My 3090 Settings for my G2 (2.7.11)

[ShaunX]

9 hours ago, speed-of-heat said:

OP Updated for 2.7.

 

Fantastic update/rewrite of what is surely THE definitive tuning guide for DCS VR. I particularly appreciate the addition of methodogy and baselining sections.

 

Thank you for taking the time to put this together.

 

Now, I just have to try and chase your frametimes...Damn !

[speed-of-heat]

Thanks @ShaunX,i’m glad people find it useful, as a starting point for their own settings. 
 

I will also note that I did at one stage have FPS figures, I took them out, because I couldn’t get them to quite marry up to the frame times, there is some additional interaction going on there that I don’t quite understand yet. And when I am less tested out i will go back and look at the test data and see if I can divine it.  I don’t know if it’s an artefact of fpsvr or the interaction of cpu and gpu frametimes and “anything” else... but the deltas are more useful than the absolute frametimes in my opinion.

[ShaunX]

1 minute ago, speed-of-heat said:

Thanks @ShaunX,i’m glad people find it useful, as a starting point for their own settings. 
 

I will also note that I did at one stage have FPS figures, I took them out, because I couldn’t get them to quite marry up to the frame times, there is some additional interaction going on there that I don’t quite understand yet. And when I am less tested out i will go back and look at the test data and see if I can divine it.  I don’t know if it’s an artefact of fpsvr or the interaction of cpu and gpu frametimes and “anything” else... but the deltas are more useful than the absolute frametimes in my opinion.

 

I guess with reprojection enabled, FPS is irrelivent anyway so long as it's locked.

 

Just noticed..you didn't mention the new VR Mask (which incidently I'm still not seeing at all)

 

[speed-of-heat]

Still don’t know why not, I really wish I did.
 

I don’t really have a practical way of turning it off/on, in order to test anything with it, i did find a way of reducing it, which does have a performance impact, so my instinct is it is doing it’s job.

Edited April 24, 2021 by speed-of-heat

[freehand]

10 hours ago, speed-of-heat said:

Hi @skypirate,

great question, you only need one paging file, it’s a system wide resource.  Ideally it is a on a separate disk from DCS, and ideally using a separate bus resource on the motherboard to maximise performance. In some ways it’s a bit like ram if you are using multiple ssd’s make sure they are on separate scsi channels, m.2 drives should be already on a separate motherboard resource, to ssds. You might have to do so manual digging to work out what resources are shared.

 

but to answer your direct question, no you don’t need one on the dcs disk, indeed it’s preferable for performance that you don’t. So your current config is correct.

Last time i used scsi channels was over 20 years ago lol

 

I see (Serial-Attached SCSI solid-state drive) is a thing I stand corrected I used the older scsi devices on my old PCs & Amigas lol. 

Edited April 24, 2021 by freehand

[ShaunX]

@speed-of-heat can I ask you a couple of questions on your methodology and baselining please ?

 

I'm planning to redo my own based on that (although I'm currently held up by a) can't disable preallocated monitors and b) still not getting the VR Mask!)

 

1. For the baseline tests, did you leave motion reprojection off ?

2. Assuming you're using the data logging csv files that FPSVR can create ? Are you triggering that on and off with hot key within your track run ? What then, do you take an average of the results rows ?

 

Thanks

[speed-of-heat]

Hey @ShaunX, yes exactly that.

 

turn off any frame rate stabilisation, as you loose any performance improvements,  I think I have set my key in FPSvr to rctrl+rshft+F11 which didn’t seem to be in use.  then once the replay has run, I average the results of each column.

[ShaunX]

8 minutes ago, speed-of-heat said:

Hey @ShaunX, yes exactly that.

 

turn off any frame rate stabilisation, as you loose any performance improvements,  I think I have set my key in FPSvr to rctrl+rshft+F11 which didn’t seem to be in use.  then once the replay has run, I average the results of each column.

 

Luvely jubly thanks

 

...ah, forgot to ask..what about headset placement ? Ideally I guess you want to have the headset stationary in the same position for each run ? Do you place it in some sort of stand ? or am I being too picky ?

