AMD RX 6800XT vs. Nvidia RTX 3080

[JayRoc]

Way more detailed testing in our second bench run over here!

 

 

So we got the lucky opportunity to test both graphic cards in different hardware combination.

I think there can be some conclusion drawn from these tests. The problem was that it was getting really late and we have a funny behavior with one constellation that we have to investigate.

 

 

System 5900X:

 

AMD Ryzen 5900X

Gigabyte B550 Aorus Pro

G.Skill Sniper 3600 CL19-20-20-20-38

Seperate SSD´s for Windows and DCS

 

System 5600X

 

AMD Ryzen 5600X

Gigabyte B550M Aorus Pro

HyperX 3200 CL16-18-18-18-38

Seperate SSD´s for Windows and DCS

 

 

XFX AMD RX6800XT in the stock AMD design.

It was testet in stock and with OC where the clock was increased to 2100Mhz and the Powertarget to 115%. SAM (CPU direct memory access) was activated.

 

Gigabyte RTX 3080 Aorus Master 10G with Stock OC over the reference design.

We testet it stock with 1845Mhz and OC with 1920Mhz. The archived clocks where actually higher but we didn keep track of that.

 

We testet both Systems with a Pimax 8KX which has two native 4K pannels with 2x 3840 x 2160.

The Pimax was set to the narrower field of view. I dont know which resolution the headset runs in that case. But the FOV is still way bigger compared to a Samsung Odyssey.

The problem here was that SteamVR had different native resolutions at 100% and different ratios aswell. On the 5900x it was 3152x3160 and on the 5600X 3336x3948.

We used Gyphis Benchmark File for testing. The first one with more CPU load.

 

In all tests the graphic card was the limiting factor. With this settings and this mission the faster CPU didn't give an advantagte.

We have other settings testet and there the AMD was way faster, but the results seem to be off. So we will investigate an test that again.

Unfortunately that have also been the test where we deactivated SAM to see if it increases the fps.

Very interesting are the frametimes which are way more consistent with the AMD card.

 

(above frametimes for AMD)

 

(above frametimes for Nvidia)

 

My conclusions so far:

 

With SAM and the drivers we have right know the AMD 6800XT is faster than the RTX3080 in DCS with very high resolutions in VR.

The framestimes of the AMD are more consistent than with the RTX.

Faster RAM timings are better than faster clocks.

 

What we will test at a later point:

 

SAM on/off

Both PC´s with excact same settings (if possible we havent found a solution for the 100% resolution problem)

RAM with fast clocks and timings

RX 6800 non XT

GTX 1080ti watercooled

All graphic cards with an AMD Ryzen 2700

Edited December 14, 2020 by JayRoc

[MadMonty]

Great overview and test setup. Thank you so much for posting this. Very helpful for those who are looking for an upgrade like me...

[BitMaster]

That 3600-CL20 kit....oh man, who sells 3600 with a CL20 latency ??? That's like a Porsche that comes with 3 cylinders missing.

 

3600 is usually CL18, better ones have CL16 and real good ones even lower. Never saw a CL20 kit.

 

Anyway, nice testing scenario and I am smiling to see the 6800XT beating Nvidia's 3080. WHo would have thought that 6 weeks ago.

[med0ra]

It is from my old system. They where pretty decent then and offered great speed for 180€. With 4 off them, i have the same FPS like the test with 3200MHz fast CL kit. But we will bench that also with the other systems.

[San]

Thanks Jayroc - really appreciate the effort you have gone to!

 

Is there any chance you have any comparative numebrs from your pre-upgrade GPU and CPU you could share so we can get a sense of what has driven the biggest uplift out of the CPU & GPU updates?

 

From my perspective I currently have a 3700x and a 5700XT but with a Reverb G2 arriving next week for my first ever foray into VR I wanted to work out which upgrade (3700x to 5600/5900x or 5700xt to 6800xt) would offer the best uplift when considerd against value.

 

Regardless of this - your insights above are already really appreciated.

Edited November 25, 2020 by San
Fixed typo

[reece146]

Would be interesting to see 3600CL16 tests - Infinity Fabric runs at 1800MHz and getting the RAM running in multiples of that speed increases performance - or so I've read.

 

The 3600CL20 has latency of 11.1 ns; 3200CL16 is 10 ns (i.e. better). 3600CL16 would be roughly 8.89 ns - combine that with the IF uplift...

Edited November 25, 2020 by reece146

[Raven (Elysian Angel)]

My conclusions so far:

 

With SAM and the drivers we have right know the AMD 6800XT is faster than the RTX3080 in DCS with very high resolutions in VR.

The framestimes of the AMD are more consistent than with the RTX.

 

That seems to be consistent with what we hear from Aurelius here on this forum. Thanks for testing: this is very welcome confirmation! :thumbup:

Consistency and smoothness is very nice to have in VR.

 

Infinity Fabric runs at 1800MHz and getting the RAM running in multiples of that speed increases performance - or so I've read.

Yes I heard that as well.

[med0ra]

The 3600CL20 has latency of 11.1 ns;

 

The Timings of that Ram are:

 

CAS Latency CL 19 (entspricht ~10.56ns)

Row-to-Column Delay tRCD 20 (entspricht ~11.11ns)

Row Precharge Time tRP 20 (entspricht ~11.11ns)

Active-to-Precharge Time tRAS 40 (entspricht ~22.22ns)

 

[med0ra]

Thanks Jayroc - really appreciate the effort you have gone to!

