking at numbers). Something else is at play.
MilitantPotato Posted March 17, 2023 Posted March 17, 2023 16 hours ago, firdimigdi said: Indeed, but that doesn't explain why enabling CAS which by all indications in the code should be costing performance, as in this current instance is not upscaling the image, instead increases performance by such a substantial amount. Hey, could you try loading oxtk into safemode and resetting to defaults with the hotkeys? I wonder if it's pulling from old fsr/nis settings or something?
RossMarBow Posted March 17, 2023 Posted March 17, 2023 Considering fraps last update was 2013. Not too sure why it is still being used here. Not surprised it has issues with VR. OCAT or like the dude said inbuilt tools seems to be a more sensible option.
MilitantPotato Posted March 17, 2023 Posted March 17, 2023 I'd be real interested in ocat vs oxrtk internal metrics runs, if ocat is similar doing a steamvr run with ocat would probably be worthwhile. If fraps is just reading openxr results poorly that'd be wild.
firdimigdi Posted March 17, 2023 Posted March 17, 2023 8 hours ago, MilitantPotato said: Hey, could you try loading oxtk into safemode and resetting to defaults with the hotkeys? I wonder if it's pulling from old fsr/nis settings or something? No need to do all that - just go to the registry and rename Computer\HKEY_CURRENT_USER\SOFTWARE\OpenXR_Toolkit\OpenComposite_Il-2 to something else, then rename it back when done testing. It does not read any fsr/nis settings (this also is validated by reading the relevant toolkit code) - as I said before if it did change the render resolution it would be a. visible in the or_width/height entries of startup.cfg and b. quite apparent in the headset itself. I insist there's something else at play here. 1
chiliwili69 Posted March 17, 2023 Author Posted March 17, 2023 On 3/16/2023 at 8:20 AM, RossMarBow said: CPU is running stock or not. How you will define stock? with Intel you have the TurboBoost, the Thermal Velocity Boost, Undervolting, and many other tricks. With AMD you have the PBO, PBO2, and beyond. I think all this Auto turbos do a job. Investing time ton know more and get a bit more of performance is less worth than before. In a i9-13900K the base frequency for P-cores is 3.0 GHz. Is this Stock?, or 5.4? or 5.9? It wold be really difficult to determine what is stock and what is not. I just had a column for CPU freq (from the days we did manual fixed overclock) and currently is used to indicate the Max freq settings of the Turbo or overclock. Adding more details if provided in the notes field. On 3/16/2023 at 8:20 AM, RossMarBow said: Borderline unbelievable the concentration of bleeding edge results here What results are unbelievable for you? Most of the tests has been confirmed later with more people with similar hardware. I think most of people here report their results using their normal gaming turbo/overclock settings.
chiliwili69 Posted March 18, 2023 Author Posted March 18, 2023 On 3/16/2023 at 8:20 AM, RossMarBow said: If you ran 1080p or even 720p on minimum graphics settings, that would give use a far better measure of CPUs. The asnwer to your question is that some graphics options load more the CPU (mirrors, shadow, Preset, etc) and other graphics settings load more the GPU (clouds, MSAA, resolution). You can have more details about it here. During the CPU test we want to load the CPU as much as possible and unload the GPU as much as possible. Most of the graphics settings except Clouds, MSAA and resolution load essentially the CPU. The CPU is taking care of the flying models, damage models, particles of smoke/fire, AI objects, to define exactly the polygons mesh of every object in the scene. Once this is defined then the GPU is taking care of render that scene with the right lights/colors for every single pixel, the more pixels the more work for the GPU. In VR, the polygon mesh has to be calculated for every eye, and that´s why the VR performance is very much dependant on the CPU performance. And that´s the reason for the VRtest1 where the resolution is only 9.5 million pixels.
chiliwili69 Posted March 18, 2023 Author Posted March 18, 2023 On 3/17/2023 at 3:59 AM, RossMarBow said: Considering fraps last update was 2013. Not too sure why it is still being used here. Not surprised it has issues with VR. OCAT or like the dude said inbuilt tools seems to be a more sensible option Yes, last update was fom 2013 but it has been working well so far for the simple purpose of this basic test. Being old, if works, should not be a reason to discard it. I am 53 and still fully functional! ? BTW, Who had issues with fraps with VR? OCAT was suggested in the past, look here. But I discarded it since it give me issues with my NVIDIA card. Perhaps it is just me, but remember that OCAT is an AMD tool. If you find any other simple and fully free software (no ads, no data collected) which wold better than fraps I will be happy to take a look.
