VR Hardware Benchmark late 2020 - Ryzen 2000/5000 - 1080ti vs 3080 vs 6800XT

[med0ra]

On 12/19/2020 at 9:34 AM, helios123 said:

Does anyone have an explanation as to why the 6800XT is performing better than RTX 3080, when in most non-vr games 3080 is on par if not slightly better?

On our VR testing, the CPU utilisation was lower with amd 6000 cards. So there was more room for dcs.

[JayRoc]

On 12/19/2020 at 9:34 AM, helios123 said:

Does anyone have an explanation as to why the 6800XT is performing better than RTX 3080, when in most non-vr games 3080 is on par if not slightly better?

The amd are better at simple rastarisation...thats why they're faster in DCS and not in games with modern engines. 

[blkspade]

23 hours ago, JayRoc said:

The amd are better at simple rastarisation...thats why they're faster in DCS and not in games with modern engines. 

Nvidia's focus seems to be specifically 4K and raytracing. The Nvidia 3000 series cards are somehow tuned for 4k that results in them scaling poorly at lower resolutions compared to AMD 6800 cards. The limited VRAM potentially comes into play with the 3080 with supersampling in VR. Its one of the reasons why I'm not interested in stepping down to 10GB from my 1080ti.

[Gladius]

Today I replaced my AMD Ryzen 2600X with a Ryzen 5600X.
I am actually quite surprised that with the Ryzen 5600X the performance is very Stabiel, partly I had with the Ryzen 2600X 22-45 FPS now I reach satabiele 45-65 FPS with a 1080 TI and just still Oculus Rift CV1.
Two weeks ago I replaced the Ram 2666 with 3200, which also made a small contribution.
I am satisfied for the first time. So I can well wait for a 3080 or 3080 TI until also then the availability and prices votes.
Find the thread very interesting and very informative.

[VitS]

That's quite interesting results. I tested DCS with both 3080 and 6800xt, but 3080 was consistently faster than 6800xt. Did you undervolt 3080? 

Edited December 22, 2020 by VitS

[JayRoc]

No the 3080 wasn´t undervolted, it was overclocked to run consitenly on 2070Mhz.

I think you have to give a little more details on your setup and what and how you testet.

 

There is another topic where someone found that in a different scenario the CPU load with the RTX is lower.

Due to that he gets more frames in his scenario.

 

17 hours ago, blkspade said:

Nvidia's focus seems to be specifically 4K and raytracing. The Nvidia 3000 series cards are somehow tuned for 4k that results in them scaling poorly at lower resolutions compared to AMD 6800 cards. The limited VRAM potentially comes into play with the 3080 with supersampling in VR. Its one of the reasons why I'm not interested in stepping down to 10GB from my 1080ti.

 

The Problem with the AMD´s is that the infinity cache doesn´t work as good on high resolutions like 4K. And therefore they lose their advantage.

They should have made the Vram interface faster with 512bit or something in between.

Edited December 22, 2020 by JayRoc

[med0ra]

On 12/21/2020 at 7:50 PM, blkspade said:

Nvidia's focus seems to be specifically 4K and raytracing. The Nvidia 3000 series cards are somehow tuned for 4k that results in them scaling poorly at lower resolutions compared to AMD 6800 cards. The limited VRAM potentially comes into play with the 3080 with supersampling in VR. Its one of the reasons why I'm not interested in stepping down to 10GB from my 1080ti.

No. The memory bandwith on AMD cards is lower. The chose it, because it is cheaper and more power efficent. To overcome the small bandwith, they added the infinity cache 128MB direct on the gpu chip. The prediction for what to chache does not work as good as they thought. So in 4k where you need bandwith, the amds cant hold against the raw performance of the nvidia cards. The chose the just working method over new tech. Nvidia focus is on RT. But there is no card which can handle RT at 4k. DLSS is nice, but i think it should be 3.0 to get very good results.

Edited December 24, 2020 by med0ra

[Bog9y]

Interesting that your results are that the 6800XT performs better. Babeltech found this:

 

https://babeltechreviews.com/vr-wars-the-rx-6800-xt-vs-the-rtx-3080-15-vr-games-performance-benchmarked/

[Hiromachi]

On 12/22/2020 at 1:23 PM, JayRoc said:

No the 3080 wasn´t undervolted, it was overclocked to run consitenly on 2070Mhz.

