The ACC CPU Benchmark Thread

I play with medium video setting and win 10 with last 2004 updated. My PC configuration is in signature.

 

Where do you see the 100%?
In theory ACC isn't able to use all 6 cores so either the reading is wrong or something is using your cpu in the background.
However 26 ai (how many opponents visible in the graphics settings?) is quite a lot so you definitely can hit the cpu limit.
But it shouldn't be at 100%.

Instead your fps should go down because of the hidden single thread limit, while the overall cpu load shouldn't go higher than about 70-80%.

 

It is the same thinks that I thought but with win10 game bar I have this feedback. The CPU works normaly around at 85-90% but often it reaches 100% usage which coincides with boring stuttering but the fps seems the same (85-90 fps). I also check if there are something that use CPU in background with win task manager but I don't find any strange applications that occupy cpu.
The opponents visible is set to 15.
Have you to suggest me some other tests?

 

I have a 10600k so basically your cpu with hyperthreading. I went into the bios, disabled HT and did a quick race with 26 AI, lowest resolution to not cause limitation from my gtx 1070.
15 opponents visible, just like you.

2 things:
1. Interesting "bug": Taskmanager shows higher CPU load than msi afterburner or openhardwaremonitor. If I disable turboboost (go into powerplan settings, set maximum processor frequency to 99%), my CPU will only run at 4.0 GHz instead of 4.8 GHz.
This funnily fixes the wrong CPU load shown by Taskmanager and every monitoring software shows the same CPU load then.

As I mentioned, ACC can't use full 6 cores so while my FPS went down, the CPU load stayed the same.

2. This is how it looks like for me. There is NO spike to 100%. Highest spike was to 81.5% so I'm wondering where the 18.5% CPU load come from on your system..

I put ACC into windowed mode and reduced resolution to have multiple windows visible at the same time.
Please do the same, open Taskmanager and do a full formation lap from last place. This caused the highest CPU load for me.
There should only be "AC2-WinShipping64" or whatever the ACC exe is called and "idle" or whatever that's called in Italian showing any CPU load higher than 1-2%.
For me it's "Leerlaufprozess" -> "Neutral process" when using racing terms, lol.

Maybe you have another process there? Gamebar maybe?

 

I did the test as suggested by you and I did not find any problems .... at first. Continuing, however, I noticed that the CPU reached the 100% peak. A process that saturates the CPU appears to randomly trigger. I have to understand what this process is .... difficult task.

 

Windows has a tool called "perfmon" where you can create a graphs with the processor usage % of all running processes in your system. These graphs will contain lots of processes, but it should be easy to spot random peaks.

You start it, right click on the bottom and add something like "performance monitors" (sorry my windows is not english)
there you look for process - processortime (%) and below <all instances>

 

Hi there,

I am planning to upgrade my CPU as I am playing only ACC online in VR (rift S) and I want to achieve 80fps.

I run I7 7700k paired with 1080ti.
I am able to achieve 80 fps all the time unless I join a very busy server. I would say that there is a kind of wall at around 40 players. Let say that a server with 50 opponents, the FPS drop around 60/70.

I am planning to upgrade to i5 9600K.

So simple question. Will I sustain 80 fps on a biggrid server ? Should I get the boost of 10/20 fps with I5 9600K compared to I7 7700K.
I will keel my ram which is at 3600c16.

Thank you.
Cyril.

 

I dont see how the 6 core will gain you much of anything in ACC ! If anything the 9900k/10900k would be the pick just because it can pass 5GHz but even then im not sure it will gain you too much over a 5GHz 7700k .

 

We can calculate it pretty well. Download cinebench r15 (r20 used avx, acc doesn't so we'll use r15).
Then go into (top left):
File - > Preferences - > check "Custom Number of Render Threads".

Do a run for 3,4,5 Threads.
ACC runs on 2-4 big threads depending on AI, online, replay etc.
From my charts on the previous page you can see that more than 5 cores barely improve fps with full AI grid (most thread usage).

Anyway, do the 3 runs and post your points.
If you can be bothered do every run 3x and take the average or highest result.

I have a 10600k, which basically gets the same results as a 8700k and 9600k when at the same clock speed.
I can run 4.8 GHz with ht on and 5.0 GHz with ht off.
If I up the voltage a bit more, 5.1 GHz are possible without ht.

So I can provide you with my results and then you'll get minimum and maximum expected fps difference to your current cpu.

Post your results, then I'll bench my cpu too

 

Hi @Krassmus

Thanks for your reply. As I drive exclusively online should the offline test you suggest will show something? I guess to some extend yes but I already know that my system is not able to sustain 80fs with 28ai. I think that my brake even is 15ai ATM.

Anyway I will do the monza procedure and will post here the results.

Thank you.

 

Just do the cinebench runs. It runs without installation and gives clear results.

But yeah additionally we can compare the monza procedure too. I thought we could spare us all the setting up

If you have a look at my earlier post, here are the thread usages in different scenarios of acc:
(process Explorer can show cpu load of each application thread)

So 1 thread is hitting the single thread cpu limit (100% / 12 cpu threads = 8.33% Max per core).
While 3 other pretty intense threads will benefit from running on separate cores.
To have one core fully available for that one big thread at all times.
Rest of the system and game will use another core that will barely be loaded but will handle the background stuff etc.
So 5 cores is what will give you enough cores to have the maximum power available for that one biggest thread.

It's a bit weird to understand this all if you never heard about it. Complicated stuff but Taskmanager won't show you anything useful as long as your cpu isn't completely maxed out at 100%.

The hidden single thread limit can only be "shown" by checking process explorer. Sadly only on real time.. No history graph or anything available.