 

Is there any chance you have any comparative numebrs from your pre-upgrade GPU and CPU you could share so we can get a sense of what has driven the biggest uplift out of the CPU & GPU updates?

 

From my perspective I currently have a 3700x and a 5700XT but with a Reverb G2 arriving next week for my first ever foray into VR I wanted to work out which upgrade (3700x to 5600/5900x or 5700xt to 6800xt) would offer the best uplift when considerd against value.

 

Regardless of this - your insights above are already really appreciated.

 

We will test my old System with Ryzen 7 2700 and GTX1080ti.

[Lurker]

Very interesting. Thanks for the quick review. Looking forward to more detailed testing, maybe with downsampled resolution to see how it would perform with a Rift S type device?

[Lange_666]

5600 (with 3200 CL16) is 3 frames faster then the 5900 (with 3600 CL20). Would have been nice if they were compared with the same RAM.

[Cassiop]

eyh ! thanks for the insight. that's great to have those feedback.

 

if you want to save time, i did an article on reddit about how memory affect performances in DCS VR, check it out

 

also, could you please share Vram usage of the two card please ? On SteamVR i think we will see 3080's vram almost capping its 10gb but i wonder how the 16 gb of the 6800xt are filled. but i trully dont know the Pixmax so maybe its vram usage will be different from a regular htc Vive.

benchmark usually put the slightly 3080 in front of the 6800xt in 80% of games but since DCS have nothing to compare to anything... =D

[Leaderface]

We could end up in an scenario where both cards perform very similar when Drivers gets up to date in a couple of months and those are very good news from the buyer perspective.

AMD doing well in Vr: Great to know!

[Kazansky222]

Hey man I run an 8kx as well, I have this little executable that will re-write the steamvr.settings file to give you the same steam vr settings between restarts, just open it, and press Y 3 times and it resets everything,, else disables some of the steam crap that degrades image quality. Its also sort off a poor mans way to disable the steam vr auto resolution scaling.

you do have to run the executable every time you restart steam vr though.

 

You're getting about double my frames running an 8600k 5.1ghz, ,3200mmhz cl14, and 10800ti OC.

I have to run in upscaled mode to be playable. But in native you're about double my fps, if not a little more then double.

 

https://drive.google.com/file/d/1LnC...ew?usp=sharing

 

And if anyone is curious, vrfps shows 10.9 of 11gb of vram used.

 

@JayRoc Its the steamvr.settings "gpu speed" setting that is messing with your resolutions. just set it to 2000 with this executable or manually to "disable" resolution scaling. But it automatically changes after restart to you have to set it every time you want to play. Also can you test with amd smart caching on vs off. Curious to what the difference is in dcs.

[JayRoc]

Thanks mate, we will have look into it.

[prichardgs]

If only DCS would update it's graphics think of the possibilities. Still using an outdated graphics environment

 

 

Sent from my iPhone using Tapatalk

[eatthis]

what ingame settings do you run for this?

[Gladius]

Very interesting tests. When do you do more tests?

I would be interested in the effect with and without SAM on the RTX 3080.

How much is each GUP memory used?

[JayRoc]

The SAM feature is not released by Nvidia, its just with AMD. Not sure, i think it was 7,8GB.

Will all be more detailed in the next test that we will run probably next week.

[Andrei]

Thank you very much for those tests. There're so little benchmark info available for DCS done in consistent manner, that every work like this is extremely valuable for the community.

[Mustang]

The SAM feature is not released by Nvidia, its just with AMD. Not sure, i think it was 7,8GB.

Will all be more detailed in the next test that we will run probably next week.

 

SAM is coming to nvidia, on my ASUS STRIX Z490-F board I updated my BIOS and it now has the Resize BAR Support option

 

https://wccftech.com/nvidia-confirms...mory-sam-tech/

 

got the RTX 3080, just need the driver to make use of it :)

[Lurker]

@JayRoc

 

Is it possible for you to test using a Rift S, or maybe just to downsample the PiMax so that us peons with less resolution (Im using a Rift S) could see how those two cards perform at more standard VR resolutions?  Thank you. 

 

(Now if only I could find a 3080 or a 6800XT somewhere in stock) 

[EAF51_Jimmi]

On 11/25/2020 at 4:08 PM, JayRoc said:

Gyphis Benchmark File for testing

What is it?

i would like to make a similar test with my system...

On 11/26/2020 at 11:34 AM, Cassiop said:

if you want to save time, i did an article on reddit about how memory affect performances in DCS VR, check it out

Great article Cassiop much appreciated!

[Waxer]

@JayRoc Coming late to this topic: thank you for posting the methodology and conclusions. I am tempted by all this shiny new silicon being released and I have been considering an upgrade. Information is power etc. I'm not paying scalper prices so I have not rushed into anything, but maybe prices will be more sensible in the new year.

 

Appreciated!

[Mustang]

Update regarding SAM feature, the ASUS Z490-F has had a recent beta BIOS that permits the use of Resize BAR Support the same way the AMD x5xx boards do, just waiting on the software from nvidias side now

 

https://www.tomshardware.com/uk/news/asus-firmware-resizable-bar-support-intel-z490-motherboards

 

Edited December 3, 2020 by Mustang