I think you have to give a little more details on your setup and what and how you testet.

Could you be so kind and share your Radeon settings ? 
I have received yesterday 6800XT Asus Strix LC and thought I could compare it with RTX 3080. So first I tested 3080 using your methodology in same benchmark with same in-game settings. GPU only had increased power limit to 110 %, Temperature limit and memory speed by 400 Mhz. So it was a bit overclocked.

Result was this:

Spoiler

fpsVR Report:
App: DCS HMD: HP Reverb Virtual Reality Headset G20 (90.001 Hz, IPD 63.3)
GPU: NVIDIA GeForce RTX 3080 (27.21.14.6089) CPU: AMD Ryzen 9 5900X 12-Core Processor            
Delivered fps: 62.21  Duration: 2.6min.
GPU Frametimes:
Median: 12.1 ms
99th percentile: 16.4 ms
99.9th percentile: 17.3 ms
frametime <11.1ms(vsync): 19.6%
CPU frametime:
Median: 8.6 ms
99th percentile: 14.2 ms
99.9th percentile: 17.4 ms
frametime <11.1ms(vsync): 92.1%
Reprojection Ratio: 31.0% (for Index/Vive/VivePro headsets only)
Dropped frames: 3 or 0.0% (for Index/Vive/VivePro headsets only)
Max. SteamVR SS: 100%
Render resolution per eye: 3160x3092(by SteamVR settings, Max.) (HMD driver recommended: 3162x3093)

 

Than I swapped to Radeon. First I wiped drivers with DDU, than installed newest Adrenalin. I did not fiddle much in GPU driver and Adrenalin, switched only those three options

• Morphological Anti-Aliasing - OFF (i.e, "Disabled").
• "Surface Format Optimization" - OFF  (i.e, "Disabled").
• "Tessellation" - change from "AMD Optimised" to "Override Application Settings" then set at 16x

as user recommended in another thread:

I also presume that Radeon Image Sharpening for VR is not beneficial for FPS at all ? I also set it as user recommended but can remove it. This is the result of same situation test:

Spoiler

fpsVR Report:
App: DCS HMD: HP Reverb Virtual Reality Headset G20 (90.000 Hz, IPD 61.9)
GPU: AMD Radeon RX 6800 XT (27.20.14501.33001) CPU: AMD Ryzen 9 5900X 12-Core Processor            
Delivered fps: 57.58  Duration: 2.6min.
GPU Frametimes:
Median: 11.6 ms
99th percentile: 15.9 ms
99.9th percentile: 17.4 ms
frametime <11.1ms(vsync): 30.2%
CPU frametime:
Median: 8.2 ms
99th percentile: 13.5 ms
99.9th percentile: 16.2 ms
frametime <11.1ms(vsync): 97.1%
Reprojection Ratio: 48.3% (for Index/Vive/VivePro headsets only)
Dropped frames: 1 or 0.0% (for Index/Vive/VivePro headsets only)
Max. SteamVR SS: 100%
Render resolution per eye: 3160x3092(by SteamVR settings, Max.) (HMD driver recommended: 3162x3093)

So GPU delivered slightly fewer FPS but also had lower frametime. Max clock I have observed was 2412 Mhz, which is quite a bit less than this GPU can actually reach. Also GPU did not hold its clocks and I saw drops. 
I'm not proficient in Adrenalin at all so having your settings used for Reverbg G2 testing would help me a lot  

[exil]

53 minutes ago, Hiromachi said:

Could you be so kind and share your Radeon settings ? 
I have received yesterday 6800XT Asus Strix LC and thought I could compare it with RTX 3080. So first I tested 3080 using your methodology in same benchmark with same in-game settings. GPU only had increased power limit to 110 %, Temperature limit and memory speed by 400 Mhz. So it was a bit overclocked.