MilitantPotato Posted March 19, 2023 Posted March 19, 2023 4 hours ago, chiliwili69 said: Yes, last update was fom 2013 but it has been working well so far for the simple purpose of this basic test. Being old, if works, should not be a reason to discard it. I am 53 and still fully functional! ? BTW, Who had issues with fraps with VR? OCAT was suggested in the past, look here. But I discarded it since it give me issues with my NVIDIA card. Perhaps it is just me, but remember that OCAT is an AMD tool. If you find any other simple and fully free software (no ads, no data collected) which wold better than fraps I will be happy to take a look. I believe nvidia has fixed the issue of monitoring tools causing regular and high frametime spikes. At least i don't get them anymore and people aren't complaining about it as much anymore. I wonder if that was what you experienced with ocat?
firdimigdi Posted March 19, 2023 Posted March 19, 2023 OCAT, FrameView and CapframeX are all based on PresentMon Both fraps and PresentMon for the basic enumeration of "frames rendered" query the same source. This is a metric that is the same either in VR or in flat screen, the VR runtime used plays no role in this (fraps or presentmon don't query OpenXR or OpenVR for this info). PresentMon on the other hand is able to provide further detailed metrics via ETW (Event Tracing for Windows) and it also supports DX12. All of the above PresentMon based apps can be setup in exactly the same way as fraps for the purposes of this benchmark. Would the extra metrics help? Maybe. Has the benchmark in its current form helped? Yes, definetely. 1 2
=SFG=BoostedStig Posted March 24, 2023 Posted March 24, 2023 (edited) Haven't had a chance to run the VR stuff yet, but will probably get around to it in a week or so. Motherboard: Asus ROG Strix X670E-F CPU: AMD R9 7950X3D CPU Freq: 5.7 Ghz (5 Ghz on X3D cores) Cores: 16 Threads: 32 RAM size: 32Gb (2x16GB) RAM Freq: 6000 MHz (DDR5) FCLK: 2033 MHz RAM timings: 30-38-38-68 GPU: Gigabyte Gaming OC 4090 Windows 10 1080P CPU Test: 2023-03-24 00:21:59 - Il-2 Frames: 9161 - Time: 60000ms - Avg: 152.683 - Min: 133 - Max: 198 Edited March 24, 2023 by =SFG=BoostedStig
chiliwili69 Posted March 24, 2023 Author Posted March 24, 2023 On 3/19/2023 at 1:52 AM, MilitantPotato said: I wonder if that was what you experienced with ocat? I had spikes with the CPU test, but the big issue I had is some erratic behaviour when using in VR, don´t remember exactly what it was but I uninstalled the tool and problem was corrected. I can evaluate it again, or any other tool suggested, but I have another bigger domestic "problem" right now. My 15-yr son spent too much time playing with the PC I use for IL-2 and he didn´t put enough time for his studies. So, I resolved to keep the PC in the closet until he recover better study habbits. It means no IL-2 for me for 1 month or 2. ? On 3/19/2023 at 9:31 AM, firdimigdi said: OCAT, FrameView and CapframeX Thanks for this. I will not be able to evaluate this tools as an alternative to fraps (not PC for a month or two...) What I miss from fraps in VR tests is to get some more useful data like average frametimes and percentiles frametimes. If you or someone else can evaluate these tools (or any other tool), and give feedback, then it would be great. Keep in mind that the tool and test procedure has to be simple. 2 hours ago, =SFG=BoostedStig said: Haven't had a chance to run the VR stuff yet, but will probably get around to it in a week or so Thank you for this test. Quite aligned with the other 7950X3D. Let see how it goes in VR.
