MSFS RTX 3080 vs. 3090 Benchmarks

Hello everyone,
As from another thread, I mentioned that I would have a brief window in which I had both a 3080 and 3090. They are both MSI Suprim X versions, so I thought I would compare their performance in various MSFS scenarios.
Of course, as always, take all these numbers with a pinch of salt. I’m not claiming these are objectively true, but rather the results I had found. There are most definitely variances in the numbers just because of some third party factor(i.e. MSFS stutters on something that it didn’t do in another run).

TL;DR - Provided you’re not being limited by mainthread, the RTX 3090 offers around 10 - 15% greater performance over the 3080. At the best of times, this will amount to 6 - 8 more FPS. However, even with a 5800X, CPU remains the dominant limiting factor that will prevent better performance, particularly with VR. VRAM usage in my testing was not wildly different between the card, and never exceeded the 3080. Nonetheless, it is certainly possible this can happen once MSFS moves to DX12. The 3080 is a vastly more attractive option for its value, and performs extremely well. It remains to be seen how much of a difference there will be once DX12 happens. Unsurprisingly, the 3080Ti would be the best choice once released.

This is my rig:

ASUS Prime X570 Pro motherboard
Ryzen 7 5800X OC’ed with CTR 2.1 at 4.725Ghz
Coolermaster ML240R
G.Skill Trident Neo 4*8 GB 3800Mhz CL16 DDR4 RAM
Corsair 750W PSU
Corsair 680X Crystal Series Case
Sabrent Rocket Q4 2TB NVMe SSD

I have tested the two cards in two separate conditions. That is, both of them were run at stock, and overclock. For stock settings, I did also open MSI Afterburner and pull the power limit to its maximum to make sure they get the wattage they need to stay at higher clocks. For overclock settings, they both had the same core and memory clock increases - interestingly both crashed on Unigine Heaven when I put them above +100 core clock. OC settings were +100 core, +250 memory clock.

Scenario 1:
Shoreham airport sitting on runway(freeware scenery from flightsim.to) 20 ultra settings with Sim Skunk Works Fiat G.91:
Scenario 1

Notes: This was in a situation in which both GPUs were allowed to rise up to full 100% usage, and is the most accurate example in performance difference between the two cards.

Scenario 2:
Flying along the coast of Brighton seafront(freeware scenery from flightsim.to) around 800ft 250KIAS, Fiat G.91 Ultra Settings
Scenario 2

Similar to Scenario 1, though there were some flickers showing it was being limited by mainthread in split seconds.

Scenario 3:
JFK Landing Challenge Ultra Settings
Scenario 3

Notes: Scenario was dominantly limited by mainthread.

Scenario 4:
Shoreham Airport Runway 20 Ultra Settings SSW Fiat G.91 Custom(around high to low ultra) Settings, VR Rift S
Scenario 4

Notes: VR uses quite a lot of CPU performance, and MSI Afterburner demonstrated that both cards fell to sub 80% usage.

Scenario 5:
Flying along the coast of Brighton seafront around 800ft 250KIAS, Fiat G.91 Custom(around high to low ultra) Settings, VR Rift S
Scenario 5

Notes: Similar to scenario 4, though I suspect since I was moving rather than static on the runway, the 3090 was less of a bottleneck hence marginally better FPS.

Otherwise, my only other notes are that:
The 3090 only ran about 1 celsius hotter than the 3080(72/73 rather than 71/72)
VRAM usage between the two cards was very similar(this was using the developers FPS tool) and the highest I saw(in these scenarios) was 8.5GB.
As with flight sim in general, the CPU remained the dominant limiting factor, in particular with VR.
I did not have any traffic enabled or multiplayer given that the dominantly affect CPU usage.

If my Reverb G2 arrives before I have sold my 3080, I will test the results… this is also contingent on the Reverb working on my X570 board!

I hope these results are of interest to people, and don’t hesitate to ask any questions if you want more information.

Cheers,
Peter

8 Likes

would be interested to know the FPS results with stock aircraft that Asobo probably tuned for optimum performance rather than systems accuracy but your results also provide some proof that a 3090 is, as far as games are concerned, not worth the money.

Well this helps towards making my decision the 3090 doesn’t look worth double the cost in some cases for around 5 fps gain. Though almost impossible to get hold of anything right now.

@ToastedVengance was this done in full 4k and 100 res in sim?

Render scale was at 100, and this was performed on a 4k monitor, so 3840 x 2160.
While in VR, I change the game to windowed mode and set it at a resolution of 800 x 600 to free as many resources to the headset rather than the mirror. Makes a substantive difference, especially if mirroring on 4k monitor.

2 Likes

I wouldn’t be so sure.
The SSW G.91 is extremely friendly in terms of performance. In my experience glass cockpit aircraft eat up a lot of CPU performance, so a classic steam gauge aircraft is more than fine.
Nonetheless, the stock 747 was used in the landing challenge.

If your only seeing 8.5 GB VRAM usage and the limitations with current CPUs - even a 8GB 3070 may be in striking range of the 3090 with MSFS.

CPU limitations are mainly seen when sat on runways or flying close to the ground. Once if you a couple of thousand feet up then it’s more rare to be CPU limited.

On days with more clouds then the GPU is quite a limit. Often a 3090 is only just holding 60, so then a 3070 is going to be hovering around ~40 FPS. Playable, but not within striking distance.

Thanks for doing these tests. I guess I’ll keep waiting for my 3080 to arrive - won’t spend another $1200AUD to get a 3090 now but with very limited increase in performance - I’ll keep that money towards the future 4080 or whatever it will be next.

I have a 3090-the big question is whether 10GB VRAM will be enough if you also fly P3Dv5 which is DX12 albeit unoptimised (my 3070 and its 8GB were woefully unable to run it with complex addons).
No such concerns with the beast that is the 3090

I share your concerns too. With the 2080Ti I had no qualms about P3d v5 but an 8GB card was never up to the task. Similarly the 3080 10Gb card didn’t feel right. From a P3D perspective I wonder how a 6800 or 6800XT performs?

Until there’s a 3080Ti(which I think will be a while away and heavily demanded), radeon may be the better call for P3D aside from the 3090

It is rumoured for April, but it is worth noting that Nvidia have downgraded their proposed 3080ti to just 12GB instead of the original 20Gb.

Even if it launches, it could very well be a paper launch. Perhaps they dropped VRAM down to 12GB to keep it distinct to the 3090, or because of the VRAM shortage?

Thanks for these Peter. I wondered if your G2 ever turned up and if you noticed a much bigger difference between the cards due to its higher resolution? Thanks a lot !

I did get my G2, but after my 3080 left I’m afraid.
In my view I definitely think the 3090 would contribute a fair amount when using a Reverb G2. I can run comfortably at 80% resolution, though I think you could sneak it to 85 without the card being a bottleneck

1 Like