The ACC CPU Benchmark Thread

Thanks for the detailed response.
I only use the pc for sim racing so single core performance is the way to go.

 

Makes sense!
So 9600k would definitely be good enough for your needs.

The other question still stands though..:

Ryzen 4xxx will 99% surely beat the current and next Intel generation in single core performance, while running cooler with lower power consumption and probably be cheaper.
So I would recommend going the AMD route if you don't really need the maximum performance right now.

Only if you want to buy the best performance now and don't want to fiddle with CPU and cooler at the end of the year again, Intel would be better suited for you.

 

Too many times I’ve held off until the ‘next thing’ arrives or bought not-quite-what-I-need now because I’ve been thinking about a future upgrade so now I don’t worry about that stuff. When I get the funds to upgrade I just look at whats the best for my budget now. Saves stress.

 

Totally understand that!
I was so fed up with my i7 2600k dropping below 60 fps in basically all games apart from stuff like..
Fortnite, urgh.

So I simply waited for the 10600k to be available and the prices to stabilize within the first week.
Then bought everything and never looked at prizes again.
Mobo and cpu 20€ cheaper now? Who gives a crap!
Stable 85+ fps in acc now, above 100 in all other sims.

I did regret not saving money and getting the 9600k.
Until I found assassin's creed odyssey to be quite a lot of fun and having a lot better minimum fps with 4.8 GHz + ht instead of 5 GHz without ht (deactivated it for simracing).

 

I would suggest ryzen 5 3600 as a cheapest option. It is has the best performance per dollar and should be completely enough for current gen games including ACC.

 

Indeed.
However if you have a 90 hz vr Headset, the 3600 will struggle to maintain the needed 90 fps at race starts for example.

Simracing is rather "last gen". Or looking at the multithreading of them.. Rather "mid 2000's gen", lol.

 

VR is another beer, and in my opinion currently there is no desktop/consumer grade build that would allow running ACC in VR maxed out especially the higher res vr headsets.
Few days ago a friend of mine came with his VR (rift s) to test it on my machine (ryzen 3950 with 2080 ti both with moderate overclock) and we could not get max details at 80hz and the gpu was the limiting factor. After some compromises and tweaks we could get good picture at 80hz with 80%-90% load on the gpu and 10-15% load on the cpu, but that was on single player hot lap.

 

True.. Although even with lowest details you might have problems keeping the fps above 90 when adding some ai or online opponents.

With your 32 thread cpu and acc using 4 cores, the maximum overall load you can theoretically see is
4/32*100= 12.5%

Due to windows scheduling and optimizations, you might max out at 20% but not much more..

So at 20% overall cpu load you'd hit the cpu limit in theory. Practically, one or two game threads are really pushing to the limit. So the limit for your cpu is probably at around 15%.

Would be interesting what's your average fps are for 20 ai, 15 opponents visible, medium to high settings and then the resolution scale put to minimum to keep the graphics card from being the limit.

My guesstimation would be 85-100 fps, depending on the conditions and how close the cars are.

BTW I get almost the same cpu limited fps when on Monitor or in vr.
Vr adds 1-3% load on a separate thread so barely any impact.
But not dropping below the refresh rate is a lot more important in vr.

 

@Krassmus
We didn't log or analyze detailed data during the test. The values I posted are only from rough observations from msi afterburner sensors. It was more like an evaluation if VR is possible or worthy.
Later, if you like, I can give you more data with my current setup (monitor).

 

@Krassmus

Here the process detail during a multiplayer session. The only thing I understand is that those processes have peaks that send my CPU in crisis, for the rest of the explanations (solutions) I trust in your help!

 

Hi,
sadly I don't really got my head around these perfmon graphs yet. It seems like we could see a lot in there but somehow what I see doesn't really make sense for me.

Did you monitor your frame times too so we could see if there actually was a spike?

About optimizing your cpu for acc:
Do you know about the ccx communication latencies and all that stuff?

It might be worth a try to assign acc only to the first 4 cores of your cpu and only the real cores.
That would mean setting the affinity in Taskmanager to core:
0,2,4,6 as due to smt, 1,3,5,7 will be virtual threads.
(you can also swap these numbers. Afaik they're both "virtual" so it doesn't matter which half of the core you put it on.)

If you assign it to 0,1,2,3 it means only using core 2 cores.

And if you wonder, like I did and tested this by looking at the temperatures of the cores when setting the affinity to different cores during a benchmark:
Taskmanager shows your cpu threads like this:
"core 0+1" = 1st core
"core 2+3" = 2nd core

So selecting every second core in Taskmanager will result in using all cpu cores but basically disabling hyperthreading/smt.


So you could try if the performance for acc gets better when you assign it to only 4 real cores. If the fps are bad, select every second "core" in Taskmanager to use another ccx of your cpu.

I've read quite some reports about improved fps when also putting everything else in the background to the other ccx to keep them free for the game.