firdimigdi Posted March 24, 2023 Posted March 24, 2023 1 hour ago, chiliwili69 said: Thanks for this. I will not be able to evaluate this tools as an alternative to fraps (not PC for a month or two...) What I miss from fraps in VR tests is to get some more useful data like average frametimes and percentiles frametimes. If you or someone else can evaluate these tools (or any other tool), and give feedback, then it would be great. Keep in mind that the tool and test procedure has to be simple. Can't promise on a timeframe on this as my computer time is mostly taken up by work, but I'll see what I can do. I doubt it would be any more complicated as a procedure to perform the tests, they operate in a very similar way AFAIK (set a 'start' key and a time limit), but the data-gathering/presentation might be.
=SRS=SLaXOR Posted March 25, 2023 Posted March 25, 2023 Hi guys. After some straggle in maintaining stable 90fps on my old rig I ended up upgrading CPU+Mobo. Below is old hardware and then new one. Old rig: Motherboard: Asus Z170-AR CPU: i7-7700K CPU Freq: 4.5 Ghz Cores: 4 Threads: 8 RAM size: 32Gb (4x8GB) RAM Freq: 3000 MHz (DDR4) RAM timings: 15-17-17-35 GPU: PNY RTX 4080 Windows 10 1080P CPU Test: 2023-02-25 15:36:04 - Il-2 Frames: 5039 - Time: 60000ms - Avg: 83.983 - Min: 66 - Max: 111 The same rig but CPU OC up to 4800 Ghz (I could get up to 5100 but no gains in IL-2 performance after 4800) 2023-03-02 23:58:58 - Il-2 Frames: 5447 - Time: 60000ms - Avg: 90.783 - Min: 74 - Max: 125 New rig: Motherboard: Asrock Steel legend H670 CPU: i5-13600K CPU Freq: 5.1 Ghz Number of cores: 14 Number of threads: 20 Core Set 0 P-Cores, 6 cores, 12 threads Core Set 1 E-Cores, 8 cores, 8 threads RAM size: 32Gb (4x8GB) RAM Freq: 3000 MHz (DDR4) RAM timings: 15-17-17-35 GPU: PNY RTX 4080 Windows 10 2023-03-25 21:08:06 - Il-2 Frames: 8124 - Time: 60000ms - Avg: 135.400 - Min: 116 - Max: 181 2 1
chiliwili69 Posted March 25, 2023 Author Posted March 25, 2023 8 hours ago, slaxor said: CPU: i5-13600K Congrants for your new upgrade! You just upgraded from 7700K to a 13600KF, keeping same RAM and GPU and you got +45 fps in the CPU test! That´s a nice improvement. It is the first 13600KF tested here and the results are very close to their bigger 13th gen brothers! If one day you want to get some performance you may consider upgrade the RAM, for 200$ you can get DDR4 32GB with first word latency below 8 ns: https://pcpartpicker.com/products/memory/#b=ddr4&Z=32768002,32768004&S=3000,8000&F=6000000,8085106&sort=price&page=1 But perhaps you don´t need it right now. It would be interesting if you could run the VR tests...
=SRS=SLaXOR Posted March 27, 2023 Posted March 27, 2023 On 3/26/2023 at 6:21 AM, chiliwili69 said: Congrants for your new upgrade! You just upgraded from 7700K to a 13600KF, keeping same RAM and GPU and you got +45 fps in the CPU test! That´s a nice improvement. It is the first 13600KF tested here and the results are very close to their bigger 13th gen brothers! If one day you want to get some performance you may consider upgrade the RAM, for 200$ you can get DDR4 32GB with first word latency below 8 ns: https://pcpartpicker.com/products/memory/#b=ddr4&Z=32768002,32768004&S=3000,8000&F=6000000,8085106&sort=price&page=1 But perhaps you don´t need it right now. It would be interesting if you could run the VR tests... Thanks a lot! I really enjoy 120 FPS on my Oculus 2 with the new hardware - the smoothness is stunning Yes, I only upgraded Mobo and CPU, RAM is the same. Thanks for the memory advice but I think I'm good for now. I will try VR test when I have some spare time. ?
