A bunch of FS24 test results

[BegottenPoet228]

I spent most of the day running CapFrameX tests. My goal was to look at a few variables in my default test aircraft, the Bonanza G36. I haven’t figured out how to get access to my G36 Improvement Project, so I flew the Asobo version. My existing SPAD profile worked fine for the purpose of the test.

Gigabyte X670E AORUS Pro X, Ryzen 7950X3D (per core undervolts) RTX 3090 Ti 24 GB, 64 GB DDR5-6400/CL32. nVidia driver 566.14

The only overclocking was mildly tightened tertiary RAM timings, and FCLK= 2133. MCLK:UCLK ratio set to 1:1. I have a 1 Gb fiber router connection to a WiFi 6 chipset on the computer.

Notes:

• All tests done @ 4K, with HDR on, Windows Game Mode OFF, Xbox GameBar disabled, RBAR enabled, HPT disabled, ISLC running, Process Lasso CPU affinities set, flightsimulator2024.exe set to High Priority, and Bitsum Core Park set to disable core parking. Ultimate Power Plan.

• I detailed the long process I used to decide on these settings in this thread: My journey from AM4 to AM5

• Overall experience was very positive. The sim loaded quickly every time I launched it (faster than FS20, for sure) and the only stutter I experienced was flying over NYC, using the Ultra preset and using TAA, with TLOD=400 and OLOD=200, and it was just one brief stutter. Once I dialed TLOD back to 200 it was always very smooth. I did not experience any ‘object popping’ at that setting. I mostly fly GA at lower altitudes, so I don’t need high TLOD anyway.

• I definitely need to get ReShade up and running again. I experienced the ‘washed out’ look others have reported.

• Maximum RAM usage was 34 GB. Normally around 25-28 GB.

• Maximum VRAM usage was 11 GB, but it was mostly between 6-8 GB.

• I started each flight airborne, and did not do extensive cockpit testing.

• Rolling Cache was set to 100 GB, running on my fastest M.2 drive. The custom location never changed, and the file size remained at 100 GB.

I’m dumping a lot of test screenshots here. You can draw your own conclusions, but for my system I think I’ve settled on DLSS Quality, Latency ON, and assigning the sim to the Vcores only (although as you’ll see, it didn’t seem to matter a whole lot if it was assigned to Vcore, Non-Vcore, or even All-cores.) This tells me I’m GPU limited. Cockpit instruments were a little blurry using DLAA, but I couldn’t tell DLSS Quality and TAA apart. And DLSS Quality was the clear performance winner.

Throughout the tests I was seeing the rapid switching between Main Thread and GPU limited, which is what I want to see.

I’m quite happy with the performance over NYC, although I know that performance will get worse with more complex aircraft. I was really pushing my system hard, and I feel like I’ve got a lot of headroom if I need to decrease some settings.

Temps throughout were excellent. Both the air-cooled CPU and liquid-cooled GPU ran 60-70°C throughout.

Here we go…

First, the Vcore affinity tests. I’ll start with the Sahara, then NYC. I tested DLSS DLAA, DLSS Quality, and TAA, along with Latency ON, OFF, and ON+Boost. Latency settings didn’t have much impact, so I won’t include that variable here.

Vcores - DLSS DLAA - Sahara - 55 FPS

Vcores - DLSS Quality - Sahara - 81 FPS

Vcores - TAA - Sahara - 57 FPS

Vcores - DLSS DLAA - NYC - 47 FPS

Vcores - DLSS Quality - NYC - 62 FPS

Vcores - TAA - NYC - 48 FPS

Here’s what the metrics look like on the Vcores when the sim is loading:

And here they are in flight:

The next tests were with the sim assigned to the Non-Vcores only. I only did the Sahara test. I can add NYC if interested. I simply didn’t see a significant difference between the core affinities.

Non-Vcores - DLSS DLAA - Sahara - 56 FPS

Non-Vcores - DLSS Quality - Sahara - 83 FPS

Non-Vcores - TAA - Sahara - 57 FPS

Now with the sim assigned to ALL Cores.

All Cores - DLSS DLAA - Sahara - 56 FPS

All Cores - DLSS Quality - Sahara - 83 FPS

I did not do TAA. I’m not going to use it.
At this point I didn’t bother doing the All Core test over NYC.
I reverted back to Vcores only, since this allows me to keep all other process (except for default Windows processes, which should not be touched) on the Non-Vcores.

Here’s the difference between TLOD=400 and TLOD=200

Vcores - DLSS Quality - NYC - Ultra Preset - TLOD=400 - OLOD=200 - 45 FPS

Vcores - DLSS Quality - NYC - Ultra Preset - TLOD=200 - OLOD=200 - 57 FPS

And finally, a night flight over Greenland with the settings I’ve settled on (for now.) 74 FPS

I’m going to start looking deeper into how well the sim is working, i.e.

