Finding out whether your FPS are CPU or graphics card limited

[dud]

Do just take the "unreal is cpu intensive" unmeasured blah-blah statements.

Do this:

• get framerate in a reproducible way that you like
• go into the BIOS and downclock your CPU by 1/3rd, repeat benchmark
• download an overclocking utility for your graphics card and use it to downclock your GPU by 1/3rd, repeat benchmark
• for added fidelity, do same for:
• RAM
• graphic card RAM

I don't have time right now but I'll post mine later.

We need a reference thread about hardware-to-fps anyway.

 

[Opus131]

@suspectmonkey Yep, just like me - somewhat GPU bottlenecked in solo mode, CPU bottlenecked in AI races. I can only run around 6-8 AI opponents before the CPU is no longer able to keep up.

Being GPU bottlenecked is certainly a lot better, since you can drop some setting to manage. With CPU bottleneck, you're pretty much screwed

Not only that but it's easier to upgrade GPU.

I have an AMD FX-8350. This thing is a fossil at this point, and in order to upgrade it i have to get a new motherboard as well as new RAM.

I'll definitely be upgrading my graphic card soon. Once i do we'll see how important the CPU is. I'll report back here if this thread still exists then, in the name of science.

 

[Case_]

I'll definitely be upgrading by graphic card soon. Once i do we'll see how important the CPU is.

For ACC - a lot. I wouldn't bother with a GPU upgrade on AMD FX, honestly, but that might be just me. I wouldn't bother with it even on my overclocked i5 2500k - doesn't make much sense.

 

[Pawel Ziolkowski]

• get framerate in a reproducible way that you like

I have noticed that there is not much difference between actual game and its replay's performance (using cockpit view for both) , could "using the replay function" be a good way for having the same piece of gameplay to test our frame rates in different settings/clocks ?

It's a "good enough" method to test GPU, but if you ever get CPU bound replay will run at higher FPS. And since cpu vs gpu is the actual topic of this thread, the answer is a definite no.

 

[RasmusP]

Isn't it the opposite? If your GPU is pegged at nearly full usage, but your CPU is not, then that means the app is using the full potential of the GPU and would therefore benefit from any more potential that a more powerful GPU would provide -- i.e., the app is currently GPU-limited.

Yeah I think that was a typo

Oops!
Corrected to "but below"!
I meant that if the gpu is at full potential but not being used up to it, it's the rest of the pc that's limiting
But mobo and ram barely have such a big influence that it would make sense to swap them out instead of a better cpu. If they do, it's the 1%

 

[Jempy]

Well ... with my config:
- I5 4670k not overclocked
- GTX1060 OC V1 MSI 6GB
- RAM 8GB

2 screenshots will speak by themselves

In such conditions, ACC will sleep in its HDD until optimization.
CPU already at +- 100% low settings and just my car !

NB: No problem for frame rates ... as you may surely see
On MID settings 55 fps with 15 cars at start, between 60 and 70 fps in race.

At this 1st stage of ACC, I may understand ... I just hope it will be optimized

 

[Neilski]

CPU already at +- 100% low settings and just my car !

Your second screenshot has 140 FPS - it wouldn't bother me at all for the CPU to be nearly flat out at that frame rate. You can safely jack up the graphics settings quite a bit (maybe even all the way to EPIC) with that GPU.
You don't feel that 55+ FPS is usable for racing? I could live with that no problem (I normally use vsync on a 60 Hz monitor).
Edit: any reason why you aren't overclocking that CPU?

 

[Case_]

@Jempy That's some pretty high CPU temperatures for not being overclocked, IMO. I'm at similar temperatures with my i5 2500k overclocked at 4.5 GHz if it runs at 100 % for a while, and at least in practice mode with me alone on the track, I'm at like 60-70 % CPU utilization, and that's with several things running in the background.

If I were in your position, I'd look into better cooling or check what is causing those high temps, and then overclock your CPU - it is a "k", after all, so you should be able to get some noticeable improvement in performance without much problems.

I'd also say it might be counterproductive for you to lower the settings so much, your 1060 should be perfectly capable of running the game at close to max settings at 1080p.

Edit: Oh right, I didn't even notice the framerates in the corner. Well, I guess my standards are different, then.

 

[Jempy]

Your second screenshot has 140 FPS - it wouldn't bother me at all for the CPU to be nearly flat out at that frame rate. You can safely jack up the graphics settings quite a bit (maybe even all the way to EPIC) with that GPU.
You don't feel that 55+ FPS is usable for racing? I could live with that no problem (I normally use vsync on a 60 Hz monitor).
Edit: any reason why you aren't overclocking that CPU?

In fact, I tried lowering the graphics settings in order to see if the CPU use might be lower than 100% with Mid or Epic graphics settings. .... but the gain was negligible as you can see.

