Jump to content

Radeon slows down CPU in DCS, really? Testing Nvidia RTX 2080Ti vs RTX 3090 vs Radeon 6800 XT


Recommended Posts

I benchmarked Nvidia RTX 2080Ti, RTX 3090 and Radeon 6800 XT using the same settings in DCS.

 

While RTX 3090 is of course clearly the fastest of the bunch, I found a weird thing considering the CPU frametime values recorded using fpsVR.

It appears as if both Nvidia cards have similar CPU frametime and performance despite the large difference in GPU performance. This is expected.

 

The Radeon 6800 XT, however, shows much higher CPU frametime values. The difference is enough to be of practical importance.

Actually, sometimes it seems that the 6800 XT is resulting in lower fps rates than my old 2080Ti due to higher CPU load.

I very much prefer lower CPU load, because that is usually what is limiting my frame rate in VR (graphics load is tunable by various settings).

 

I'm not sure if something went wrong when reinstalling SteamVR etc due to the entirely different set of graphics drivers.

 

I attach an example of fpsVR graphs. In this specific case:
-all possible settings the same for Nvidia and Radeon

-mission is the F-18 instant action "ready on the ramp" in PG

-30s fpsVR recording sim stopped, then 60s more fpsVR recording with sim running (plane just idling on platform) --> 2 peaks in graphs corresponding to pause on and pause off

-steamVR set at 202% with Valve Index (2864x3184)

-CPU is i7 9700K @4.9GHz, Z390 chipset, 32GB RAM @3200MHz

-Nvidia CPU frametime is around 15.5-16.5ms paused, 17-18ms unpaused

-Radeon CPU frametime is around 18.5-19.5ms paused, 20-21ms unpaused, around 15% worse than Nvidia!

 

Can anyone else verify that Radeon drivers eat this much more CPU than Nvidia drivers in DCS and VR ?

Or did I mess up something when switching drivers?

 

R6800XT-B-2.png

RTX2080Ti-B-2.png

RTX3090-B-2.png

RTX3090-B-2.png

RTX2080Ti-B-2.png

R6800XT-B-2.png


Edited by RedX
tag graphs

Intel Core i7 9700K@4.6GHz|ASUS Strix Z390-H|G.Skill DDR4 32GB@3200MHz|MSI GeForce RTX 2080 Ti|2xSSD 500GB Samsung 970 EVO M.2 RAID0|SSD Samsung 850 EVO|Corsair HX 750W|Fractal Design Define R6 Blackout|UWQHD 34"|Valve Index|Logitech G9|MS Natural Ergonomic Keyboard 4000|VPC WarBRD Base+MongoosT-50CM2 Grip|VPC MongoosT-50 Throttle|VKB T-Rudder|Beyerdynamic DT770|Boom mic|Windows 10 Home

Link to post
Share on other sites
  • RedX changed the title to Radeon slows down CPU in DCS, really? Testing Nvidia RTX 2080Ti vs RTX 3090 vs Radeon 6800 XT

That is strange.

I know the Radeon cards are touted as running best with the AMD boards.

Don B

 

My VKB Gunfighter MK III Pro L Review

EVGA Z390 Dark MB | i9 9900k CPU @ 5.2 GHz | EVGA RTX 3090 FTW3 Ultra | 32 GB G Skill Trident Z 3200 MHz CL14 | Corsair H150i Pro Cooler | Samsung 970 Pro, 970 Evo 1TB m.2 Nvme |VKB GF MK III Pro L Ultimate Grip| Virpil CM3 Throttle |Reverb G2 | 

Link to post
Share on other sites

exil, did you clean up nvidia drivers with utility like DDU when you switched the card? Or did you just insert the new card and leave all the old drivers around? Did you reinstall steamVR? I wonder whether it matters which output port is used for VR headset, too.

 

The RTX 2080Ti is back in the computer and CPU frame times are the same as before as recorded by fpsVR.

I think I'll try to install 6800 XT again and see if the result changes by tinkering with the installation procedure or settings.