So anyway, we have 1 big thread, 2 slightly bigger, 1 small and a chunk of background stuff.
Meaning doing a cinebench of 3, 4 and 5 rendering threads will come pretty close to what acc uses.

 

ok so here are the results
3 -> 587
4 -> 692
5 -> 795

 

So where does this leave the 7700k with 4 cores and 8 threads ?

 

At "4 cores HT", since online racing in ACC still uses 4 "bigger" threads so needs 5 cores to maximize performance (1 extra core everything else than the game).
But we'll see it in Cinebench pretty clearly.

At 4.8 GHz (ht disabled), my points are:
3 -> 621 (+5.8%)
4 -> 813 (+17.5%)
5 -> 1012 (+27.3%)

At 5.0 GHz (ht disabled), my points are:
3 -> 645 (+9.9%)
4 -> 856 (+23.7%)
5 -> 1058 (+33.1%)

Looking at how much you gain when going from 4 cores HT to 5 cores in ACC, but the big game-threads only being 4:
I'd say the gain will be between Cinebench's 4 and 5 threads.
So 17.5% in the worst case and 33.1% in the best case.

Probably around 22%.
So instead of 80 fps, you'd have 97.6 fps.

If you want to have real ACC numbers, we can do "the procedure". But I might not be able to do that before Sunday.

 

What do you mean "bigger" threads ? and how did you come up with ACC needs 5 cores ?

 

By testing and observing. Check the previous post and video, the one with the graph he quoted here.

 

@Krassmus

Thanks for the answer. This is valuable source of information.

If I follow the conclusion, I would be able to achieve 80fps in VR (since VR Perf is very sensitive to CPU and ram) where I am around 65fps today.
And if I upgrade my Ram to 4000 or even more, I would gain some fps to be more comfortable.
I am pretty much convinced that my bottleneck is CPU RAM when joining biggrid server as the visible cars are set to 15 and I have no issue to maintain 80fps on server up to 40 opponents.

My last 12 hours endurance race (50 opponents) was with ASW on at the start and I was able to unlock ASW later during the race when reaching 40-42 opponents after some DNF.

So now I have to decide if a 500€ upgrade is worth the biggrid experience!

 

I’m looking at upgrading my i5 4960k with a new MB and CPU.

I’m no tech-wizard by any stretch of the imagination but I think I understand enough about this thread to save a bit of money and get an i5 9600k rather than the newer 10600?

should I be looking at anything else for similar bang/buck?

I’m running a 1660ti atm so definitely cpu bottlenecked.
I run in stereo 3d so not as intensive as vr but more than 2d.

solo lapping I can get a steady 60fps on most tracks with my graphics options (dips to high 30’s at race start in sp and low-mid 40’s in mp), I’m hoping for a steady 60fps to match my projector.

 

Process Explorer.
Screenshots of different thread usage:

The maximum of any thread is 8.33% in theory. So since I'm in the CPU limit during these screenshots, the 8.22% or whatever are close enough to be called "Single thread performance limit".
Conclusion from this is that you want to have 1 separate cpu core for each thread that reaches above 8%.

To understand this better: If you put cinebench to 24 threads on my 12 thread cpu (6c+ht), Process Explorer will show only 4.2% for each thread (if they all get the same priority ofc).

For maximum fps you need to cut the process explorer list into pieces of "8.33%". You can't cut single threads though.
So 6.0% + 1.5% cpu usage threads are okay to run on 1 core. But 5% + 5% will make one of these threads having to wait at some point.

So looking at my screenshots above you'll see that 4-5 cpu cores are enough for ACC. HT makes things better with 4 cores but actually worse with 5+ cores! For whatever reason the scheduling with HT seems to slow down the single core's throughput.

What you can also see here though:
AC and rF2 only run on 2 big threads! So there you need to compare the "3 thread" scores of Cinebench.
As you can see, it wouldn't be worth it at all there.

I wouldn't upgrade the RAM! You can see from my big testing page that 3200 to 3400 barely gives an fps boost. 3400 with better timings does neither.
So 3600 to 4000 won't gain you much but is very pricey!

I can recommend the Gigabyte z390 Gaming X. Very good VRMs, BIOS has enough settings for everything, 8 USB instead of the 6 other cheaper mobos have and it runs very nicely with the 5.1 GHz 9600k of my best buddy.
My z490 Gaming X is very nice too.

In Germany that would make the upgrade:
120€ Motherboard
189€ 9600kf

 

Sorry for the double post but I wanted to split topics.

The problem with VR isn't the 1-2% additional load from the VR controller. The problem is that your fps get halfed + interpolation the moment you drop below the refresh rate of the headset.
You could also disable that but stuttering is really really awful in VR!

About CPU upgrade:
yes, get the 9600k! If you play new AAA games like Assassin's Creed you should get the 10600k since these games can utilise more than 6 CPU threads.
But simracing doesn't look to achieve similar level of multithreading anytime soon...
Without HT the CPUs usually can be overclocked higher so you even gain performance.

Look at my 10600k: 4.8 GHz with HT stable, 5.0 GHz without HT stable. Same vcore!
Less heat too so I could push the vcore a bit higher and run 5.1 or 5.2 GHz too.

If you want to "be smart" and are able to wait a little:
Get a b550 motherboard and the Ryzen 3600 (not the x, not the xt) and when Ryzen 4xxx comes out at the end of the year, you'll be able to buy a new CPU and put it into the same motherboard!
Ryzen is way more efficient and almost gives you the same fps. But if you really really need every fps you can get, Intel is the way to go for simracing.

I wanted the maximum performance NOW, so I got the 10600k. But the smarter plan would've been to get a 3600 now and the 4700x in December for example.