TESTED: Game mode and Hardware Accelerated GPU Scheduling (HAGS)

After trying to research and finding only opinions with little-to-no evidence one way or the other, I decided to test this morning, to try and establish what affect, if any, the Windows feature “Game Mode” and “Hardware Accelerated GPU Scheduling” (HAGS) have on MSFS 2020.

CAVEATS: This is not a rigorous test, I am certainly not professional at this kind of thing. Also, this is obviously only one my single PC I use to play, and don’t have the opportunity to test it on better or worse hardware . I have my settings basically on Ultra across the board, with LODs reduced to about 150. I do still get stutters sometimes just like everyone else, and would love some better performance, but I do typically get stable 40+ FPS while playing, so 1 FPS up or down here is maybe 1-2% difference, and therefore if you’re normally around 30 FPS, you may see literally no difference at all - your mileage may vary!

SYSTEM: AMD 5900x CPU, MSI Suprim X RTX 3090, MSI X570 Tomahawk motherboard, 64GB RAM, 2TB m.2 SSD. I also have attached Alpha yoke, Bravo throttle, TPR Rudder, Next Level Racing v3 motion platform, SRS Wind simulation, 6 Saitek FIPs. This test is being done on a 4k 60Hz 27" Dell monitor, using the latest (at the time) nVidia drivers of 465.89.

PROCEDURE: Load into London Heathrow (EGLL) at Runway 27. Cessna 172 G1000. Set weather to clear skies, set time/date to April 11th, 11am, no mutiplayer traffic, Live AI traffic on. Ready to fly, then wait 1 minute for FPS to settle, and then 1 minute capture with Riva Statistics while idling on the runway. Between tests of Game Mode on/off or HAGS on/off, I would fully restart the system.

I wanted to test the affects the affect of various background tasks. I tested with my usually-loaded background tasks on reboot (which includes a bunch of ‘unnecessary’ things, including both Razer Synapse and Corsair iCue for RGB, a Dell monitor manager software, and some other bits). I would then do the test, close everything non-essential, restart/reload the ‘mission’ (i.e. using Esc, then Home, then Enter), again hitting ready to fly, waiting a minute to settle, then 1 minute bench. Finally I would then start up some extra software I usually have running while playing the sim (e.g. the motion platform software, SPAD.next for the FIPs, and SRS for the wind) and do another test. These are reported in results as BG, No-BG, Sim-BG

Values reported are average FPS, and 1% lows in the format Av/1%.

RESULTS:
Game Mode on + HAGS on:
BG: 50/43
No-BG: 50/42
Sim-BG: 48/41

Game Mode off + HAGS on:
BG: 50/42
No-BG: 49/39
Sim-BG: 48/38

Game Mode on + HAGS off:
BG: 51/46
No-BG: 51/45
Sim-BG: 51/38

Game Mode off + HAGS off:
BG: 52/49
No-BG: 51/48
Sim-BG: 50/40

ANALYSIS:
I must say, I was surprised by these results. Mainly in that the differences were pretty minimal, and until stepping back to really look at it, I wasn’t sure if it even had any beyond run-to-run variance. I will also say that I think MOST people have the attitude “yeah, turn that ■■■■ off”, with seemingly no reasoning behind it, having read about what these features actually do, I was kind of hopeful that they’d actually have a positive impact.

What I found however, is that whilst the impact is pretty insignificant, I will be leaving HAGS off, as I think it’s fair to say that in any scenario I tested, there’s a 1 or maybe 2 FPS increase in performance. Take this with a pinch of salt, since I cannot explain why when I closed background tasks, most of the time performance dropped slightly - this could indicate a problem with my method!

As for Game Mode, I think this possibly made a very small positive impact when HAGS was on, but with HAGS off, made no difference to average FPS, but potentially hurt 1% lows slightly … somehow … for some reason.

CONCLUSION:
It probably makes no difference one way or the other for most people. I’ll be leaving HAGS off, and I’m on the fence about Game Mode, I like the idea that it stops Windows from doing update downloads and other irrelevant stuff in the background while a game is running, but the data seems to suggest it has a tiny negative impact. Perhaps it would help with a longer test where you fly around streaming new data, and perhaps Window has just happened to download an update at the same time. I will probably just leave it off for now and possibly retest at a later date.

Please let me know if you interpret anything different from my results, or if you have done similar tests with contrary/complimentary results. Also, feel free to ask me to test something else - having said that, I’d rather spend a few hours flying than testing, so … you know, don’t ask me to redo the whole thing :smiley:

Hope it helps someone!
Dave

19 Likes

Thanks for the detailed test results. Very appreciated with your sharing.

Did I miss about what is BG ?

Edit: Oh sorry. It was background tasks. I have just found.

Kind regards.

The result does not surprise me: the simulator is not a game!!! :grinning:

Dieter

1 Like

thanks for your work and feedback.

HAGS off gave me about 5fps.

That’s interesting, what sort of CPU/GPU is that on?

9900K/1080

Spawning at Heathrow gate 225, I was getting 30-35fps before, with HAGS on and Nvidia driver 460.89. Switching to 465.89 with HAGS off got me to around 41fps. The driver alone got me a couple of fps if I recall.

yep:

I have it off… it cause more trouble as it is worth.

Only known exception I know is VR users because a nvidia issue:

Ah … now that’s interesting. I do use VR, but hadn’t done this test in VR. I typically had HAGS off, and had recently noticed that VR was stuttery … kinda hence how I started looking into HAGS again to get a definitive answer.

Looks like I shot myself in the foot then, coz where I have stuttering is VR, and I had just assumed that HAGS would applying equally to VR.

Might have to go back and do more testing :-/

Thanks for taking the time to post your results. I don’t want to be a killjoy but the results you found are not conclusive.

