Having already proven the benefit of upgrading from a 5800X to 5800X3D back in 2022 (see this post) I have recently upgraded from the 5800X3D to a 9800X3D and thought I would show the benchmarks here, for both MSFS 2020 and MSFS 2024.
Testing conditions were chosen to put the maximum strain on the CPU, which I chose to use the Fenx A320 on the ground at EGLL with traffic, and a second scenario flying low over the photogrammetry of Tokyo.
TL:DR The upgrade to a 9800X3D showed performance improvements for both MSFS 2020 and 2024 when CPU-bound, even at 4k resolution. In the most demanding CPU scenario the 9800X3D is enough to bottleneck the 3090 in MSFS2020.
First, a few caveats. Moving from AM4 to AM5 platform, the RAM has also been upgraded in capacity and speed, so it isn’t a straight comparison of CPUs, but this is likely to be a common upgrade for people moving up to AM5. The comparison between MSFS 2020 and 2024 isn’t like-for-like, as in the Heathrow test I use the iniBuilds EGLL in 2020 but the Asobo Premium Deluxe version in 2024 (iniBuilds version not released yet for 2024). Another difference between the 2 sims is the use of FSTraffic from JustFlight in 2020 and the standard traffic in 2024, again due to FSTraffic not being available yet in the new sim. It isn’t entirely fair to compare the performance between MSFS2020 and 2024 for these differences, especially since the new sim has different scenery (grass, bushes, seasons, etc.), but it is an interesting comparison to make.
Test Conditions:
FE 3090 24GB
Upgrade 32GB 3600MHz (2x16GB) → 64GB 6000MHz (2x32GB)
M.2 SSD 2TB
Win11, HAGS On, Game Mode Off
MSFS 2020 DX12 / MSFS 2024, Ultra Preset, TAA + 100% Render Scale 3840 x 2160, Terrain LOD 200, No FG mods (although I do normally use these but not for this benchmark), 500Mbps internet
Scenario 1:
Fenix A320 taxi @30 knots East to West along one of the taxiways at London Heathrow, cockpit view, scattered clouds, traffic on 100%. I also chose to set the time to an early winter morning where the shadows are long. This is probably one of the most demanding CPU usage cases in both sims, and where any performance improvement would be best felt.
Looking at MSFS 2020, this is much the same story as the upgrade from the 5800X; there is a measurable (and noticeable) increase in smoothness. Average FPS and 0.2% lows are boosted (16% increase in FPS). Frametime variance is improved, as is evident in less variation in the graphs. The dev mode showed I was Main Thread limited with the 5800X3D, but GPU limited with the 9800X3D. It now seems the 9800X3D is enough to put the bottleneck on the 3090 even in the most CPU-demanding scenario, which would imply that any future CPU upgrades wouldn’t show an improvement without a GPU upgrade.
Looking at MSFS2024, there is still a small improvement of 12% for average framerate, as well as an improvement in 0.2% lows, but generally MSFS2024 performs much worse and there are a lot more spikes in frame time, presumably when assets or scenery are loaded in. Dev mode showed CPU limited for both CPUs.
Scenario 2
Beechcraft Bonanza starting on runway at Tokyo Haneda, take off and fly at 1000 feet North over Tokyo, scattered clouds.
This test starts to shift onto being GPU bound, as the Dev Mode shows, but is a better representation of the FPS when actually flying. In both 2020 and 2024 there is very little difference between the 2 processors. Interesting that in 2024 the older 5800X3D has better numbers, but I think this is within the repeatability of the testing. Much as my previous analysis of the 5800X3D, this scenario is GPU bound at 4k and doesn’t show any improvement for the newer CPU.
A few other observations, at no point did I ever saturate the 24GB of VRAM on the 3090. However, in 2024 the 32GB of system RAM was saturated and the PC started using the page file. This was only in 2024, and obviously this did not happen when switching to 64GB RAM.
Generally pleased with this upgrade, and hopefully I can pair it with a new 5090 when they become available, to lift the GPU bottleneck in MSFS 2020. I generally use the FG mod, which coupled with a properly configured G-SYNC set up results in very nice performance.
For those interested, these are the initial screen captures for each test with dev mode turned on. Mouse over the images to get the title for each test.
*Edited to correct 1 measurement
) I’ll sum up with these few charts:
