ACC AI cpu usage?

[theguy69]

I was curious about the AI CPU usage in ACC. They are pretty good when you are side by side but apart from that they don't really do much but they seem to have a lot of cpu usage. I played gtr2 a few days ago and I could easily run 60+ AI. Is there a reason for the AI having such high CPU usage?

 

[SwannyUK]

I think it's because they are affected by the same physics system we are so more calculations going on.

 

[theguy69]

I think it's because they are affected by the same physics system we are so more calculations going on.

Ahhhh that explains a lot. Well I hope they optimize it a bit

 

[RasmusP]

I wouldn't bet on them optimizing it further after 2? years out there.
They couldn't/didn't with AC and ACC is already a lot better (basically uses one cpu core more).
But yeah, AI uses the same physics, which makes them a lot more believable, in theory.
Most other sims use some simplified physics for the AI, which is better for performance but can make the AI seem unrealistic sometimes.

 

[theguy69]

I wouldn't bet on them optimizing it further after 2? years out there.
They couldn't/didn't with AC and ACC is already a lot better (basically uses one cpu core more).
But yeah, AI uses the same physics, which makes them a lot more believable, in theory.
Most other sims use some simplified physics for the AI, which is better for performance but can make the AI seem unrealistic sometimes.

Ohhh mann. Time for CPU upgrade I suppose

 

[RasmusP]

Ohhh mann. Time for CPU upgrade I suppose

What do you have right now?
Thing is with AC and ACC.. As soon as you throw more than 10 AI cars in the mix, no cpu on the planet can run "high" fps.
Vr headsets often have native 90 hz/fps and with 144 hz monitors becoming so available, most people now want higher fps.

My overclocked 10600k at all cores at 4.9 GHz drops below 80 fps at race starts, no matter what I do..

Problem is that ac only uses 2-3 cores and ACC uses 2-4 cores. Simplified.
So with my 6 cores + 6 virtual cpu threads, Taskmanager shows 12 "cores".
Only 3 can be used.. Meaning I have more than half of my cpu sitting around, doing nothing...

If you want to upgrade, get a b550 board + amd 5600x.
Best single core performance with a good price/performance ratio.
More cores won't give you anything. The more expensive cpus are only better for simracing because their single cores can clock higher due to higher manufacturing quality.

 

[theguy69]

What do you have right now?
Thing is with AC and ACC.. As soon as you throw more than 10 AI cars in the mix, no cpu on the planet can run "high" fps.
Vr headsets often have native 90 hz/fps and with 144 hz monitors becoming so available, most people now want higher fps.

My overclocked 10600k at all cores at 4.9 GHz drops below 80 fps at race starts, no matter what I do..

Problem is that ac only uses 2-3 cores and ACC uses 2-4 cores. Simplified.
So with my 6 cores + 6 virtual cpu threads, Taskmanager shows 12 "cores".
Only 3 can be used.. Meaning I have more than half of my cpu sitting around, doing nothing...

If you want to upgrade, get a b550 board + amd 5600x.
Best single core performance with a good price/performance ratio.
More cores won't give you anything. The more expensive cpus are only better for simracing because their single cores can clock higher due to higher manufacturing quality.

Thanks mate! This is really good advice. I needed it. I mostly race offline. I was under the impression I needed more cores but what you say does make sense

 

[RasmusP]

Thanks mate! This is really good advice. I needed it. I mostly race offline. I was under the impression I needed more cores but what you say does make sense

One little note though: in Taskmanager it will often look like all your cores would be utilized.
Don't let that irritate you!

That's because windows often shuffles the load from core to core and with 4 GHz = 4.000.000.000 cpu cycles per second, you will see averaged loads across all cores.

But if you would look into one single cpu cycle, only 2-4 cores will actually show the loads!

 

[theguy69]

One little note though: in Taskmanager it will often look like all your cores would be utilized.
Don't let that irritate you!

That's because windows often shuffles the load from core to core and with 4 GHz = 4.000.000.000 cpu cycles per second, you will see averaged loads across all cores.

But if you would look into one single cpu cycle, only 2-4 cores will actually show the loads!

How do you monitor this behaviour?

 

[RasmusP]

How do you monitor this behaviour?

You can read out the threads of applications via process explorer. Just google for it. Little tool that doesn't require an installation.
Here is how Cinebench R15 set to 3 rendering threads looks like in Task Manager (refresh rate "normal"):

Looks like ALL CPU cores would do "something", right?

Well here is how it looks in process explorer (right click on cinebench.exe -> properties -> threads):

You can see that only 3 threads really cause CPU load. There's one tiny thread and 4 additional threads that can be ignored regarding CPU load.

Now the math behind this is:
100% divided by 12 CPU threads = 8.33% maximum overall CPU load per CPU thread.

So if one of these application threads in ProcessExplorer goes close to 8.33%, it's hitting the maximum. All these 3 are "close".
You can't split single application threads so at one point in time, only one CPU thread can calculate that application thread. (physical cores, virtual cores from hyper threading).

A CPU core can only "do something" or "do nothing". 50% "load" actually means 50% of the time calculating stuff and 50% of the time doing nothing.

For some reason, that goes beyond my knowledge, it's more efficient to let one cpu thread calculate an application thread and then let the next cpu thread calculate the application thread.
Probably because the other CPU threads can start "caching" or something.. No idea really.