But this is a lot of work and you'd need something like process lasso to save the configuration and automatically do all the affinity setting.

 

Thx for your replay but sadly I haven't the skills to understood what you said :-(
If you need to go crazy this way for a game to work properly I think the developers should ask themselves some questions.

 

Nothing to do with the game, it's how Ryzen CPUs work (and Intel too to some extend).

But good to know what you know and what you don't know so I can adjust my wording (hopefully, a bit difficult sometimes in English hehe).

So inside your CPU you have 16 cores, SMT doubles the amount of threads so from 16 "real cores", you go to 32 "virtual threads".
Then you have packs of 4 core inside your cpu.
So for your cpu, there are 4 packs. These packs are called "CCX" (Core Complex).

Now inside your cpu you have connections between the 4 CCXs and this connection causes some latency.

Now if you go into Taskmanager, you see 32 cpu threads. But Taskmanager calls everything "cores" so here we can get mixed up with the wordings...

Anyway, in Taskmanager it's like this:
Core 0 = real core 1
Core 1 = real core 1
These are the "virtual cpu threads" of the 1st core in your CPU, on CCX 1

Then Core 2 = real core 2
Core 3 = real core 2

Core 8 = real core 5

The real 5th core inside your CPU however lays inside the 2nd CCX.

Now we come to the issue:

If you have a 4 thread game like ACC and Windows somehow decides to put the CPU load on (Tasmanager cores) Core 0,1,2,3 ; You'll have the load only on the first 2 real cores.
But you'd gain a lot of performance if Windows would put them on Core 0,2,4,6 = real cores 1,2,3,4 with one of the virtual threads of each core being idle.

This first issue however is the same for amd and Intel. Luckily Windows is smarter than this most of the times, but in ACC (and that might be down to the developers), you gain a few fps when locking ACC to Taskamanger Cores 0,2,4,6,8 so it can not put 2 game threads on 1 real core (onto the 2 virtual threads).


The second issue is the latency between the CCX packs inside your AMD cpu. If Windows is so dumb to put the load on (real cores) 1, 5, 9, 13 ; You'll end up with 1 game thread on each CCX and the latency will cause fps spikes.

The developers never assign the game threads to any cpu cores. They write their code as optimized as possible and Windows shuffles everything across the CPU in the hopefully most efficient way.
But there are reports, that in some cases, it won't work.

But we aren't talking about general issues for everyone. But if it means the world to you to have either 88 fps average on a 90 hz vr headset or getting 92 fps average.. Then we start talking about all this stuff.

 

He is using a 9600k , not Ryzen ! Or did i miss something ?

 

You are right, I have I5 9600K.

 

Oops! Mixed the usernames, sorry.
Thought you were the one with the 3950...

With the 9600k you don't have ccx and you don't have latency issues (only a very low core to core latency).

I don't know why svchost would cause any load spikes but yeah, there's a problem with them apparently. Sounds like very low level stuff.

Maybe Windows Update and driver updates might solve it. Maybe you could try to kill everything in the background, including Dropbox, one drive, mouse/keyboard software.

What you also try is to run "Latency mon" while playing.
It causes some cpu load so definitely lower fps with it on but it will record any hiccups and log the process, DLLs, everything.

 

How is possible to know the correspondence between "svchost#68" (for example) and the real process name?

 

Afaik that is the real process name. It's one of the many svchost from windows.
Sadly I don't know of any way to know what programm or service uses that svchost to so something. I never saw this happening...

I think the best thing to try would be running latency mon for a while and check what causes spikes.
For me it's mainly Nvidia (DirectX) and network, which is kinda normal.
But I also had spikes coming from a USB device

 

Struggling to understand if my i7 4790k is good enough for the RTX 2070 Super in VR and... come out that sometimes ACC uses only 4 threads (and so holding everything else back if not driving in solo) but sometimes uses all the threads and then it’s just matter of graphic settings and GPU. Tried with 2 consecutive sessions without launching or closing any other app and how many cores/threads it will use its just luck.
(monitored with FPSvr in real time and MSI afterburner afterwards).

 

That's pretty weird if you didn't change anything...
Replay, solo, online and ai are using different amounts of threads but it should always be identical if you don't change anything.

If you can reproduce this, you should check if only Taskmanager/afterburner are showing different usage or if the fps are actually throttled by the cpu.
Windows is shuffling the loads across all cores on its own so it might look a lot different while the fps are identical.
Try to lower the resolution scale. Very ugly in vr, but it's the easy way to check the cpu fps limit.

Did you check if you're running into reprojection or not? Might make a big difference too...

About your cpu in general:
For 90 fps it's definitely holding back your gpu.
Holding back is the wrong phrasing... Your cpu simply won't run at 90 fps with more than probably 8 other cars on the track.

Depending on your graphic settings it might not be holding back your gpu, but instead simply holding back your vr experience in general.

If you're running 45 fps + asw with an oculus though or similar, your cpu is totally fine I'd say!