Result was this:

  Hide contents

fpsVR Report:
App: DCS HMD: HP Reverb Virtual Reality Headset G20 (90.001 Hz, IPD 63.3)
GPU: NVIDIA GeForce RTX 3080 (27.21.14.6089) CPU: AMD Ryzen 9 5900X 12-Core Processor            
Delivered fps: 62.21  Duration: 2.6min.
GPU Frametimes:
Median: 12.1 ms
99th percentile: 16.4 ms
99.9th percentile: 17.3 ms
frametime <11.1ms(vsync): 19.6%
CPU frametime:
Median: 8.6 ms
99th percentile: 14.2 ms
99.9th percentile: 17.4 ms
frametime <11.1ms(vsync): 92.1%
Reprojection Ratio: 31.0% (for Index/Vive/VivePro headsets only)
Dropped frames: 3 or 0.0% (for Index/Vive/VivePro headsets only)
Max. SteamVR SS: 100%
Render resolution per eye: 3160x3092(by SteamVR settings, Max.) (HMD driver recommended: 3162x3093)

 

Than I swapped to Radeon. First I wiped drivers with DDU, than installed newest Adrenalin. I did not fiddle much in GPU driver and Adrenalin, switched only those three options

• Morphological Anti-Aliasing - OFF (i.e, "Disabled").
• "Surface Format Optimization" - OFF  (i.e, "Disabled").
• "Tessellation" - change from "AMD Optimised" to "Override Application Settings" then set at 16x

as user recommended in another thread:

I also presume that Radeon Image Sharpening for VR is not beneficial for FPS at all ? I also set it as user recommended but can remove it. This is the result of same situation test:

  Hide contents

fpsVR Report:
App: DCS HMD: HP Reverb Virtual Reality Headset G20 (90.000 Hz, IPD 61.9)
GPU: AMD Radeon RX 6800 XT (27.20.14501.33001) CPU: AMD Ryzen 9 5900X 12-Core Processor            
Delivered fps: 57.58  Duration: 2.6min.
GPU Frametimes:
Median: 11.6 ms
99th percentile: 15.9 ms
99.9th percentile: 17.4 ms
frametime <11.1ms(vsync): 30.2%
CPU frametime:
Median: 8.2 ms
99th percentile: 13.5 ms
99.9th percentile: 16.2 ms
frametime <11.1ms(vsync): 97.1%
Reprojection Ratio: 48.3% (for Index/Vive/VivePro headsets only)
Dropped frames: 1 or 0.0% (for Index/Vive/VivePro headsets only)
Max. SteamVR SS: 100%
Render resolution per eye: 3160x3092(by SteamVR settings, Max.) (HMD driver recommended: 3162x3093)

So GPU delivered slightly fewer FPS but also had lower frametime. Max clock I have observed was 2412 Mhz, which is quite a bit less than this GPU can actually reach. Also GPU did not hold its clocks and I saw drops. 
I'm not proficient in Adrenalin at all so having your settings used for Reverbg G2 testing would help me a lot  

Could you be so kind and post your in-game settings and the mission you benched with as I would like to compare your results to my 6900XT?

 

Edit: and I am, of course, also interested in the Radeon settings. 

Edited December 24, 2020 by exil

[Hiromachi]

Same settings and benchmark as in OP. You just set VR preset in settings. And mission is available to download in OP (Gryz benchmark mission). 

[exil]

8 minutes ago, Hiromachi said:

Same settings and benchmark as in OP. You just set VR preset in settings. And mission is available to download in OP (Gryz benchmark mission). 

Thanks a lot! 

[exil]

By the way: did any of you amd owners already tested the Sam VS. non-SAM in DCS? If it's close to 10% I would think about updating my mainboard. 

[Hiromachi]

I will do that test. First I want to tweak GPU settings to optimize it. Than repeat benchmarks to compare with tuned RTX 3080. Next I will OC it, since its liquid cooled and in all reviews they managed to push it to at least 2.7 Ghz, I will try that and bench again. Last I want to try SAM vs no-SAM (my Crosshair VII with 5900X already has that).

[exil]

1 hour ago, Hiromachi said:

I will do that test. First I want to tweak GPU settings to optimize it. Than repeat benchmarks to compare with tuned RTX 3080. Next I will OC it, since its liquid cooled and in all reviews they managed to push it to at least 2.7 Ghz, I will try that and bench again. Last I want to try SAM vs no-SAM (my Crosshair VII with 5900X already has that).