Edited April 26, 2021 by ShaunX

[speed-of-heat]

yep, I will update on this cause they are great questions! thats exactly what I do the headset is on a stand. So if i want to you know play the game, rather than just test it... i end up having to do another "baseline" run to establish the correct baseline for that "view", but as I am mostly looking at deltas in frame time from a known base that should be enough!   

 

The downside, of course is the "looking out the side problem" e.g. you dont, i also keep radar mode in air to air, again  minimise the number of variables that change.

Edited April 26, 2021 by speed-of-heat

[Mathieas]

Hi there Speed of Heat,

 

First, I would like to thank you for your amazing and thorough job testing settings and their impact on performance! I had a couple questions regarding your 2.7 update and the mods. I have adopted your mods as you suggested but was wondering if your final shadow setting in DCS is still set to low? You mention "medium" in one of your description for the mods but I did not see a change to a medium setting in any of your test descriptions. Are you using medium or low in DCS now with the 3 mods(VR Shaders Mod/Flat Shadow removal/Shadows Reduced Frame Rate Impact)? 

 

I was also wondering if you might be willing to share your test flight file? I would like to see how close my performance mimics yours. I have a 5800x with a 3090 FTW3 Hybrid. 

 

I was also hoping to understand a bit about how you test. I am very new to VR and so do not know the ins and out of what folks are doing. Do you leave the HMD off your head but running and just use the desktop? I personally have been putting FPSVR on my left monitor, alt-tab my right monitor that is now displaying a distorted left eye from my HMD and leave my HMD on the stand. 

 

Thank you sir!

Edited April 26, 2021 by Mathieas

[speed-of-heat]

@Mathieas, Thanks for your questions!

 

1) shadows yes it is still set on LOW in the UI, but the effect of the MOD is to give "Medium shadows in the cockpit".

 

2) sure performance track.trk

 

3) I leave the HMD on a stand and run the tests from my 2d monitor, just to minimise the total variations of head movement and so forth. 

 

4) I have two  monitors, but I only run DCS on one of them at a time, because my set up   is strange I have a single PC, on my left my main monitor in front of my main work space on my desk , and the to my far left I have monitor facing "the cockpit" area, which has my HOTAS setup and so on, on the far left end of the desk, but I normally disable that monitor when I am using the main screen and visa versa (not for any performance reason, though I might test it just for fun at some point, mostly just to pay a token nod at my electric bill)

 

Edited April 27, 2021 by speed-of-heat

[Mathieas]

@speed-of-heat Thank you very much for the detailed reply and your test file. Finally a benchmark to use to see how things are changing but with the added benefit to have a comparison with others.  

 

I have ran this using your final recommended test 26 settings but with Clear Water IC PASS. 

 

Alt-Enter used

GPU Frame Time: 9.22ms

CPU Frame Time: 6.11ms

 

No Alt-Enter used

GPU Frame Time: 9.64ms

CPU Frame Time: 6.32ms

 

This is my PC spec just incase anyone is interested:

CPU: AMD Ryzen 5 5800x - Fclk@1800mhz

Mother Board: Asus Crosshair VIII Formula
GPU: 1x 3090 FTW3 Ultra Hybrid @undervolt 0.987v +165mhz offset (~2070mhz - 2040mhz)

RAM: 2x 16GB G.Skill Trident Z Neo @3600mhz CL14 (B-Die) 

SSDs(NVME): 1X Samsung 970 EVO Plus 2TB, DCS and PAGE file here

Sound: Creative Sound Blaster ZXR
PSU: EVGA SuperNOVA 1000W T2

Monitor(Main): Acer Predator XB271HU (144hz IPS)

OS: Windows 10 pro 64bit

Edited April 26, 2021 by Mathieas

[speed-of-heat]

Hi @Mathieas,

I'm really glad you found it useful.  My intent is not to create a benchmark, though I am happy if you choose to do so, but that is not my intent or goal; Really it's to help people understand what's going on when they flip certain switches in DCS in a more systematic way. 

 

 

 

[Mathieas]

@speed-of-heat Totally understood! From my perspective, I would say you have met your goal 100%.  Thank you again! It was not until I saw your performance increase from the mods that I even considered trying them (generally try to keep my games stock). 

[speed-of-heat]

Awesome@Mathieas,I'm glad you have found it helpful.

