DCS VR performances with the new AMD's 5600x CPUs

[Cassiop]

Hello guys, i wrote on reddit a post about tests i made during the past 3 days about DCS VR and how it rely on Hardware. Especially CPU and Ram which are often neglected for GPUs and are, in often, the actual problem.

 

If you guys have any questions please ask them on reddit as i might forgot to come back here to check if there are comments and answer them. Hope it might help.

Moderators, Feel free to move this post to any section of the forum it i'm in the wrong one !

 

the post: https://www.reddit.com/r/hoggit/comm...re_and_dcs_vr/

 

the content

 

Hi Everyone, I noticed a reccurent problem into the hoggit's community. Either people dont know shit about computers and just go for most expensive parts, (which kinda works...) Either peoples dont know shit about computers and dont buy PC parts. Because they're broke. Or have better things to do than putting 1600$ for a GPU and 1000$ for a CPU that aren't even available anyway. =D So i decided to take a bit of my precious time (few fucking days, actually) to demistify things for you.

 

Important notions to keep in mind concerning DCS's behavior with hardware. And if you dont agree with that you should really have VERY SOLID arguments AND proof.

• GPU just need to keep up with CPU
  
• The goal is 45 fps in Asynchronous reprojection. (we're poor, remember ? also realistics.)
  
• DCS's core engine is cursed.
  

 

 

 

HARDWARE

 

now the setup for you to have an idea of what to expect. Its the "cheapest" setup i could do with amd's lastest release. (i like small motherboards and got a very good deal on this one. i've nothing to compensate ok? same for the ssd =D)

1. Gigabyte B550i Aorus Pro AX
   
2. AMD 5600x
   
3. 2*16Gb 3733C16 Dual Rank Micron E-Die
   
4. Sabrent 500gb PCIe gen 4 SSD
   
5. GTX 2070 super 8Gb
   

what you must remember from that :

1. any b550 motherboard will do. X570's features are irrelevant.
   
2. this is the cheapest new Zen 3 processor
   
3. This is actually a 3000c15 kit running at 3733c16. we will discuss that later
   
4. any SSD will do, just avoid HDD
   
5. the GPU is what it is. it's borrowed (i own a gtx1080) and i dont have any better so...
   

I also borrowed and tested the whole computer coming with the 2070s which is

• Asus TUF z390-Pro Gaming
  
• I5 9600k
  
• 4*8Gb 3433c16 Single Rank Samsung B-die
  
• Samsung 970 Evo Plus 500gb (Pcie gen 3 SSD)
  
• same GTX 2070 super
  

note : even if i didnt cover overclocking, on Intel's plateform is not doable without a Z390-490 chipset and a K cpu.

 

 

 

 

BENCHMARK

 

I setuped a F18 scene around Kutaisi. a pass at low altitude without TGP, a loop looking in the sky then the ground without TGP and a pass at low with TGP. Those 3 scenario cover things good enough for the purpose of the demonstration. you can have a look at the scene in the link below.

 

Does i benchmarked Multiplayer ? Well, yes but not much. the reason is that Kegetys's Shader doesn't IC Pass so most servers are inaccessible. i disabled it and did few tests on the Grownling Sidewinder server though but since my GPU couldn't keep up it was not very relevant.

 

Informations are given to you in Millisecond from FPS VR. That's quite inacurate but the best i can find and good enough since we wont notice a difference below 1ms (or even more than that.)

 

Settings will be detailed in a picture. I'm running last OB patch with Shader VR Mode from Kegetys (man if you're reading this, big shootout to you \o/) DCS's oversampling is set at 1.3 and SteamVR's at 1.5 for a total of 1.95. why ? because that's my sweet spot for readability, and GPU's performance.

 

The marging of error is quite big and nothing belows 1ms of Delta will really matter

 

 

 

 

INTERPRETING THE NUMBERS

 

(edit)I might have skipt this entire section... sorry for that =D.