Intel Core i7 9700K@4.6GHz|ASUS Strix Z390-H|G.Skill DDR4 32GB@3200MHz|MSI GeForce RTX 2080 Ti|2xSSD 500GB Samsung 970 EVO M.2 RAID0|SSD Samsung 850 EVO|Corsair HX 750W|Fractal Design Define R6 Blackout|UWQHD 34"|Valve Index|Logitech G9|MS Natural Ergonomic Keyboard 4000|VPC WarBRD Base+MongoosT-50CM2 Grip|VPC MongoosT-50 Throttle|VKB T-Rudder|Beyerdynamic DT770|Boom mic|Windows 10 Home

Link to post
Share on other sites
21 hours ago, RedX said:

exil, did you clean up nvidia drivers with utility like DDU when you switched the card? Or did you just insert the new card and leave all the old drivers around? Did you reinstall steamVR? I wonder whether it matters which output port is used for VR headset, too.

 

The RTX 2080Ti is back in the computer and CPU frame times are the same as before as recorded by fpsVR.

I think I'll try to install 6800 XT again and see if the result changes by tinkering with the installation procedure or settings.

Please never ever use any weird tools for driver cleanup. You can have both drivers in the system without any problem. If you want to deinstall any driver clean, boot in safe mode and do it there with standard windows tools. Windows knows best. It is not Win98 anymore.


Edited by med0ra

| F-16C | FC3 | TM Hotas Cougar | CH Products Pro Pedals |

Pc1: AMD 5900x, RX6800xt, 32Gb 3600 CL19, Gigabyte B550 Aorus Pro | Pc2: AMD 2700, GTX 1080ti, 32Gb 3200 CL16, MSI x370 Gaming Carbon

Samsung Odyssey, Pimax 8kx, HP G2 (incoming), Pimax 5k+ (ex), Pimax 4k (ex), Vive Pro (ex), Oculus CV1 (ex), Oculus CV1 (ex)

Link to post
Share on other sites

@RedX
Can you bench it with our bench setup from

Maybe delete the shader files after swapping cards.


Edited by med0ra
shader..

| F-16C | FC3 | TM Hotas Cougar | CH Products Pro Pedals |

Pc1: AMD 5900x, RX6800xt, 32Gb 3600 CL19, Gigabyte B550 Aorus Pro | Pc2: AMD 2700, GTX 1080ti, 32Gb 3200 CL16, MSI x370 Gaming Carbon

Samsung Odyssey, Pimax 8kx, HP G2 (incoming), Pimax 5k+ (ex), Pimax 4k (ex), Vive Pro (ex), Oculus CV1 (ex), Oculus CV1 (ex)

Link to post
Share on other sites

@med0ra

Tested with that mission 1 you suggested, both 6800XT and 2080Ti. And, well, would you guess, the result was the opposite: now 6800 XT produced slightly faster CPU frame times than 2080Ti. Key figures below.

6800XT:  CPU 50% 13.6ms;  CPU 99% 22.1ms;  GPU 50% 10.6ms;  GPU 99% 14.6ms

2080Ti:  CPU 50% 14.5ms;  CPU 99% 22.1ms;  GPU 50% 11.5ms;  GPU 99% 15.7ms

 

Despite all the card swapping, driver reinstallation etc. the results for instant action missions (my first post) are still repeated and there, when standing on the ramp / ship, the Radeon card is more taxing on the CPU than Nvidia.

 

I can only conclude that the changes in CPU load may vary greatly depending on the mission. Perhaps the computational workload of different graphics items is differently distributed among the CPU and GPU. Now I have no clue whether I should keep running 6800XT instead of 2080Ti despite the larger amount of VRAM 🤔

Intel Core i7 9700K@4.6GHz|ASUS Strix Z390-H|G.Skill DDR4 32GB@3200MHz|MSI GeForce RTX 2080 Ti|2xSSD 500GB Samsung 970 EVO M.2 RAID0|SSD Samsung 850 EVO|Corsair HX 750W|Fractal Design Define R6 Blackout|UWQHD 34"|Valve Index|Logitech G9|MS Natural Ergonomic Keyboard 4000|VPC WarBRD Base+MongoosT-50CM2 Grip|VPC MongoosT-50 Throttle|VKB T-Rudder|Beyerdynamic DT770|Boom mic|Windows 10 Home

Link to post
Share on other sites
2 hours ago, RedX said:

Perhaps the computational workload of different graphics items is differently distributed among the CPU and GPU.

 

That would be the conclusion...and very interesting. Maybe the Def´s would be so nice and let us know about some details @BIGNEWY?

In your case i would have a look at the scenarios you play most and see which one runs better there.