That sound great but also a lot of work! Eagerly waiting for the results! 

[Hiromachi]

Well, I wont be making 10 or 20 attempts for each setup. Three at most. So thats 9 runs, each is about 2.6 minutes. It's not that bad. 
I just need good config for Radeon Adrenalin that OP used to try keeping tests uniform.

[JayRoc]

What me really wonders is why 2.6 minutes? Ours have been 3.2 Minutes for the whole benchmark run. 

 

We used the standard settings that adrenaline comes with.

Same with the Nvidia driver. 

[JayRoc]

@Hiromachi

 

Your benchmark results make no sense to me. The Median Frametimes and the percentiles of the radeon ist less, which is better and faster. But still the fps are supposed to be less? Our duration has been 3.2minutes on every run. 

 

I'd say there is something wrong there. 

Edited December 24, 2020 by JayRoc

[Hiromachi]

Umm, I dunno. It shows to start benchmark and I press reset, than it shows to stop and I press reset again. No idea. 

 

Well, than it is weird. I will hit reset button on my Adrenaline and try it again. 

 

Edit. Wait, you sure about default settings ?

Cause this directly says 6800XT OC:

And I dont see how without overclocking I can make 13 FPS.

Edited December 24, 2020 by Hiromachi

[Hiromachi]

Well, again, from the moment it says start benchmark to end benchmark its roughly 2.6 - 2.7 min (benchmark v1 complex mission). I defaulted driver and reset DCS to VR preset. And results are fairly similar:

fpsVR Report:
App: DCS HMD: HP Reverb Virtual Reality Headset G20 (90.000 Hz, IPD 61.9)
GPU: AMD Radeon RX 6800 XT (27.20.14501.33001) CPU: AMD Ryzen 9 5900X 12-Core Processor            
Delivered fps: 58.2  Duration: 2.7min.
GPU Frametimes:
Median: 11.5 ms
99th percentile: 15.5 ms
99.9th percentile: 15.9 ms
frametime <11.1ms(vsync): 34.2%
CPU frametime:
Median: 8.1 ms
99th percentile: 13.9 ms
99.9th percentile: 16.5 ms
frametime <11.1ms(vsync): 96.8%
Reprojection Ratio: 48.3% (for Index/Vive/VivePro headsets only)
Dropped frames: 1 or 0.0% (for Index/Vive/VivePro headsets only)
Max. SteamVR SS: 100%
Render resolution per eye: 3160x3092(by SteamVR settings, Max.) (HMD driver recommended: 3162x3093)

[exil]

Are you sure you're benching the correct mission? Iirc the 'heavy' mission lasted a bit longer than the not so heavy mission. That could at least explain the missing frames. 

[Hiromachi]

Yes, as quoted: benchmark v1 complex mission

 

Anyhow, the problem may be that GPU does not seem to be holding usage. It jumps between 65 and 95 %. 

[exil]

11 hours ago, Hiromachi said:

Yes, as quoted: benchmark v1 complex mission

 

Anyhow, the problem may be that GPU does not seem to be holding usage. It jumps between 65 and 95 %. 

I thought that too because mine was doing the same, but this seems to be pretty normal in modern GPU architectures. GPUs are altering every now and then to not overheat. If your Gpu would run at >90% all the time it would simply overheat. 

At least this is what some far more experienced users in a hardware forum told me when I asked the same question. 

[Hiromachi]

Hmm, but that changing usage sometimes leads to FPS going below 45 and re projection point which in turn reduces smoothness.

I've noticed another user also reported that:

 Also, my 3080 whenever I did not have issues, was maintaining stable 85+ % usage and did not drop frames in such manner.

[JayRoc]

That's not true. 

Modern CPU clock down and throttle the power that goes through a CPU is they're getting too hot. The usage of the CPU will be 100% if it is fully used no matter if it clocks slower or faster. 

 

The usage of the CPU doesn't really matter because its mainly 1 thread of your CPU that is used to full extend from DCS which limits your frames.

 

@Hiromachiwhat hardware do you use? Did the other guy use DDU to deinstall the drivers aswell?

Edited December 25, 2020 by JayRoc