 

I know those numbers are not the FPS we're used to. Long story short : The Lower The Better. Those numbers are the time needed for either the CPU or the GPU to render a frame. To allow a certain amount of FPS you need those number to be below the time that sit between two frames. For example at 90 fps you need to render a frame each 1000/90 = 11.1ms. which give the following table for the 2 commons refresh rate found on HMD's

 

90 fps for HTC Vive/Oculus Rift kind of HMD's

• 11.1ms for 90fps
  
• 22.2ms for 45fps with asynchronous reprojection at 100%
  

120 fps for Valve Index and such (which support 80 and 90 fps by the way)

• 8.3ms for 120fps
  
• 16.67ms for 60 fps with asynchronous reprojection at 100%
  

To have a rough idea where performances would sit with any HMD on the market you can extrapolate HMD's resolution x Oversampling. As stated, i'm using here 150% on Steam VR and 1.3 on DCS's Pixel density. So 1.5x1.3=1.95. I'm using a HTC Vive which use a 2160x1200 panel so i need to push at least 4212x2340 pixels every 22.2 milliseconds into the HMD in order to achieve those 45fps with 100% reprojection.

 

to keep the Valve Index for example you should achieve similar result running Steam VR at ~112% and DCS's Pixel Density at 1.3

 

 

 

 

Chapter One : The CPU

 

First question : Does the CPU Clock matter : MEH

• 5600x runing at 4.65ghz max 12/10/12ms
  
• 5600x running at 4.85ghz max 12/10/12ms
  

Of course CPU clock matter but that's not relevant for a ~5% increase and i didn't worked on CPU clock yet. Tools aren't available (yet ?) to figure what core can go how high and i wansn't willing to do that search by hand.

 

Second question : Does the CPU Matter at all ? Well...

• MP 5600x stock (4.65ghz) : 25.6/18/21
  
• MP 9600k stock (4.2ghz) : 27/18/21
  
• SP 5600x stock (4.65ghz) : 12/10/12
  
• SP 9600k stock (4.2ghz) : 14/11.4/15
  

At this point i have to clarify something. AMD Zen 3 cpus are known for having the best single core performances on the market currently. Cinebench R20 Single core performance of the 5600x is 604 while 9600k's is 481. that's ~20% More raw performances. While it translate rather well in SP, in MP it doesn't. I suspect the GPU bottlenecking the CPU here since GPUs stats are as follow

• MP 5600x GPU : 22/22/22
  
• MP 9600k GPU : 22/22/22
  
• SP 5600x GPU : 15/15/15
  
• SP 9600k GPU : 15/15/15
  

Also for the rest of the tests, Globaly GPU always sit between 13 and 17 as long as the CPU can keeps up.

 

Does DCS care about SMT ? (AMD's equivalent of hyperthreading). NO

• MP SMT Off : CPU:20-25ms, GPU 20-25ms
  
• MP SMT On : CPU:20-25ms, GPU 20-25ms
  

yeah, i know those numbers are probably GPU limited but in fact, almost all my tests are done SMT off since i forgot to turn it back on And its another variable i dont have to worry about anymore... its a win win !

 

 

 

 

Chapter 2 : the Ram !

 

Here is what i spent days doing. fine tunning those fuckers just to discover that Zen 3's bios are fucked right now and cant go beyond 1900 of Fclck. Mine actually is stuck at 1866.)

 

all thoses tests were made in SP with Shader mod On except those where i explicitly tell that they are Multiplayer.

 

First : Does the speed of the ram have an impact ? YES. well, a bit

• 2133C16 UnCoupled : 20/13/20
  
• 3200C16 UnCoupled : 14/12/15
  
• 3733C16 Uncoupled : 13/12/15
  

as long as you go past 3000 or 3200 you should be fine though

 

Do i need to run ram in Coupled mode (fclock = ram speed/2) or not : YES, BY A LOT

• 3733C16 UnCoupled : 13/12/15
  
• 3733C16 Coupled : 11.5/10/13.5
  

Does DCS Care about Ram's timing and subtimings. running at 3733 Coupled ? No.

