RTX 3080 vs 6900 XT: Some findings

User Support Hub Hardware & Peripherals
1

Dear all,

since I’m facing the decision whether to keep a RTX 3080 Founders Edition or an AMD RX 6900 XT and all the “evidence” on the Internet is quite inconclusive, I did some testing of the cards on my own. In the following, I will illustrate some of my findings as this might be interesting to some people which are facing a similar decision.

Test system:

i7-10700K@5.0GHz
32GB DDR4-3600 RAM
OS and MSFS on SSDs

Scenarios:

In order to compare apples to apples, I tried to use scenarios which I could easily replicate and were not subject to unwanted variation. So for all the testing, I set the date to Jan, 15th of this year and the time to 12 noon. I used the Ultra preset in all scenarios and I’m running the sim in 4K. I also turned off all air traffic (so no AI or real time traffic). All fps and VRAM figures are from inside the cockpit. So no outside shots!

I chose scenarios where I would start at the runway of a large airport (exception Courchevel and Yosemite) facing the direction of a large city and I picked the runway so that the airport buildings were in sight. All the numbers for “on runway” were recorded while sitting on the runway, ready to take off. I waited a couple of seconds before recording the numbers to let the system shake off a bit. Then I took off and maintained runway heading and leveled off at 2500-3000 feet at a speed of roughly 250 knots. That way, I flew across the city to record the “Over city” numbers. The average fps were measured from take-off until I crossed the city. In the case of KLAS, which I used as a VRAM worst case, I had to make a sharp turn to the right after take-off to fly over the city (of course ;-)).

The KJFK, EGLL, KLAS and Sydney landing challenge scenarios were all flown in the stock A320. Courchevel in the TBM and Yosemite in the C172 G1000. The Yosemite scenario used an older version of the Yosemite Valley add-on from VFXSimmer and I started the scenario airborne at Half Dome and flew around Half Dome.

Both graphics cards were at stock performance. No overclocking. The VRAM is usage from the dev OSD in MSFS. So this represents usage not just allocation.

So here are the results:

6900 XT:

RTX 3080:

My key findings:

  1. The cards perform roughly the same in 4K. I really do not understand how some Youtubers and other testers on the net find that the 3080 outperforms the 6900 XT. This is just not the case. Apart from the scenarios shown here, I tested a lot of other scenarios and the 6900 XT never performed worse than the 3080.
  2. Even though the 6900 XT often uses more than 10GB of VRAM, the 3080 somehow manages to use less without performance suffering a lot. Why am I saying “a lot”? Because my very subjective impression is that the 6900 XT performs way smoother with much more even frame times. Most likely this is because it can use more VRAM.
  3. Another very subjective finding is that the 6900 XT has much sharper textures in the far distance, e.g. in mountainous terrain. Could also be, because it can load more textures into its VRAM.
  4. In the Yosemite scenario, which is also heavy on VRAM, the 3080 really struggled. The performance was choppy and some textures didn’t load at all. So Half Dome looked way better on the 6900 XT.

What’s my conclusion?

The 3080 can handle MSFS 2020 in 4K well. However, the 10 GB VRAM is a limitation. Not to the extent that performance tanks, but it has an impact on frame time variability and maybe also texture sharpness. Given that MSFS 2020 is going to move to DX12, I would assume that 10GBs will be even more of a problem. This is not saying that you are not able to enjoy MSFS in 4K on a 3080. It’s just that more VRAM improves the experience.

The 6900 XT performs excellently in MSFS2020. The performance is always smooth and the card really seems to be up to the task at all times. Together with the fact that XBox also is based on RDNA2, I speculate that AMD graphics cards will benefit from the move to DX12 and the optimizations that are coming to MSFS 2020 in the summer. All this is the reason why I’m going to keep the 6900 XT and sell the RTX 3080.

Best regards

7 Likes
2

Good luck with AMD drivers.
(Hint: read amd drivers thread :wink:)

1 Like
3

The one reason I ditched AMD years ago; poor computational performance compared to Nvidia, with poor drivers.

4

That story is getting really old. But good luck with your NVidia cards.

5

Sorry I don’t understand what you mean. From practical experience AMD has always suffered from poor computational performance in applications such as 3DS Max, and the drivers have at best been flaky.

1 Like
6

Dear, I’m waiting next gen cards, I decided to skip 30x0 and 6x00.
I’ll buy the best card we’ll have, red or green is the same for me.

7

Wasn’t the 6900xt supposed to compete with the 3090? I feel like it was kind of a fail that they are just holding up with the 3080, and for hundreds of dollars more on the MSRP than the 3080.

Meanwhile, the 3000 series Nvidia drivers have been problematic (so it’s not just AMD), and the prices of Nvidia GPUs just went up. In some cases massively. The 3090 is now going for a whopping $1900.

1 Like
8

Yes, but if you choose AMD, you need to convince yourself in some ways.

9

With Vulkan in Xplane, AMD holds its own against Nvidia comfortably. It may well be the same with DX12. If it does, the buying decision will be RAM, drivers, reliability under load and prices.

Personally I will wait for dedicated DX12 benchmarks before choosing the best.

1 Like
10

You’re starting to sound like an Nvidia fanboy there.
Seems lately much more people have been having issues with Nvidia drivers rather than AMD ones (especially if you count VR).