At Epic settings ... with a "grid" of 15 cars ... in daylight without rain or night, this meant +- 42 fps at the start and mostly between 45 and 60 on track

That's why I prefered first to have a lower graphics setting with some High settings and other with Mid settings in order to get a 60-70 fps on my other monitor for gaming: 21/9 with a 2560*1080 resolution ( 29" LG ).

Now about overclocking the CPU .... I never did this kind of operation and am not a real PC specialist. That's why I hesitate to overclock it and cannot be sure it will not make its life much too short.
But of course, it's a choice I'll have to do .... overclock or replace the CPU by a better one and maybe being forced to change the mother board also as I'm not sure the MSI Z87 G45 Gaming may support this kind of CPU..... ( and maybe add RAM !? )
And this second choice should be forgotten for financial reasons of course.

@Martin Fiala
Idle temperatures are round the 35-36°C ... that seems +- correct for this kind of CPU ?
I just checked with 3DSimed opened + Gimp + Discord ... and AV running of course.

 

[RasmusP]

Idle temperatures are round the 35-36°C ... that seems +- correct for this kind of CPU ?

It's not about the idles but about the load temps of around 80°c.
Without oc my I7 2600k won't go above 60° while at 4.4 GHz it will max out at around 75°.
I have a pretty warm case but the thermal right Le grand Macho (one of the best air coolers) and optimized the airflow as much as I can.

Your cpu will run a bit higher temps but it shouldn't be that high at only 3.6 GHz!
I mean it's totally fine but there's either something odd going on or you just don't have a really good cpu cooler. (not good = not meant for high overclock)!
May I ask which one you got?

 

[Case_]

At Epic settings ... with a "grid" of 15 cars ... in daylight without rain or night, this meant +- 42 fps at the start and mostly between 45 and 60 on track

With 15 cars, you're almost guaranteed to be CPU limited in ACC with that i5. I can run around 6-8 AI before I start bottlenecking at the CPU, and I'd say my CPU is more or less equal to yours or even slightly better, considering my overclock.

Idle temperatures are round the 35-36°C ... that seems +- correct for this kind of CPU ?

Yeah, that sounds OK. But it's not really idle temps you have to worry about most of the time, they will almost always be pretty low with these, unless you have some really serious cooling issue like failing fan or something.

 

[Neilski]

I haven't paid attention yet to just how many threads ACC really works hard - thus how many cores you need to get the best out of it. (Grateful for any pointers that you guys know of.)
Depending on the precise code, for single-threaded stuff a 4670K is about 75% of the speed of the fastest CPU that you can buy today. For heavily multi-threaded stuff it could fall below 50% of the fastest CPUs...

 

[Jempy]

@RasmusP
That's the original CPU cooler ... nothing more ( else said: surely not the best ! ).

And that is surely the reason why temperatures might be so high and I'd hesitate for overclocking the CPU.

Adding to this, at 1st floor ... PC not on the ground but on desk due to lack of place and ... worse.... 2 cats at home . .... that's surely 23-24°C for now that our hot summer is gone.

@Martin Fiala
I suspect of course that the CPU might be the biggest problem: 4 cores while 6 are recommended and surely also only 8GB Ram .... but it's a real fact that this Unreal Engine seems very very heavy for CPU.

NB: at max settings with AC, 20 cars on track +- 100 Fps and only +- 65% CPU use.
The difference is huge between this and ... only 1 car on track with ACC.
I really think the UE is really the big reason.

 

[RasmusP]

@RasmusP
That's the original CPU cooler ... nothing more ( else said: surely not the best ! ).

And that is surely the reason why temperatures might be so high and I'd hesitate for overclocking the CPU.

Adding to this, at 1st floor ... PC not on the ground but on desk due to lack of place and ... worse.... 2 cats at home . .... that's surely 23-24°C for now that our hot summer is gone.

@Martin Fiala
I suspect of course that the CPU might be the biggest problem: 4 cores while 6 are recommended and surely also only 8GB Ram .... but it's a real fact that this Unreal Engine seems very very heavy for CPU.

NB: at max settings with AC, 20 cars on track +- 100 Fps and only +- 65% CPU use.
The difference is huge between this and ... only 1 car on track with ACC.
I really think the UE is really the big reason.

Wow so the boxed cooler from Intel?
Then these temps are kinda low, lol!

ACC has a lot more complex physics going on and to make the experience believable, the AI runs the same physics (same für ac).
That means your car + 1x AI is like 2x your car + driving it instead of you giving inputs!

I doubt that the engine has anything to do with it. I'd rather think most other engines would be running even worse!

Also ac1 might be cpu limited too. 65% could just mean that it purely doesn't run on enough threads to use multiple cores to the full extend.

Gothic 3 from 2006 is cpu limited at 13% cpu usage. Runs on one thread. 100% / 8 cpu-threads (i7, 4 cores + HT) = 12.5% per thread.

