Powerful system (9800x3d, 5070 Ti), but Overwatch 2 on the lowest settings has suboptimal 0.1% / 1% lows

[netborg]

If the problem happens during gameplay but not during a replay then I think this might point to high polling rate mouse/keyboard causing it since you're not moving them during the replay. You can temporarily lower the polling rate of both down to 1000Hz for testing purposes.

Look at what I've just written. The replay does not reproduce the real thing, it's rendered much more inefficiently.
Or like is there a misunderstanding, what "reproducing the real thing" means? I'm interested in optimizing the gameplay which is the real thing.

https://github.com/netborg-afps/dxvk/releases
Put this into your Proton/Linux install and you'll get the best input lag this game has to offer (roughly 20% less latency compared to what you get on Windows).

[TN_fun]

If your gameplay replay codes do not show problems, then it is a network code or a network stack problem on your side.
But since there are videos on YouTube where 1 and 0.1% are also low, I think it is a network code problem

The replay plays back with completely different GPU usage and thus was not suitable to gather data from for my frame pacer. I need 100% GPU usage with ~500 fps (which then gets converted to 95-99% GPU usage to minimize input lag - or less if the fps limiter is engaged).

I don't quite understand how you achieve 100% GPU load even at 500 fps, is this ultra settings or a low/middle end video card?

[netborg]

I don't quite understand how you achieve 100% GPU load even at 500 fps, is this ultra settings or a low/middle end video card?

Medium settings, 1080p on a RTX 3080.

[TN_fun]

I don't quite understand how you achieve 100% GPU load even at 500 fps, is this ultra settings or a low/middle end video card?

Medium settings, 1080p on a RTX 3080.

How about a all settings low or all low + render 75%? Do such a test. I do not ask you to play all the time like this, but will the problem persist on such settings?

[netborg]

How about a all settings low or all low + render 75%? Do such a test. I do not ask you to play all the time like this, but will the problem persist on such settings?

Maybe I'll have a shot at this. Going to 75% render scale is not necessary for me in OW (it's barely faster, and 1080p is my native display res), but going low settings possibly makes the game use simpler shaders which are faster to compile, I don't know. Actually noteworthy that shader compilations are a bit slower on Linux, cause it first translates the shader to SPIR-V (Vulkan related intermediate language) before the Nvidia driver compiles it to native.

[Maelstrom]

How about a all settings low or all low + render 75%? Do such a test. I do not ask you to play all the time like this, but will the problem persist on such settings?

Maybe I'll have a shot at this. Going to 75% render scale is not necessary for me in OW (it's barely faster, and 1080p is my native display res), but going low settings possibly makes the game use simpler shaders which are faster to compile, I don't know. Actually noteworthy that shader compilations are a bit slower on Linux, cause it first translates the shader to SPIR-V (Vulkan related intermediate language) before the Nvidia driver compiles it to native.

The shader compile is likely a red herring, since all shaders ought to be compiled at first launch of the game, or when you have a new driver.

[Maelstrom]

With uncapped FPS in practice range the 1% lows are around 350. Once I cap the FPS to 350 suddenly the 1% lows are down to 180. And so on as I cap the FPS lower and lower, the 1% lows are scaling downwards as well.

Is this expected behaviour? My initial thinking would've been that the 1% or 0.0001% lows would be relatively constant, unrelated to the framecap.

[0xfffffffadc]

Today is interesting: new Nvidia (hotfixed) drivers, new AMD Chipset drivers, and a new Overwatch major update. Since starting this thread, I've also reinstalled Windows (to an earlier version, 23H2), and made sure to barely change anything on my OS (for testing purposes).

Unfortunately, the game still feels pretty choppy for me... curious if others have significant improvements. Convinced it's my RAM. I will mention it's better than it used to be though!

[kyube]

Today is interesting: new Nvidia (hotfixed) drivers, new AMD Chipset drivers, and a new Overwatch major update. Since starting this thread, I've also reinstalled Windows (to an earlier version, 23H2), and made sure to barely change anything on my OS (for testing purposes).

Unfortunately, the game still feels pretty choppy for me... curious if others have significant improvements. Convinced it's my RAM. I will mention it's better than it used to be though!

You're gonna have a tough time getting smooth gameplay in OW, considering that you (very likely) have unstable RAM & have a 50-series GPU.
50'000 tREFI can only be stabilised if you removed your heatsinks, any RGB and have a RAM fan or some other type of active cooling to stabilise the temperature fluctuations. Then, you need to hope your IMC can take it.
To avoid the this hassle, I would recommend you to disable EXPO and/or any RAM overclock setting entirely, relying on this data: https://kingfaris.co.uk/blog/intel-ram- ... /overwatch
Extrapolating from this, one may conclude that UE games benefit the most from any manual RAM timings and that other titles are very tame in performance increases.
i do not know whether OW's engine benefits from by disabling SMT, you may try doing so.

The 50-series drivers have been atrocious, I don't know what to recommend you other than to hope Nvidia fixes it for Blackwell owners...
I've also seen mentions of MPO breaking with 50-series GPUs, specifically with AM5 X3D CPUs, so that also might contribute to your choppy feel.
Not only that, the ALC4080 may also cause some DPC/ISR overhead since it's USB-based (not to mention that it adds severely to your overall audio latency too)
I haven't seen anything in regards to the 5GbE NIC that your motherboard has, this can also somewhat contribute. I've also seen Realtek networking NICs execute their interrupts on the wdf0100.sys driver (the USB one)

[netborg]

The shader compile is likely a red herring, since all shaders ought to be compiled at first launch of the game, or when you have a new driver.

Not all shaders are compiled at launch. They are compiled during gameplay too!

If you have more information how the shader cache works on Windows, I'd gladly read. I'm sure the cache will be limited though, so if you play some other games in between, the cache will be deleted. The shader compilations during gameplay also means it will take a long time of playing until all shaders are loaded into the cache.