  1. HAGS reduces the workload on a high priority CPU thread that submits frames and commands to the GPU
  2. Game mode is supposed to prioritize the game by giving low priority to background tasks. One important thing that game mode does, is disabling Windows Update while you are playing. All this stuff is, again, on the CPU.

At 4K/Ultra with the Cessna 172 you were probably 100% GPU limited, which reduced the benefit of the first feature to slightly reduced input lag.
Moreover, staying idle on the runway is a good way to get a reproducible scenario but when you fly the sim has to load terrain and objects and make the aircraft work, which generates much more workload on the CPU (primarily) and GPU. In other words you don’t have stutters when you are still, and the system is not fully taxed.
HAGS and Game mode are supposed to give relief to the CPU, you may not see a measurable framerate increase but maybe reduced stutters and higher 0.1% minimums in the timeframe of a full flight from taxiing to landing.
Personally I have a different approach: I leave them ON because according to some tests I’ve done, they don’t do harm, so they may do some good in a situation I have not encountered yet, and if the OS developers put them there, there must be a good reason. Or maybe not :slight_smile:

1 Like

I leave them in Default Off, because depends on windows update x ( or nvidia driver ) , these feature let reboot my pc randomly. In newer updates it seems fixed (again) , but as said: at the moment not worth the trouble :slight_smile:

Yes, I agree with basically everything here, and I tried to make clear in my post that I don’t at all think that my test is definitive or conclusive at all … and even if it were, it’d only be so for my specific hardware and background tasks that I use.

I also tend to have more faith in developers (being one myself … or perhaps despite being one myself :rofl:) that it’s there for a reason, rather than trusting hearsay of people just saying “turn that ■■■■ off, it doesn’t work”. Hence why I wanted to try something on it myself, I quietly expected positive results for HAGS and Game Mode.

I did also say in my conclusion, referring to Game Mode that “Perhaps it would help with a longer test where you fly around streaming new data, and perhaps Window has just happened to download an update at the same time”. And hence again, I agree with you that on the face of it … that seems like a spanking good idea! But would basically not be ‘testable’, just sometimes you’d be playing and thinking “what the hell is going on, it’s running awfully”. So yes, I’m kinda still on the fence with Game Mode, because I did seem to see a tiny performance loss, but can’t explain why that would be (except run-to-run variance), and it seems like I should want it on, for the reasons you give.

I will also say that I did some flights last night, and used the same riva stats to record average and 1% lows, something I’d not actually done before. I was surprised at the 1% lows being quite low, and I did have some noticeable stutters - I didn’t however record my results.

Tonight, when kids are in bed, I might take an hour to do a couple of repeated 15 minute flights from takeoff to landing, recording average FPS, 1%, and 0.1% lows, with HAGS on and off. I’ll also do the same in VR, but the numbers might not matter to me as much as a general opinion of “how smooth it feels”.

I’ll post results when I’ve done this, and make clear that even still this won’t be a definitive test, as the area I’m flying in (EGSG to EGMC) is probably in my rolling cache already, and such a short flight might not ever require any additional loading of textures anyway. But I think it’ll be useful anyhow - I certainly would take a few frames off my average FPS, to add them onto my 1% or 0.1% lows!

1 Like

In my tests I left HAGS off as I found it lost me around 5fps. Your results may vary on your GPU and/or driver version.

1 Like

With MSFS it’s very difficult to obtain a reproducible scenario and come to a conclusion when it comes to performance. Too many variables on a custom flight, including Internet streaming, traffic, etc.
Usually I fly the NY JFK landing challenge, which has fixed route and weather and it’s very heavy both for CPU and GPU, the first more stressed on final and the second more at the beginning of the approach.
You can also use CapframeX to record the data and make a comparison. This is an example of one of my NY landing runs with LOD 100 and 200:

2 Likes

What are the two graphs comparing?

the LOD … :slight_smile:


yes… therefore we have in #self-service:wishlist some about a “Benchmark”

2 Likes

An in-game benchmark would a fantastic addition.

I’m wondering if anyone has played around with any of the available flight recorders that seemingly would let you record an entire flight and then play it back. I don’t know what effect if any that would have on FPS, but if it works, that’d be a good way to come up with your own ‘benchmark mode’ that you can just replay, record FPS, and then make graphical tweaks to see the impacts.

I can imagine it wouldn’t actually … ‘work’ that well for precisely reproducing the flight, and I’d imagine there’d be overheads for using the tool itself. But if anyone knows either way, that’d be handy to know.

intention of such a benchmark is, that it is same for all users. We get comparable values, without any sideeffects like short-connection-hicks, player spawn on server, the ATC trinks a coffee ( you know what I mean :slight_smile: ). Of course custom benchmark mode should be possible… but as said: one of these Search results for 'benchmark #self-service:wishlist order:latest' - Microsoft Flight Simulator Forums

With such a tool perf-tests for e.g. HAGS are repeatable and comparable.

And also for the replay-feature exists somewhat in whishlist ( and I think I have seen this onto the devs sheet ).

Whatever the overheads might be, an initial run without making any changes would establish a baseline.

Current replay systems (as far as I know) only record the plane itself. Everything else remains variable around the plane - e.g. traffic, weather etc.

So i think that if a “standard” flight was recorded on clear skies, no traffic of any sort, multiplayer off etc, then you could get reproducable results.

HOWEVER - I’m not sure this is real, because if the sim is not calculating the position of the plane (i.e. the actual in sim aerodynamic calcs) and is instead being fed that by the replay file, I don’t think this would show the true FPS that would be obtained when truly flying the flight plan “live”.

The best answer would be to let the copilot fyy the route, with traffic off, clear skies etc…but you’ll probably end up in a mountain or a never ending journey to the void.