1 Like
11

Thanks for this.Is that AMD card compatible with the G2 Reverb?

12

Great post and thanks for doing the hard work comparing the cards. I would argue that this is only relevant for this very simulator, since NVIDIA cards still have things AMD has not yet built in. DLSS is one feature missing from AMDs portfolio. Raytracing is anohter - both do not play a role in msfs 2020 yet. ( And you can argue about the value of raytracing in general I guess…)

Having that said, I recently invested in VR. I do have a AMD card and can confirm that the AMD drivers for this are less than optimal. I get constant “Graphic driver timeouts” which always lead to a CTD.

System:

  • AMD Ryzen 7 3800X
  • 32 GB RAM
  • 1 TB NVME Flash drive for Simulator
  • Sapphire RX 5700 XT

First I tried the Oculus Quest 2 and must say - hands off if you have an AMD card (no matter if 5x or 6x). The Oculus Link “Technology” is only poorly running on AMD cards. The reason for that is that your card needs to do 2 things in paralell:

  • Render the game in VR resolution (which is high for the Quest 2 - good display!)
  • Encode the rendered came content into a video stream to send to the Oculus Quest 2 HMD

The Encoder in AMD is very poor. It performce not nearly as good as the NVIDIA cards. Reason for that is simply that NVIDIA has its own business branch focussing on enterprise graphics (Quadro) and they invested a lot in hardware encoders. The 3x series (and even the 2x series) have this tech built in.

Now this is only relevent if you are usng a Quest 2 - but I can tell you - It is no fun with an AMD card (drivers and encoder performance).

I managed to get my hands on a HP Reverb G2 (which also is tricky to run on any AMD card - but it runs! - Yay!). This one runs sooo much smoother even though the headset requires even more GPU power (higher res). But I can tell you it runs at least twice as good as Oculus link. Now, I only get driver timeouts once a week or so… :confused:

Now I would wonder how these two cards (the 6900 and the 3080) compare in the VR space - since it seems you cannot simply take results from “non VR” benchmarsk and transfer them.

I am mediocre happy with my AMD card, that is why I actually plan to switch to a 3080 once the prices become reasonable again (summer ;D?)…

Just my experience…

5 Likes
13

Yes and that about sums it all up perfick!

14

Thanks for posting this. So many opinions on the internet but good objective data is hard to find. If you said the 6900 XT performs better but I’m going to sell it instead I’d be suspicious, haha.

So um, how much do you want for the 3080? :grin:

15

Hmmm. Running at 4K ultra using sim dev tools I never see VRAM usage above 9 GBs on the 3080, even in places like Tokyo and New York. No stutter in GPU bound scenarios either (when limited by mainthread, there is obviously some stutter)

This might be due to driver overhead. What’s weird is that the 3080 shouldn’t encounter any performance loss until it uses more than 10 GBs, so there’s likely an external factor involved.

I’d still take the 3080 over the 6800 XT though, since the 3080 does outperform the 6800 XT in most benchmarks.

6900 XT is more expensive by about $300, so I can’t say I’m surprised it performs better lol

My “ranking” would be 3090 > 6900 XT > 3080 > 6800 XT (slightly better for non raytraced games at 4K) > 6800 > 3070

If MSFS gets raytracing though, which will hopefully happen (it would improve a few aspects of the sim imo), then this will all change

16

According to what I’ve seen online, the 3000 series cards perform better in VR in most games. I haven’t seen MSFS tested however. I’ve tested the 3080 with VR using the Quest (original) and it’s flawless with little to no stutter (all stutters are caused by the 3700X) but I have nothing to compare it to. In MSFS though, get the cheaper card and a better CPU imo, CPU seems to be the limiting factor more.

1 Like
17

?

I haven’t had any issues, stability has been perfect, haven’t had a single crash or issue

1 Like
18

many people flying in VR have reported very low framerates with recent drivers. Search around, plenty of reports. That you don’t have issues, doesn’t mean that others don’t either (goes for both NVidia and AMD).

19

In MSFS though, get the cheaper card and a better CPU imo, CPU seems to be the limiting factor more.

I can tell you that in my case it is 100% GPU Bound in VR. You can check in developer mode with “Show FPS” enabled - tells me right away that my GPU is maxed out… VR is just a massive challange for any GPU.
I mean, 4k is nice and all - but in VR your card needs to render two different imiages for stereoscopic 3d and than your VR Headset comes along and says: Better make this 2x 2160 x 2160.
Your CPU just goes: Up to you Buddy and raises eybrowes at your GPU.

But dont get me wrong - when I run Half Life Alyx on this GPU - I can tune it to the second highest detail setting and it runs great. So I absolutely agree that it also depnds on the VR implementation in the title.
(Keep in mind, in a Game in VR the designer can choose to only expose you to a limited surrounding (in a building, in an alley, etc…), while in MSFS 2020 you always see the world oustide of the aircraft - not really compareable. So I do not meant to say they implemented VR in a bad way. All I want to say is - VR in MSFS is awesome, but very very hardware hungry in oder to work nicely.

1 Like
20

Nah, that is W10 memory management for you. W10 will always use as much memory as it can find and will precache more textures or not unload textures as quickly when more VRAM is available.

W10 has a very dynamic memory management system.

2 Likes