Months of testing 6900xt vs 3080 - final conclusions.

[TED]

As some may have read in my previous posts, I ended up in the lucky problem of having both a amd. 6900xt which I ordered a while back, having previously ordered a rtx 3080 which had a years waiting time! Now I have had both for the last few months and run them through their paces on dcs. 
Both are great gpus but one is better for my use. They seem to behave very differently and have their own strengths and weaknesses. How they render, image quality and interaction with the software and hardware. 
 

im using a reverb g2 headset, 32gb cl14 3600mhz ram and an AMD 3800xt cpu.
 

I spent the first few months solely with the 6900xt and it worked very well although needed quite a bit of tweaking. It runs cooler than the 3080 and in sp I can run pretty much max settings. The amd software is nice and pretty easy to configure although many of the settings are not working with dcs. In general in sp I had no problems maintaining 80fps (without mv) in a harrier even at low level on Syria map. 
I have the steam resolution set to as close to g2 resolution as I could which works out around 52% for me then set the per app setting to 100% to keep it simple. Motion smoothing on. Wmr motion vector enabled. 
The issues I started to have however seems to be once the map gets more populated and frame times would ramp up quickly to orange levels 17ms plus. Interestingly it was cpu frame times that seems to suffer the most but gpu spikes and frames were also suffering. I started to suspect my cpu was bottlenecking or throttling down and ran a number of checks. Once into MP servers frame times really struggled and I was getting numerous spikes with the settings described above. I turned shadows off, clutter and grass right down, trees down and forest visibility down. Nothing really helped and my cpu and gpu frame times were regularly spiking leading to a very jittery experience. 
 

back and forwards many times between the 6900xt and the 3080 flying the exact same missions and profiles proved to me that the 3080 for some reason works better in vr with the g2. With the exact same settings and mission as I tested on the 6900xt , Syria from Ramat low level to the north, engage some sa8 sites then a troop of btr80. Both frame times for cpu and gpu were significantly lower and almost always in green levels and under 12. In fact the cpu was hardly ever out of the green. This really confuses me as all I did was switch the gpu, however with the 6900xt installed the cpu is also working considerably harder. I have absolutely no explanation for this. I did have some issues with dcs crashing with the 3080. It would start with the audio stopping completely, followed by a big gpu spike then sometimes ctd, sometimes continuing with no audio. After several days of frustration thinking my cpu was screwed I noticed by resetting the audio device in dcs options it would work again, although often only to fail later. The culprit ended up most likely being the usb cable from the g2 or the USB port on the motherboard. Either way switching out to another Ss usb solved this. I never experienced this with the 6900xt, but there I was able to use the usb c directly into the gpu without using an adapter. The 3080 has no usb c port. 
 

I switched over to MP with the 3080 and again the same results with a a very stable 45fps with mv enabled and nice low frame times (10 - 15ms mostly) and a noticeably smoother experience. In fact going straight from one gpu to the next the difference was very obvious in the lack of any jitters on the 3080. 
 

Nvidia cp settings I had used were almost everything off with the exceptions of negative LOD BIAS clamped, triple buffering enabled, vr pre rendered frames set to 4. 
 

in AMD software I tried all sorts of combinations from rage mode to undervolting but could not get the frametimes down to the same levels as the 3080. 
 

For the most part I had used @BIGNEWYguide for vr settings as a start point and it’s pretty much bang on. I can certainly increase these settings quite a ways for sp but I am most interested in mp, so his settings work very nicely. I am not using any shader mods. I tried Fholger fsr for a while with both and it works nicely but I didn’t like the image so much so ditched it. 
 

I am fairly sure with a bit more technical knowledge the 6900xt can be tweaked to improve the results but I didn’t want to get into too much messing about with system settings. What I really dont understand is why both the gpu and cpu were working so much harder and both producing upwards of 40% higher frametimes on the exact same mission setup. One possibility is that my 6900xt was pushing out more than the cpu could keep up with and an overload of some sort is occurring. Possibly over clocking my cpu more or upgrading the cpu may have helped.
 

Anyway i may continue testing but for now it looks 3080 suits me better. There are so many variables it’s very likely others will have completely opposite experience and results to mine but hope this helps anyone looking at gpu setups.

[streakeagle]

I turned off motion smoothing, now my 6900 XT runs butter smooth with a first generation G2 with the new usb cable. I can fly in an F-14A from the supercarrier and sustain 45 fps with graphics settings very high, with mirrors, shadows, and heat blur. In many situations, the frame rate rises up to 60 fps or more. At this point I am very happy with the 6900XT. If the price had not been ridiculously high, I would have liked a 3090.

[DeltaMike]

Thank you so much for posting your results.  

