Profile of Seb

tl;dr no new conclusion (cpu => 1 core only, gpu mostly underused, smoke kills gpu)


More interesting graph with the awesome Process Explorer

Both graphs are in sync, over a spectate of 20min.



The colors on the CPU Usage are for each core. Pretty much only one core is used. The total never went above 25%.

This means my earlier comment about thread scheduling is somewhat false. Windows doesn't artificially make it look like the game is using more cores, it's just that it puts other running applications on the other cores. Or in case of i7, yeah it's actually putting each of the game 25% in 12.5% * 2 hyper-process (of the same core) or whatever the name. In practice the result is the same though and conclusion correct: only single core performance matters for this game at all, and having more than 2 cores is totally irrelevant except for streaming or other stuff next to it.


RAM 1.8 GB constant, the max for a 32 bit app... Yeah this is getting old.


IO at the start (loading) and end (saving replay). Not much to say, good that there is no IO during the game (having to load stuff from the disk would be pretty bad). I guess it's storing the additional data somewhere else in RAM and swapping it at runtime? I'm not really familiar with the mechanic to pre/reload more stuff when the 32bit limit is reached.




The 2 gaps to 0% in the early are when I paused the game (through the app to freeze the process).

GPU load typically 100% at the start and then gets lower. CPU can't get higher at the same time since it's already maxed on 1 core, but that means fps gets lower, and particularly goes below the critical threshold of 60fps, which is really where optimization should be aimed at.

Well, it goes below 60gps for my CPU that is. And that means the most recent such as 6700K (or older 4790K, how is it so good??), whilst not that much better, would maybe allow to keep it above the critical 60fps mark all the time, which would change everything, in regard to fluidity, and more importantly when considering streaming at a true 60fps which is currently impossible for this game otherwise. That said since I'm already at 4.35 GHz so this is not even sure.


What's interesting is the VRAM collapse from 1.9 GB at the start to then averaging 1.0 GB for the rest of the game. wtf?? This doesn't seem to match what GPU-Z reported for VRAM earlier though and will need more testing.


System GPU memory is always like 60MB whatever that changes.


Last note, multiple times the GPU load went back to 100%, it was when there was smoke (and sometimes I zoomed it on it). It killed my fps, fully loaded GPU, and significantly reduced CPU (we can see those better on the CPU graph), GPU being the main limit in this rare case.

Basically that is really bad, because it means I'd have to decrease AA to keep a constant framerate for the rare occurrence of smoke, when it's already not using my GPU the rest of the time. There really should be a separate setting to reduce only the smoke details.


Finally, running threads during the game :





The number of them doesn't matter so long as they depend on each other, or are dedicated to (too) small tasks. Total around 24% CPU at that time.

One last note for today, actually 4k DSR with low AA doesn't even max my GPU either, and back to being CPU limited.

But this pretty much confirms what I read before that COH2 medium AA is equivalent of 2x DSR and high AA is 4x DSR. Not sure which looks best though, if there is even a difference.

But there's no point of running all together on 1080p monitor like my previous test did (except for maxing out GPU load for a test of course). That was like running 16k I suppose...

But probably someone with a 4k monitor who does want AA will need to hit a similar performance threshold as my previous test.

Quoting myself from the other thread with relevant data.

The only reason the CPU load appears even across all cores is because Windows thread scheduling makes it look like so.

You will never see each of them at 100% though. The sum of all of them is what you would get if it used 1 and half (or at best 2 cores) if they were used exclusively.

You can force it by defining a processor affinity to the game in the task manager, you will see only those selected used at 100% like you would expect, and the others empty or used for other apps, but this is probably not useful.

No matter what you do from my experience you get averaging a 25% load on a quad core.

The reason being that this game (or any game in general) has a lot of sequential work to do. It's much better for gaming to get a better single core than any number of multiple cores. This is also why for gaming it's usually advised to get an i5 (compared to i7) in regard to the FPS/price ratio.

There's always more work the game developer could do to multi-thread more stuff (I strongly believe Relic hasn't done enough in COH, partially because the game engine is too old), but it has a limit for any game with a clear timeline, because you have a time-frame for each event (1/8 of a second for CoH if I remember correctly) and (obviously) cannot calculate events in the future with the spare CPU cores, and (obviously) doing it in the past would be useless (as opposed to compiling, video encoding, or other work that can be split in blocs independent of each other).

I had a relatively similar issue for the GPU and I fixed it using Display Driver Uninstaller to remove everything, and then reinstalling the latest driver. It's now always around 100%.

GPU-Z can show those numbers and more (clock, VRAM use, fan, temp, ...)

The CPU will never be maxed because it's impossible for the game engine to use all cores, too much sequential work has to be done. I don't remember the exact number but it should use around 1 and half core max (35% at best on a quad core).

My game seems to be capped at 72 fps to prevent that (I assume 144 / 2), I didn't have to do anything manually.