Test: CPU Core count and RAM scaling in ACC, AC1 and R3E

[2stains]

yes

How does hyper threading effect the amount of Ai ? Do you expect similar results as here or is it beneficial ?

 

[Dan Costa]

Really impressive amount of work here
One question though , you did these tests yourself with a direct die cooled 9900k @ 5.2GHz yet you have a 2700X @ 4.3GHz as your own gaming rig according to your sig . Do you prefer to game on the Ryzen chip for any reason in particular ?

I dont have the ryzen anymore. The cpu is good enough for most games @ 60fps but wasn't good enough for VR. The 9900K is leaps faster

 

[RasmusP]

Really impressive amount of work here
One question though , you did these tests yourself with a direct die cooled 9900k @ 5.2GHz yet you have a 2700X @ 4.3GHz as your own gaming rig according to your sig . Do you prefer to game on the Ryzen chip for any reason in particular ?

Look at the core amount scaling, then look at the single thread performance of both CPUs (pcmark website comparison for example).
You'll see that the 9900k is a lot better suited. Although the 9600k, if overclocked, basically reaches the same fps for simracing

Which leads to Dan's reply

I dont have the ryzen anymore. The cpu is good enough for most games @ 60fps but wasn't good enough for VR. The 9900K is leaps faster

 

[2stains]

Sig needs an update then

 

[RobertR1]

@Dan Costa been messing around with adaptive/offset the past few days. Works quite well. You still on fixed?

 

[Dan Costa]

@Dan Costa been messing around with adaptive/offset the past few days. Works quite well. You still on fixed?

Yep I'm still on 1.390v fixed with LLC Turbo. I read your discussion in overclockers.net, but i'm still to try it out. What would be your starting point for adaptive offset if your fixed stable voltage was 1.390v?

 

[RobertR1]

Yep I'm still on 1.390v fixed with LLC Turbo. I read your discussion in overclockers.net, but i'm still to try it out. What would be your starting point for adaptive offset if your fixed stable voltage was 1.390v?

So I was at this for fixed:
1.385v LLC High. 53x core and 47uncore. This gave me load vrvout 1.29v (give or take) when pushed like R20 etc.

I'm now at:
Vcore normal
Ac loadline power save
Llc low
Intel speed shift enabled
Core 52
Uncore 47
Offset +.080

For 53x offset = +0.110

So maybe start off with +0.110 offset and scale down from there. I'd recommend starting clean with "load optimized defaults." Had all sorts of issues trying to switch from fixed to adaptive in a modded profile.

edit: end of that experiment. couldn't get bf5 to stop crashing and even going up in offset was just causing heat issues. Went back to static and all is well. Experiment over.

 

[Dan Costa]

did you manage to hit 53x with all 8 cores enabled?

 

[RobertR1]

did you manage to hit 53x with all 8 cores enabled?

Yeah. 53x core 47x uncore. All core HT disabled. 1.39v in bios with LLC high. Load under R20/P95 is 1.29 vrvout. High LLC proved to be more stable for me than Turbo. Been running that daily for a few days now without issue. Even BF5 is happy and that's never easy.....

 

[Dan Costa]

Couldn't get mine to pass IntelBurnTest 5.3ghz even with only 2 cores. You are lucky!

 

[RobertR1]

Couldn't get mine to pass IntelBurnTest 5.3ghz even with only 2 cores. You are lucky!

Yeah there's no secret sauce here. Just pure silicon lottery luck. The diff between 52/53 is virtually nothing.

 

[Jens Roos]

Might be dumb question, but what are the numbers in the graphs indicating (scaling)?

 

[Dan Costa]

Might be dumb question, but what are the numbers in the graphs indicating (scaling)?

eh, forgot to put "FPS - Higher is better". Oh well, next time

 

[m_mirk]

Great Tests! Thank you for that. Do you also plan tests with different CPU Clock?

 

[Jens Roos]

"FPS - Higher is better".

Great work, so it more or less confirms that with 4 or more cores there is little to no difference in performance in the tested titles (as indicated in other threads as well).

 

[Dan Costa]

Great Tests! Thank you for that. Do you also plan tests with different CPU Clock?

Its not likely because my vacations are now over and its very time consuming. I can do one game, maybe. But not the 3. Which one do you suggest?

Great work, so it more or less confirms that with 4 or more cores there is little to no difference in performance in the tested titles (as indicated in other threads as well).

That may well be true for R3E and AC1. But for ACC, that game was being bottlenecked by my 1080TI even at 720P. It may be scaling beyond 4 cores if I had the 2080TI.

 

[RasmusP]

But for ACC, that game was being bottlenecked by my 1080TI even at 720P. It may be scaling beyond 4 cores if I had the 2080TI.

I smell someone going 800x600 with minimum resolution scale

 

[RobertR1]

That may well be true for R3E and AC1. But for ACC, that game was being bottlenecked by my 1080TI even at 720P. It may be scaling beyond 4 cores if I had the 2080TI.

I can test on my 2080ti if there's a way to get the replay file? see if there's a bottleneck there.

 

[Dan Costa]

I can test on my 2080ti if there's a way to get the replay file? see if there's a bottleneck there.

Sure i'll get you the replay when I get home.

But there would be no correlation with my results, you would need to run all benches from scratch.

It doesn't have to be my replay either. You can use any replay with full grid and set the camera on the pole car looking backwards. I chose a race I won at brands hatch because in a small track you can see geometry from other sections of the track even though you are driving somewhere else. And that main straight has a lot of detail. With 30 cars its a massacre.

 

[RobertR1]

I'd only be doing the core scaling test with ACC if you felt that you were GPU limited. AC tests aren't going to benefit anymore because you're at high CPU bound fps already and I don't own R3E.

If that doesn't take forever, I might do freq scaling. Going from 4.8 to 5.0 to 5.2 to see how much you gain. I'm also stuck on C15 3600mhz 1T so can't do a like for like there.