Unfortunately we dont have the F18 or Carrier Module to test your bench setup.


Edited by JayRoc

FC3 | F-14A/B | F-16C | SA342  R5 5600X | 32GB | RX 6800XT | TM Cougar | T-Rudder mk.IV | HP Reverb G2

Thrustmaster Cougar Throttle USB standalone MOD

Link to post
Share on other sites

Dont forget to check MP, if you play that too.

| F-16C | FC3 | TM Hotas Cougar | CH Products Pro Pedals |

Pc1: AMD 5900x, RX6800xt, 32Gb 3600 CL19, Gigabyte B550 Aorus Pro | Pc2: AMD 2700, GTX 1080ti, 32Gb 3200 CL16, MSI x370 Gaming Carbon

Samsung Odyssey, Pimax 8kx, HP G2 (incoming), Pimax 5k+ (ex), Pimax 4k (ex), Vive Pro (ex), Oculus CV1 (ex), Oculus CV1 (ex)

Link to post
Share on other sites

CPU frametimes on MP is a difficult test to perform reliably as the workload on MP servers can vary a lot during the time it takes to swap the cards. To tackle this, a cycle of quite a few card swaps would be required in order to make the differences of 10-20% observed reliably. Maybe 3-6 times back-and-forth. I don't plan to perform such a laborous testing at this time. To avoid it, one could have a custom, non-changing MP server with no other clients on it, but it would probably not be any different from SP testing.


Edited by RedX

Intel Core i7 9700K@4.6GHz|ASUS Strix Z390-H|G.Skill DDR4 32GB@3200MHz|MSI GeForce RTX 2080 Ti|2xSSD 500GB Samsung 970 EVO M.2 RAID0|SSD Samsung 850 EVO|Corsair HX 750W|Fractal Design Define R6 Blackout|UWQHD 34"|Valve Index|Logitech G9|MS Natural Ergonomic Keyboard 4000|VPC WarBRD Base+MongoosT-50CM2 Grip|VPC MongoosT-50 Throttle|VKB T-Rudder|Beyerdynamic DT770|Boom mic|Windows 10 Home

Link to post
Share on other sites

I would take a more practical aproach. Test a couple of crowded servers with one card, smap it and do the same again.

That might not be scientific but if there is a difference of 10-20% that should be noteable in the favor of one of them.

 

I dont have the RTX3080 anymore, so i cant test it.

FC3 | F-14A/B | F-16C | SA342  R5 5600X | 32GB | RX 6800XT | TM Cougar | T-Rudder mk.IV | HP Reverb G2

Thrustmaster Cougar Throttle USB standalone MOD

Link to post
Share on other sites
On 12/20/2020 at 11:33 AM, RedX said:

exil, did you clean up nvidia drivers with utility like DDU when you switched the card? Or did you just insert the new card and leave all the old drivers around? Did you reinstall steamVR? I wonder whether it matters which output port is used for VR headset, too.

 

The RTX 2080Ti is back in the computer and CPU frame times are the same as before as recorded by fpsVR.

I think I'll try to install 6800 XT again and see if the result changes by tinkering with the installation procedure or settings.

Sorry for the late response. No, i uninstalled nvidia drivers without using ddu and put my 6900 in. But i didn't measure scientifically what happend to my CPU frametimes. I just saw, that the were mostly in the green in fpsvr and was happy with it.

 

But i noticed another strange behavior: My GPU frametimes sometimes went up to around 15ms in GPU heavy scenarios on the syria map. When that happend with my 1080 and reprojection on (HP Reverb G1) i was still able to maintain 45fps.

Now, with the same frametimes in the 6900XT it is sometimes locked at 30fps (although the headset is running with 90hz). When i switch off reprojection i am seeing around 50 - 60fps in the same area. So idk if its a WMR problem or an AMD drivers problem that causes it. (But maybe that is worth another topic and i don't wan't to hijack your threat).


Edited by exil
Link to post
Share on other sites
2 hours ago, Lurker said:

Quick question what do you guys use to check FPS in-game and or frametime's? I am on the Oculus Rift S, I use standalone DCS, and don't have SteamVR installed. Thank you. 

We use fpsvr, a tool which measures and indicates your gpu and cpu frametimes Realtime. I am afraid you can only use it for wmr headsets using steamvr. It won't work with oculus software. 