[blueline308]

Please post a TL;DR

[BegottenPoet228]

No can do…
Scan the captions above each screenshot.
Recognize that my settings work for my system. If you can glean any helpful info from the work I’ve done, great. But you’ll have to dig in…

[FlyingsCool5650]

I think mostly we’re all wondering, 2020 vs 2024, what’s your overall opinion of the difference in performance on your machine.

Good / bad / indifferent?

[Baracus250]

@FlyingsCool5650

indifferent

I’d say “different”. I’m definitely hard GPU Limited now where in 2020 it was pretty much always CPU (7950X3D and 4080 on Quest 3). So it tells me they have indeed done something to the core engine and it’s using the CPU much better now.

But I am having a lot of inconsistencies in frame rate. It’s all fine but then suddenly just drops really badly.

I have not had much tinkering time to try and solve it but for sure I had to massively reduce my resolution to not hit 100% GPU Usage (not VRAM, that was always within limits but just graphics processing power I think).

I’ve narrowed it down to a combo effect of having lots of trees plus photogrammetry. That totally kills me.

There could be other factors. Traffic maybe (not tried turning the Realtime Aircraft off yet).

Could also be some airports I bought from the store but have not been optimised yet. For example I flew into Bilbao (Best Model of the World one) and it totally crippled my frames.

Strange thing is that was repeatable to a degree but after a fresh boot and fly past it was FINE the first time, then I flew a circuit and back past again on the exact same path and it was back to bad.

So maybe it had not loaded in traffic on the first pass. As always there is a lot of factors and it’s hard to isolate and may just be a compound effect.

EDIT: sorry, I tell a lie here maybe. I bought that airport from iniStore, so unless it found its way into my 2024 sim somehow — I’ve not added any mods yet — then this is a new bespoke airport maybe?! It looks like BMW’s! I’ll have to check. Maybe I need to disable one and add the other. But we can’t disable items in My Library yet?! Bit of a mess!!

Overall though, when it runs smooth it IS a lot smoother than 2020, but I had to sacrifice 30% of my resolution to achieve it.

The true 3D trees are nice but I definitely found that making them Low detail solved all the stuttering in tree-heavy areas, but that makes them back into the Sprite-based ones from 2020. So conclusive evidence that they are HEAVY.

On Medium it seems to be using 3D trunks and the Sprite-based leaves except for the really close trees to you, so looks a bit weird as you fly past closer ones.

On High they are still using Sprites for the middle distance ones and beyond. But that is fine as you don’t really notice the effect at that distance.

I guess Ultra pushes that distance further. But I dare not go there!

Was an interesting discovery to see that they are still leaning on the previous tech to fill in the hundreds of thousands of trees often on display in dense forest areas though. Sensible decision I think!

Anyway I have much more tinkering to do before I settle on something. Just when I think it’s good, a situation arises that throws me totally. But yeah it could just be other bugs with specific scenery (Rome seems pretty broken for example!) so it’s REALLY hard to know.

[BegottenPoet228]

I’ve hardly flown the new sim. Problems with the launch, demands of work, and mostly some hardware problems I had to get sorted kept me grounded.

Yesterday was the first time I actually spent a full 8 hours flying, but it was repetitive flights over the exact same two areas. I plan to do a little exploring today, adding in some takeoffs and landings.

• Overall it feels like the sim loads faster.
• At first glance I saw more ‘popping in’ of objects in FS20. Hardly seeing that in FS24.
• As I said in my original post, I’m definitely noticing the washed out colors, low contrast, and haziness that others have seen from inside the cockpit. Gotta fix that.

TBH, I haven’t done a true comparison of the graphics settings yet I was using in FS20 vs. what I’m using in FS24. I think they’re pretty much the same.

Here are CapFrameX results of the same flight in the same plane in FS20 and FS24. Statistically FS20 (1% low 45) is a bit better, but FS24 (1% low 41) ‘feels’ a bit smoother. I think FS24 exposes GPU limitations.

Bonanza FS20 - NYC

Bonanza FS24 - NYC

[Airbumps2501]

@BegottenPoet228

No can do…
Scan the captions above each screenshot.
Recognize that my settings work for my system. If you can glean any helpful info from the work I’ve done, great. But you’ll have to dig in…

Unless I’m mistaken it appears to show that the V-Cache cores are making very little difference to performance. Makes me wonder if higher frequency and more cores is better than V-Cache

[BegottenPoet228]

@Airbumps2501

Unless I’m mistaken it appears to show that the V-Cache cores are making very little difference to performance. Makes me wonder if higher frequency and more cores is better than V-Cache

I’m puzzled by that as well. I see no significant difference when I assign the sim .exe (and nothing else) to the vCache, non-vCache, or even All-cores.
If core frequency was king, I’d expect to see a significant bump running on the non-vCache cores.

I wonder whether my GPU is limiting things, and not letting the CPU (especially the vCache cores) work to their potential. Or maybe the code is simply not optimized to take advantage of the L3 cache like it seemed to be in FS20. Or it could be other settings.

I may try running the sim at 1080p, just to remove the GPU bottleneck and see how the CPU responds. It’s easy enough with Process Lasso to set the core assignments.