• C16-20-16-16-38 : 11.5/10/13.5
  
• C16-20-20-20-40 : 11/10/14
  
• C18-20-20-20-40 : 11/10/14
  
• C20-20-20-20-40 : 12/10/15
  

Well, ok. yeah it does a liiiiittle bit. but if you're running your ram at C20 you are doing it wrong.

 

Wanna go further ? Lets go further !

 

Does DCS Care about running very high frequency of ram ? No.

• 2*8gb at 4400c18-20-18-18-38 : 11/10/14
  

had to use the samsung B-die kit for that. the ram was unstable but i managed to get results anyway.

 

Can DCS run on 8gb of Ram in Single Player? eeeehhhh...

• 1*8gb 12.5/12/15. Ram usage : 7.3gb
  
• 2*8gb 11/10/14. Ram usage : 10.9gb
  

Can DCS run on 16gb of ram in Multi Player ? eeeeehhhh...²

• 2*8gb CPU 20-25ms, GPU 19-20ms, Ram usage 14.8gb
  
• 2*16gb CPU 18-23ms, GPU 19-20ms, Ram usage 18gb
  

that's why you often hear to get 32gb of ram at least. Choosing a Kit of ram is trully a whole different topic and i wont go into that. You can go for Crucial Ballistix 3000-3200 2*16gb kit (that's all the same), they're perfect for the job and rather cheap. Also keep in mind that my windows is fresh and running only DCS, monitoring tools and OBS for recording.

 

Conclusions

 

Thanks for reading all that. Long Story short, This new generation of processor wont revolutionize DCS's performances in VR but it is now trully doable without an expensive intel plateform. Especially with mid range GPU coming along (~3060 ~6600-6700xt), i believe we will be able to run DCS VR in Okay condition within the next year. I didn't do all those bench with my GTX 1080 but, even if slighly slower than a 2070s, it runs the game totally fine with the same settings.

 

i hope anyway it helps people like me, that doesn't want to spend thousand in hardware but still want to get decent performances and readability in VR. I trully cant play on flat screen anymore and even if it's not perfect in all case scenario, i'm now able to play my favorite game in VR. Coming from a 3770k, it's an overall big leap in performances for me anyway !

 

If you have further question, please ask, i'll do my best to answer ! just doesn't ask me to tweak graphics settings to be able to play well in MP, i'm already at the bare minimum, every aspects wise =D

 

To ED : if you read this guys, thank's for all the work you've done on optimisation. Its trully something we need and we'll always be thanksful to you to do this ungrateful work ! I know you know about Kegetys's work on Shader VR Mod and i hope one day you could impletement something like that or create an exception rule to IC pass for servers to allow that kind of mod and reduce the GPU's load.

 

Thanks guys and fly safe.

 

DCS Settings : https://imgur.com/gallery/uGAkzBI

Kegetys's mod : http://www.kegetys.fi/

 

Multi player footage (Shader mod Off) :

Single Player footage (Shader mod On) :

 

(edit : the "interpreting the numbers" paragraph)

[Northstar98]

Awesome stuff, appreciated! :thumbup:

[DarkFire]

Very interesting, thanks for doing this testing. I'm currently running what is by now a fairly old i7 6700K and will be looking to swap to a 5950X in the next couple of months so this test is very useful and relevant.

[divinee]

Nice test, thanks!

 

Are you sure about that resolution? I've understood that SteamVr SS % and DCS PD work differently. First you must calculate resolution after DCS PD -> (2160*1.3)*(1200*1.3)=4 380 480 pixels (or 2808x1560). Then you use SteamVR SS% to multiply that pixel amount -> 1.5*4 380 480=6 570 720 pixels (3439*1911). Someone wiser correct me if my calculations aren't correct..

 

btw, are you able to test i5 CPU with about 5ghz OC? Would be interesting to see how it compares..

 

edit: I learned that Vive's native resolution is propably not at the 100%? So we have to calculate the amount of increase in supersampling and my calculation is not correct. The result is propably bigger depending on the native SS% number..

Edited November 17, 2020 by divinee
correction

[Cassiop]

Nice test, thanks!

 

Are you sure about that resolution? I've understood that SteamVr SS % and DCS PD work differently. First you must calculate resolution after DCS PD -> (2160*1.3)*(1200*1.3)=4 380 480 pixels (or 2808x1560). Then you use SteamVR SS% to multiply that pixel amount -> 1.5*4 380 480=6 570 720 pixels (3439*1911). Someone wiser correct me if my calculations aren't correct..

 

btw, are you able to test i5 CPU with about 5ghz OC? Would be interesting to see how it compares..

 

edit: I learned that Vive's native resolution is propably not at the 100%? So we have to calculate the amount of increase in supersampling and my calculation is not correct. The result is propably bigger depending on the native SS% number..

 

yeah someone pointed this resolution maths to me, appearently i didn't understood well how PD and SteamVR SS% works. but i'm too lazy to check =D and the overall result would be not that relevant anyway since maybe Valve Index doesn't need that much (or more?) oversampling to be readable because of its native increased resolution.

 

No i didn't tried to overclock the i5 because that's not my computer and i dont want to breaks it =D that said, performances scale well with frequency. I5 9600k have a turbo mode that allow it to go to run at 4.6ghz (and i believe it does in DCS since its monocore and shouldn't trigger thermal throttling. it has a corsair H80 on it. But i didn't checked). going up to 5ghz would result in an increase of performances of about 8%.

[Cassiop]

Very interesting, thanks for doing this testing. I'm currently running what is by now a fairly old i7 6700K and will be looking to swap to a 5950X in the next couple of months so this test is very useful and relevant.

 

a 5950x will indeed make quite a difference =D even in DCS ! that said, you'll still struggle to get 45 fps in heavily loaded MP servers. Playing in not that crowded MP servers will should be fine though !

[Qiou87]

That's quite a detailed and thorough analysis, thanks for that. I have been playing a bit with Oculus debug tools to map my frametimes in repeatable scenarios and try to find out what I can improve for VR, while I was looking for the best compromise in supersampling between readability and smoothness.

 

I found that in most cases, the frametime I get is around 10-15ms from the CPU (in SP) and 18-25ms from the GPU. Based on this, I thought that I should update my GPU first (RTX 2070 SUPER) and only worry about the CPU later (Ryzen 5 2600X). I currently get a very decent experience, using SS x1.4 in the Rift S, my game is mostly smooth (except some airports in Syria and Supercarrier) with decent details, but I want to push SS a bit more (my cockpit is readable, but only just).

However your post seem to indicate the opposite (CPU is more important than GPU), which surprises me. My understanding was that the GPU takes ~20ms to calculate the frames as it is, so as long as the CPU is done before that, it is not bottlenecking. I would like to understand more about this reasoning, mostly because I am always happy to learn about PC hardware, but also to spend my money in the most efficient way. :smartass:

[Lurker]

Looks like you are GPU limited in nearly every test there. Makes sense since it pushes so many pixels at that resolution. I think these processors will only really come into play when paired with the new GPUs. Still a decent test, though you might want to change how you present your results, it's not really clear from the way you structured your results: XX/XX/XX doesn't really tell me much, when just looking at the numbers.

[Raven (Elysian Angel)]

DarkFire said:

I'm currently running what is by now a fairly old i7 6700K

So am I. I'm looking to upgrade early next year to 5900X and 6800XT or 6900XT, and am very interested in how the new SAM architecture will perform in DCS

Edited December 6, 2020 by Elysian Angel
typo