But all this leads to all CPU threads being used more or less.
On AMD Ryzen CPUs, you will see a different picture in Task Manager. On AMD Ryzen systems, you would probably see 3 CPU threads being close to 100% and the other CPU threads barely showing any load.

The end result is always the same: Apart from when Task Manager shows FULL load on ALL CPU threads, you can't see if your CPU or your GPU is the FPS bottleneck.
Worst case would be an application only running on a single thread and you having a 20 thread CPU.

100% divided by 20 = 5% max overall load for one CPU thread.

So while you would only have 5% overall CPU load, your CPU would still be the bottleneck because you simply can't use more than a single thread.


To know what part of your system is the bottleneck, we luckily can just look at the graphics card load! When it's above 90%, then the graphics card is the bottleneck. In theory it would be 99% or 100%, but modern graphics card have so many different sections like Ray Tracing cores, Video rendering cores, DLSS Tensor Cores...
So some games/applications can't use more than 90% of your GPU. But I've never seen a graphics card limit lower than 90% so that's the value for now.

 

[Pulp Quango]

I think it's because they are affected by the same physics system we are so more calculations going on.

In addition to that, I have a creeping suspicion that Windows 10 doesn't release memory resources 100% reliably, so should you encounter 1st lap stupidity and restart, sometimes, the AI behave even more stupidly when you take the restart... Quitting ACC and restarting it, when the AI misbehave first time around.... yeah.... It tends to misbehave again. Restart PC and suddenly they behave.... running 16Gig of RAM....

Very rare that the AI behave as badly when I reboot and go straight into ACC....

 

[theguy69]

In addition to that, I have a creeping suspicion that Windows 10 doesn't release memory resources 100% reliably, so should you encounter 1st lap stupidity and restart, sometimes, the AI behave even more stupidly when you take the restart... Quitting ACC and restarting it, when the AI misbehave first time around.... yeah.... It tends to misbehave again. Restart PC and suddenly they behave.... running 16Gig of RAM....

Very rare that the AI behave as badly when I reboot and go straight into ACC....

To be honest I don't think that's a memory thing. I think that's just an AI thing. But honestly its really funny. The game gives each AI the same physics as the player and the physics are really complex but then AI anyway do AI stuff . All of that processing power wasted.

 

[maelstrom]

I can see how people would want higher FPS when using VR. However, if you’re just playing on a monitor, why not just aim for smooth graphics/game play and forget about CPU usage and FPS? I may just be lucky but my relatively old system (stock 6700k with an AIO cooler, GTX 1080, 16GB RAM, Windows 10) runs ACC smoothly at 1440p with mainly epic/high settings, even at the start of a race with 20+ AI. I have no idea what FPS I’m getting but the fact that it’s smooth and the CPU stays relatively cool is good enough for me.

 

[RasmusP]

I can see how people would want higher FPS when using VR. However, if you’re just playing on a monitor, why not just aim for smooth graphics/game play and forget about CPU usage and FPS? I may just be lucky but my relatively old system (stock 6700k with an AIO cooler, GTX 1080, 16GB RAM, Windows 10) runs ACC smoothly at 1440p with mainly epic/high settings, even at the start of a race with 20+ AI. I have no idea what FPS I’m getting but the fact that it’s smooth and the CPU stays relatively cool is good enough for me.

"Smooth" depends a lot to the person sitting in front of the monitor. There's sensitivity of that person and conditioning.
And of course there's gsync/freesync with vsync vs no sync. Without any syncing, everything stutters for me below about 180 fps. Then the tearing borders and the unsteady frame pacing become so small, that I can't really notice it anymore.
With vsync, any fps drop will instantly result in a visible stutter, since the current frame will be displayed a second time, doubling the frame pacing for a short moment.

With gsync or freesync, things become a lot smoother if you can't maintain extremely high or fps above the vsync value 100% of the time.

For me:
I can clearly see the difference between 50 fps to 60 fps and then 80+ fps (then it stops for gaming. Mouse cursor smoothness improves noticeably until 100 Hz, then not anymore).
I tested this with my own 100 Hz gsync monitor, limiting at various fps values and the 240 Hz monitor of a friend

So I try to keep the fps above 80 fps to have this nice, constantly fluid experience but I have to say that my overclocked last gen i5 10600k can't achieve this at all times.
If I increase the settings beyond high, the fps will drop into the 70's for short moments.

However ACC can't use all cores so although hitting the CPU bottleneck, it stays cool, like you say. I get about the same temperatures when doing a 3 thread cinebench run.

I agree though that a lot of us (me definitely) should try to enjoy the fun side of things more instead of becoming kinda obsessed with numbers

 

[maelstrom]

And of course there's gsync/freesync with vsync vs no sync. Without any syncing, everything stutters for me below about 180 fps. Then the tearing borders and the unsteady frame pacing become so small, that I can't really notice it anymore.

Oddly, even though for most games I usually have to have vsync on, for some reason with ACC I don’t. I’m normally quite sensitive to stuttering and tearing but, fortunately, don’t seem to get either with ACC on my system. My experience talking with people who have much more recent, more powerful graphics cards than me is that they often seem to have more graphical/performance issues than I do. I can only really think that it’s down to my particular combination of hardware, drivers and system setup but I certainly seem to have hit some sort of sweet spot.

 

[zzjorzz]

Thank you for this very clear explaination,

RasmusP!​