 

 

 

[ShaunX]

13 hours ago, Mathieas said:

@speed-of-heat Thank you very much for the detailed reply and your test file. Finally a benchmark to use to see how things are changing but with the added benefit to have a comparison with others.  

 

I have ran this using your final recommended test 26 settings but with Clear Water IC PASS. 

 

Alt-Enter used

GPU Frame Time: 9.22ms

CPU Frame Time: 6.11ms

 

No Alt-Enter used

GPU Frame Time: 9.64ms

CPU Frame Time: 6.32ms

 

This is my PC spec just incase anyone is interested:

CPU: AMD Ryzen 5 5800x - Fclk@1800mhz

Mother Board: Asus Crosshair VIII Formula
GPU: 1x 3090 FTW3 Ultra Hybrid @undervolt 0.987v +165mhz offset (~2070mhz - 2040mhz)

RAM: 2x 16GB G.Skill Trident Z Neo @3600mhz CL14 (B-Die) 

SSDs(NVME): 1X Samsung 970 EVO Plus 2TB, DCS and PAGE file here

Sound: Creative Sound Blaster ZXR
PSU: EVGA SuperNOVA 1000W T2

Monitor(Main): Acer Predator XB271HU (144hz IPS)

OS: Windows 10 pro 64bit

 

 

Interesting to see figures from another AMD user.

 

I have a 5900X and 3090 and not quite getting your CPU result..

 

Using @speed-of-heat's track and same line 26 settings, kegetys and the flat shadows removal, I'm seeing averages

 

GPU Frame Time: 9.75ms

CPU Frame Time: 7.16ms

 

I do have an issue in that I'm not seeing the VR Mask at all on my screen mirror so I guess that could account for some of it

 

I also only have 3200mhz CL16 RAM. I have thought about upgrading but those 3600 CL14 kits ain't cheap (or easy to find these days). Were you able to compare performance with lesser RAM ?

 

Do you overclock the 5800X at all ? even PBO / auto OC ?

 

..incidently, @speed-of-heat, your FPSVR log file averaging method is really effective. I get almost identical results every time. So much more accurate than just watching the FPSVR real time figures bouncing around !

 

 

 

 

Edited April 27, 2021 by ShaunX

[speed-of-heat]

Hey @ShaunX really glad its working out for you!

 

do you also see a FT increase if you use sharper eye or is that something at my end ?

[ShaunX]

5 minutes ago, speed-of-heat said:

Hey @ShaunX really glad its working out for you!

 

do you also see a FT increase if you use sharper eye or is that something at my end ?

 

It's not just you, I definately see an increase, and I think you may also be right about the additional shimmering. The great thing is that it's an easy on/off so you can choose it for a particular scenario.

[speed-of-heat]

awesome great to know its not just me!

 

I really love the clarity! not sure I can live with the shimmer or the FPS hit ... , indeed the toggle may be the right approach to deal with it i will have a play after work tonight. 

[ShaunX]

..for good measure I just flipped resizable bar back on in the bios and Nvidia ProfileInspector..

 

Without

GPU Frame Time: 9.75ms

CPU Frame Time: 7.16ms

 

with

GPU Frame Time: 9.69ms

CPU Frame Time: 7.25ms

 

 

[Mathieas]

@ShaunX Sadly I have not tested performance with slower memory. I have read though that ram speed is really important for these CPUs. Running at 3200 would lower your Fclk to 1600 vs my 1800 which could be a big part of this esp with the 5900 having two chiplets(they communicate through the infinity fabric-fclk) while my chip only had one. Also, 2cl difference in speed at 3200 vs 3600 is much larger then it looks and is probably playing as big or bigger role then the fclk. For example, 14/3600 is 3.88ns while 16/3200 is 5ns. That is a 30% higher ram latency at the cl level. I have also read that the 5800x is so expensive for the core count because it require one perfect chiplets vs the 5600x or 5900x using 2 lesser chiplets(6cores/per vs 8cores/per). It is still less then a 5900x but only a $100 which would normally be strange for 2/3 the core count. 

 

I do use PBO on auto. I have it set to auto(I tried manually increase this, but like many others onlinr it was not helpful) and I tend to boost single core up to 4,850mhz on 1-2 cores at a time during cinebench. I did not check my boost clock during this test as those monitoring programs all cost performance.