Whenever you see software using 100% cpu it may be the case of bad optimization but it might also mean that the software can make full use of your cpu!

 

[Opus131]

You don't feel that 55+ FPS is usable for racing? I could live with that no problem (I normally use vsync on a 60 Hz monitor).

Unless you have a freesync or gsync monitor, anything that goes below the default frame rate is a no no. Vsync is little help. If you use normal vsync, the game's frame rate drops to 30 fps instantly which is jarring as hell, where as if you use adaptive vsync, you get tearing even if it drops by a single fps. Not as bad as not having vsync at all, but it is still noticeable, even in a racing game where you don't really move your camera around that often, unlike a shooter for instance, where screen tearing is a death sentence (at least for me, i seriously hate this type of stuff. Much like i hate flickering shadows, poor drawing distances, jaggies etc, things which are EVERYWHERE this days and hard to get rid of).

Personally, monitor sync is one of the best technologies to have come out this past years and it allows to stretch your hardware for longer, since as long as your fps is within the range of the monitor sync, you are fine, you don't necessarily have to match the monitor's default frequency at all costs. The problem is that nvidia messed everything up by coming up with an expensive proprietary module. Freesync is a lot more affordable and it's pretty much everywhere at this point, but if you go that way you have to deal with AMD cards, which, while they are not as bad as people make them out to be, they are not as attractive as their nvidia counterparts.

 

[Case_]

I haven't paid attention yet to just how many threads ACC really works hard - thus how many cores you need to get the best out of it. (Grateful for any pointers that you guys know of.)
Depending on the precise code, for single-threaded stuff a 4670K is about 75% of the speed of the fastest CPU that you can buy today. For heavily multi-threaded stuff it could fall below 50% of the fastest CPUs...

From what I've seen so far, there seems to be a considerable difference in ACC performance between i5 and i7, suggesting 4 threads are not enough.

That's the original CPU cooler ... nothing more ( else said: surely not the best ! ).

Surely not the best indeed. Actually one of the worst. You might be way better off investing in something better. Wouldn't have to be a big investment either - I'm running SilentiumPC Fera 3, for example, which is like 25 euro, at least around here, and really a budget solution, and I think you could expect your temperatures to drop around 20 degrees under load, giving you plenty of room for decent overclock.

but it's a real fact that this Unreal Engine seems very very heavy for CPU.

NB: at max settings with AC, 20 cars on track +- 100 Fps and only +- 65% CPU use.
The difference is huge between this and ... only 1 car on track with ACC.
I really think the UE is really the big reason.

People are kinda quick to point fingers at UE4, but I really don't think it's that simple. Even the original AC was a pretty CPU intensive game, I was at times very close to 100 % CPU even in multiplayer, and for example streaming sometimes pushed me over the limit. So not that big of a difference with ACC for me, really. There was an increase, sure, but not *that* significant. It's just that the framerate was higher, given the relative simplicity of the original's graphics engine. I'm pretty sure it's still the physics calculations that are most CPU demanding in ACC, not visuals. There's plenty of well optimized UE4 games out there (and plenty of not so well optimized, sure).

 

[Jempy]

If I was sure, Martin, that it's not only a fact of optimization still to come .... and the MSI Z87 G85 gaming might receive a new I7 CPU ( with a better cooler than the original one for sure ).... I think I might get a new one within a few months ... but keep the 8GB Ram .... to spare a bit !

It should have 6 threads in place of 4 ... and I'd be good for some years .... rather than overclocking an old 4 threads. ( Old ? ... not really in fact +- 3 or 4 years .... but PCs are getting quicker older than we do finally ! ).

 

[natedogg1867]

and I'd say my CPU is more or less equal to yours or even slightly better, considering my overclock.

Eh, Im not sure I would say that. The 4670k should still be a good bit ahead. It's 2 generations newer and had a die shrink. IPC gains in those 2 generations was pretty decent as well. 3.8GHz compared to your 4.5 isnt very much of an improvement on my own 4670k running that frequency.

 

[Case_]

@natedogg1867 I didn't do any serious research on that, but based on some benchmarks I quickly glossed over before replying, 2500k @ 4.5 seems to perform almost identically to stock 4670k.

 

[farjam]

Neither CPU or GPU are passing 60-70% usage (average GPU usage is 60% in 1920x1080 borderless, 65% for 1920x1080 fullscreen , 70-75% in 2560x1080) , whats going on?


*there are no FPS limiters active
*this is the only software/game that my GPU acting like this in it


edit: system config is 8700k + 1080ti + 16GB ram

 

[Deleted member 197115]

If it's single CPU thread that becomes a bottleneck you won't see that.
It's simple, if it's not GPU, it's CPU. I think you are doing fine with 100fps, add VSync or frame limiter and CPU usage will drop.