FoxbatRU Posted March 27, 2023 Posted March 27, 2023 (edited) Sold 12900K, bought 13700KF. Earned +30$ and slightly raised the minimum FPS. I was counting on a slightly larger increase, but given the free replacement, I'm still satisfied. I think this should be enough up to 14700K or 8800X3D. CPU test Frames: 8307 - Time: 60000ms - Avg: 138.450 - Min: 123 - Max: 190 Motherboard: Asus ROG STRIX Z690-A GAMING WIFI D4 CPU: 1300KF CPU Freq: 5.9 GHz L3 cache: 30 Mb Cores: 8 Threads: 8 (w/o HT & E) RAM type: DDR4 RAM size: 16 GB (2x8) NB Freq: 4500 MHz RAM Freq: 3600 MHz RAM Latency: 14-15-15-31 GPU: RTX 3080 OS: Windows 10 (x64, 21H2) IL2: 5.004b CPU Cooling: NZXT Kraken X73 Edited March 27, 2023 by FoxbatRU
DBCOOPER011 Posted March 28, 2023 Posted March 28, 2023 5 hours ago, FoxbatRU said: Sold 12900K, bought 13700KF. Earned +30$ and slightly raised the minimum FPS. I was counting on a slightly larger increase, but given the free replacement, I'm still satisfied. I think this should be enough up to 14700K or 8800X3D. CPU test Frames: 8307 - Time: 60000ms - Avg: 138.450 - Min: 123 - Max: 190 Motherboard: Asus ROG STRIX Z690-A GAMING WIFI D4 CPU: 1300KF CPU Freq: 5.9 GHz L3 cache: 30 Mb Cores: 8 Threads: 8 (w/o HT & E) RAM type: DDR4 RAM size: 16 GB (2x8) NB Freq: 4500 MHz RAM Freq: 3600 MHz RAM Latency: 14-15-15-31 GPU: RTX 3080 OS: Windows 10 (x64, 21H2) IL2: 5.004b CPU Cooling: NZXT Kraken X73 FoxbatRU, I have the 13700KF also and believe you possibly should be getting better FPS with that CPU. I have my 13700KF right now at cores 58x2, 57x2, 56x2, 55x2, and 54x2 with +1 TVB. With TVB, it boosts the 2 highest cores to 5900Mhz and allcore to 5500Mhz. I just did a quick CPU test and got an Average of 152 FPS, which you probably should be getting. When I set my 2 highest cores at 6000Mhz, I was averaging about 155-157 FPS on the CPU test. This is with hyper threading and the E-cores on, all stable with temps lower then stock. The link below is a pretty good guide on how to set up TVB if interested... https://www.overclock.net/threads/asus-maximus-z790-extreme-and-intel-i9-13900k-a-tuning-guide-for-beginners.1801569/ 1
dgiatr Posted March 28, 2023 Posted March 28, 2023 (edited) On 3/28/2023 at 3:17 AM, DBCOOPER011 said: FoxbatRU, I have the 13700KF also and believe you possibly should be getting better FPS with that CPU. I have my 13700KF right now at cores 58x2, 57x2, 56x2, 55x2, and 54x2 with +1 TVB. With TVB, it boosts the 2 highest cores to 5900Mhz and allcore to 5500Mhz. I just did a quick CPU test and got an Average of 152 FPS, which you probably should be getting. When I set my 2 highest cores at 6000Mhz, I was averaging about 155-157 FPS on the CPU test. This is with hyper threading and the E-cores on, all stable with temps lower then stock. The link below is a pretty good guide on how to set up TVB if interested... https://www.overclock.net/threads/asus-maximus-z790-extreme-and-intel-i9-13900k-a-tuning-guide-for-beginners.1801569/ I agree very informative tuning guide there, i OC my 13900 kf too running about the same core speed combinations at relatively low temps for that cpu. But you have to check if your mobo reads die sense voltage or socket sense. This guide talks about die sense voltages and my mobo reads socket sense voltages so core voltages are not the same as those found in the guide but a little bit higher. Edited March 29, 2023 by dgiatr 1
Recommended Posts
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 accountSign in
Already have an account? Sign in here.
Sign In Now