I always had this idea in my head that the CPU did its thing, then the GPU did its thing, then we see a picture.  I think that work flow has been diced up quite a bit in the process of optimizing VR, which I think means CPU and GPU frame times should covary.  I think.

Back in the day, it was pretty clear that almost every "graphics" setting in DCS affects CPU as much as GPU, with the notable exceptions being supersampling and msaa. 

I think what your data shows is, the 6900xt can hold it's own in the supersampling and msaa department, for example when you're just flying around by yourself. But when things get busy it has a hard time keeping up with all the stuff the CPU is demanding of it. 

I think it follows, then, that tuning the 6900xt may be a matter of limiting the CPU workload, for example by reducing vis range or trees.  Good news for people who like their anti-aliasing I guess.

Unfortunately my computer is in a million pieces right now so I can't test this hypothesis, but I suspect I'll do what I've always done -- set vis range to medium, turn off shadows and trees in MP, and just be thankful I can read my instruments now 

5600x, 6900xt (at $1350, the least bad buy on the market imo), and G2, soon. 

 

[TED]

I should just add that I have not yet replaced my power cable for the g2, but I do wonder if this was causing some of the more negative things for the 6900xt. I notice that it is very sensitive to exactly which USB port it uses and when the stuttering happens there are large spikes in both cpu and gpu. This could indicate a power supply issue to the headset. At some point I will work out how to get the newer cable and try again.

My results are purely my own experience with all the variables I have, so it is entirely possible there are other external factors at play. I was only using the usb c port for example on the 6900xt for the g2 power. This might be worth investigating. 

[DeltaMike]

Yeah.  VR is all about the latency, everything that happens between when you click the mouse, and see the result on your screen.  If you're having usb problems that's an issue for pretty much everything except the mouse innards I guess, which fortunately seem to be working at present.  Between that and power spikes, VRAM management, PCIE issues (did I forget anything?) I guess AMD's spanking new chipset is a work in progress.  

Intel's consumer-level article on VR optimization here 

Here's how I look at it.  Of the three things we might want --

1. Resolution (MFD clarity, spotting)*

2. Eye candy (clouds, shadows)

3.  Anti-aliasing (supersampling, msaa)

-- with any of the top tier cards (3080, 3090, 6900xt) and a top tier headset (eg Reverb, Pimax) it seems safe to say you can get your choice of two, but perhaps not all three at the same time.    

My only question is, is it really a meaningful test, to try to max out all three on a busy PVP server.  There, you're inclined to turn your anti-aliasing down, if not off altogether to make spotting easier.  And of course shadows and trees are irrelevant, in fact shadows could be a distraction if you're really competitive.  You get absolutely nothing by setting vis range to anything more than medium, and run the risk of being CPU bound if you do.  

I'd run the MP test without MSAA, with vis range medium, and with supersampling at 100% (or to be more precise, leave the barrel distortion correction in at 140%, but no more than that -- set the *sliders* at 100% and ignore the pixel count).  Granted, my money is still on the 3080 -- suspect its memory bandwidth gives it an advantage no matter how you slice it -- but I think it would be a more relevant test, and speaks a bit more to what some of us perceive to be the weakness of the 3000 series cards; that we are paying (a lot) for technology we aren't realistically gonna fully leverage in DCS. 

Ebay prices as of first week of Jan 2022:  6900xt $1350, 3080 ~$1800.  You make a strong argument, that the 3080 is the superior card.  My question is, is the delta worth it? 

----

*ie, not having to under-sample to handle the load

Edited January 6, 2022 by DeltaMike

[TED]

@DeltaMikeabsolutely agree. It's always a compromise and tuning and tweaking is necessary no matter what the chosen hardware is.

I should also have mentioned, the test I described was definitely not the only one I did, it was only the final one I did based on my requirements. I actually did also test as u described with msaa off and dialling back many bother settings then comparing. Way too many configurations to describe. Throughout, however was the common theme, for the same given settings in vr with the g2 i was producing lower frame times on cpu always with 3080 and GPU times were similar until I started populating the map or using mp server. At which point the 3080 pulled away. Something is going on in the interactions between my GPU and cpu when using 6900 that is not happening when using the 3080.

[JayRoc]

If you want to get rid of the jitters on the AMD...go to

SteamLibrary\steamapps\common\MixedRealityVRDriver\bin\win64OpenVRSettingsUX.exe

And turn on the Reprojection in Steam.

There are a lot of people saying its the same settings and it would make no difference...but they obviously never tried it and have no clue what they´re talking about.

Edited January 6, 2022 by JayRoc

[RED]

Yep, @JayRoc. That helped me as well. Also removed any artifacts I had.

[Archduke]

That helped me too. I had really evil looking artefacts before, lines on HUD and MFDs wriggling like mad. Since I turned motion vector in WMR off, it’s been smooth sailing.