 

Please let me know if you have any other questions, I will do my best to respond.

 

A note on ram: You may know this already but flcl really needs to match the ram freq/2. Though to be precise our ram isn't running at 3200 or 3600mhz. It is really running at 1600 and 1800mhz but completes two instructions per clock giving us 3200/3600million instructions per second instead of cycles (as the manufacturer would like us to think). Lastly, if you do upgrade ram, you want to make sure you are getting dual rank samsung b-die for the highest performance (2 stick 32gb/cl14 is almost if not always b-die, there are tools online to double check). Dual rank on 2 Dimms really seems to benefit AMD chips. You could get 4 single rank dimms but I find (due to physical limitations of the board and not knowing what memory topography your board uses) this generally results in looser  required timings then 2 dual rank chips.

Edited April 27, 2021 by Mathieas

[ShaunX]

5 minutes ago, Mathieas said:

@ShaunX Sadly I have not tested performance with slower memory. I have read though that ram speed is really important for these CPUs. Running at 3200 would lower your Fclk to 1600 vs my 1800 which could be a big part of this esp with the 5900 having two chiplets(they communicate through the infinity fabric-fclk) while my chip only had one. Also, 2cl difference in speed at 3200 vs 3600 is much larger then it looks and is probably playing as big or bigger role then the fclk. For example, 14/3600 is 3.88ns while 16/3200 is 5ns. That is a 30% higher ram latency at the cl level. I have also read that the 5800x is so expensive for the core count because it require one perfect chiplets vs the 5600x or 5900x using 2 lesser chiplets(6cores/per vs 8cores/per). It is still less then a 5900x but only a $100 which would normally be strange for 2/3 the core count. 

 

I do use PBO on auto. I have it set to auto(I tried manually increase this, but like many others onlinr it was not helpful) and I tend to boost single core up to 4,850mhz on 1-2 cores at a time during cinebench. I did not check my boost clock during this test as those monitoring programs all cost performance.

 

Please let me know if you have any other questions, I will do my best to respond.

 

A note on ram: You may know this already but flcl really needs to match the ram freq/2. Though to be precise our ram isn't running at 3200 or 3600mhz. It is really running at 1600 and 1800mhz but completes two instructions per clock giving us 3200/3600million instructions per second instead of cycles as the reported speeds would like us to think. Lastly, if you do upgrade ram, you want to make sure you are getting dual rank samsung b-die for the highest performance. Dual rank on 2 simms really seems to benefit AMD chips. You could get 4 single rank dimms but I find due to physical limitations of the board and not knowing what memory topography your board uses generally results in looser timings then 2 dual rank chips.

 

Thanks man, I'm new to AMD and although I've read a lot of that I haven't develved into in yet myself. I'll maybe give PBO auto OC a whirl on this test bed.

I've also read that PBO2 is what you really need to use to improve single core performance but it's currently only available via the latest bios and as I've really no overclocking experience I'll wait till they put it in Ryzen Master (next year?) before I lay my clumsy paws on it.

 

(..and I thought infinity fabric was just what the Avengers made their costumes out of   )

[Mathieas]

@ShaunX I know the feeling! I have been a 100% intel guy until this upgrade and all these new details were quite overwhelming. Overclocking was allot more simple with my 6900k. Do you have a sense for what your 5900x is boosting to on a single core? I use HWINFO64 to monitor while running cinebench R23. I imagine your 5900x is boosting as good or better then my chip stock. Here is the ram i bought just incase it help you any. It is quite expensive but so far I have been very happy with it. 

 

https://www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820374093?Item=N82E16820374093 

[ShaunX]

Thanks @Mathieas, it's actually those price numbers on the UK site...except it's in £££ !!

I've really done very little in the way of benchmarking as of yet. I'll let you know if I do.

 

I've just tried a basic PBO (it's literally just clicking one tab and apply in Ryzen Master for any Intel folk)

Immediately..

 

GPU Frame Time: 9.73ms

CPU Frame Time: 6.23ms

 

so looks like that's what I was missing.

 

I imagine once my lack of VR Mask issue gets resolved I'll gain a bit more.

 

 

[Mathieas]

@ShaunX Nice work!!! That is a very respectable cou frame time! Are you alt tabing as well once dcs starts?