Link to post
Share on other sites
2 hours ago, exil said:

We use fpsvr, a tool which measures and indicates your gpu and cpu frametimes Realtime. I am afraid you can only use it for wmr headsets using steamvr. It won't work with oculus software. 

 

Is there something I can use with the Rift S? Even a FPS counter would be useful at this point. 

Link to post
Share on other sites
2 minutes ago, Lurker said:

 

Is there something I can use with the Rift S? Even a FPS counter would be useful at this point. 

Ctrl+Pause shows ingame fps counter and frametime (combined cpu+gpu frametime I think)

  • Like 1

Asus Prime X570-P * Ryzen 5800X * RTX 3080 TUF OC * G.Skill 32GB DDR4 3600MHz * 2 x 256Gb M.2 * Samsung 850 EVO 512Gb SSD * Kingston HyperX 120Gb SSD * Corsair RM850x V2/2018 * HP Reverb G2 * CH Fighterstick/Pedals/Throttle

Link to post
Share on other sites

I'm not sure whether frametimes shown by DCS match those shown by fpsVR. Do they?

Intel Core i7 9700K@4.6GHz|ASUS Strix Z390-H|G.Skill DDR4 32GB@3200MHz|MSI GeForce RTX 2080 Ti|2xSSD 500GB Samsung 970 EVO M.2 RAID0|SSD Samsung 850 EVO|Corsair HX 750W|Fractal Design Define R6 Blackout|UWQHD 34"|Valve Index|Logitech G9|MS Natural Ergonomic Keyboard 4000|VPC WarBRD Base+MongoosT-50CM2 Grip|VPC MongoosT-50 Throttle|VKB T-Rudder|Beyerdynamic DT770|Boom mic|Windows 10 Home

Link to post
Share on other sites
On 12/22/2020 at 10:29 PM, exil said:

Sorry for the late response. No, i uninstalled nvidia drivers without using ddu and put my 6900 in. But i didn't measure scientifically what happend to my CPU frametimes. I just saw, that the were mostly in the green in fpsvr and was happy with it.

 

But i noticed another strange behavior: My GPU frametimes sometimes went up to around 15ms in GPU heavy scenarios on the syria map. When that happend with my 1080 and reprojection on (HP Reverb G1) i was still able to maintain 45fps.

Now, with the same frametimes in the 6900XT it is sometimes locked at 30fps (although the headset is running with 90hz). When i switch off reprojection i am seeing around 50 - 60fps in the same area. So idk if its a WMR problem or an AMD drivers problem that causes it. (But maybe that is worth another topic and i don't wan't to hijack your threat).

 

If something behaves weired, delete shader files from dcs. On reprojection with my pimax and pitool, i have allways half the native Hz of the HMD maximum!?

| F-16C | FC3 | TM Hotas Cougar | CH Products Pro Pedals |

Pc1: AMD 5900x, RX6800xt, 32Gb 3600 CL19, Gigabyte B550 Aorus Pro | Pc2: AMD 2700, GTX 1080ti, 32Gb 3200 CL16, MSI x370 Gaming Carbon

Samsung Odyssey, Pimax 8kx, HP G2 (incoming), Pimax 5k+ (ex), Pimax 4k (ex), Vive Pro (ex), Oculus CV1 (ex), Oculus CV1 (ex)

Link to post
Share on other sites

Thanks for the reply. Did delete Metashaders and fxo. 

Sure, half the native hz would be 45. But I sometimes dip down to 30fps, but my frametimes are in a 14 - 15ms range. But personally, I think it's not the card itself but more a steamvr problem. 

I'll investigate further after Christmas and let you know. 

 

By the way: Merry Christmas! 

Link to post
Share on other sites

Merry Christmas 🙂
And yes, SteamVr is a pile of ... WMR users still need to run so much software in the background to play in VR. The more software produced by different developers, the greater chance something will fall apart.

AMD Ryzen 5900X @ 4.95 Ghz / Asus Crosshair VII X470 / 32 GB DDR4 3600 Mhz Cl16 / Radeon 6800XT / Samsung 960 EVO M.2 SSD / Creative SoundBlaster AE-9 / HP Reverb G2 / VIRPIL T-50CM /
Thrustmaster TPR Pendular Rudder Pedals / Audio Technica ATH-MSR7

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
  • Recently Browsing   0 members

    No registered users viewing this page.

×
×
  • Create New...