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.

 

[Justin]

OK having now argued myself out of a new build (for now at least ), I have come across a local listing for a seemingly lightly used i7 3770K that is cheap enough for me to take a risk on and see what improvements there are. Should I acquire that CPU, I'll hopefully get it to a stable overclock at the 4.4 GHz I have my 2500K at, maybe even 4.3 to account for the IPC generational gains between Sandy Bridge and Ivy Bridge, run some before/after tests and see what the results are.

I'm well aware Ivy Bridge is notorious for high temperatures limiting overclocking unless you do stuff like delidding, but hopefully my level is conservative enough to be able to pull that off, plus I can then be safely assured my 1070 won't be hitting any PCI-E 2.0 bandwidth limitations as my board can support 3.0 with an Ivy Bridge chip installed. Even though I'm pretty positive that it was never a factor before, it can't hurt I guess. Plus it's probably safer to bank on generational gains rather than a potential higher overclock with a 2600K/2700K, especially when I'm convinced my motherboard has always restricted the potential OC of my 2500K, even if I probably did lose out on the silicon lottery there as well.

Anyway, hopefully such tests could be helpful for other people on quad-core i5s to see whether it is worth upgrading to an eight thread i7 supported by their motherboard.

Another idea I've thought about testing, for Windows 10 users anyway, is the Game Mode function that you can activate via the game bar (Win key + G by default). I've always thought of this feature as useless, but it seems like it could be beneficial in games that are hitting a CPU bottleneck, as it arranges background resources and priorities to maximise game performance at the cost of multitasking ability. Would be good to test and see what happens.

 

[Case_]

Another idea I've thought about testing, for Windows 10 users anyway, is the Game Mode function that you can activate via the game bar (Win key + G by default). I've always thought of this feature as useless, but it seems like it could be beneficial in games that are hitting a CPU bottleneck, as it arranges background resources and priorities to maximise game performance at the cost of multitasking ability.

I've just been doing some tests on this myself, being fairly desperate, and honestly, it doesn't really do anything for me that would be worth mentioning.

 

[Deleted member 197115]

In fact it can add stuttering in some titles, worst MS gimmick ever.

 

[Case_]

I certainly wouldn't say it made things worse, nor would I call it worst gimmick ever, it just didn't really do much for me. Not that I reasonably expected it to, but I was hoping it might perhaps lower my CPU usage a few percent to allow me to consistently stream again. Not really.

 

[Neilski]

Cost unfortunately is an consideration as I obviously need a new motherboard for the 8 series processors. Already have ddr4.
I only play assetto corsa so that is all the question really relates to.

Personal view: AC makes so little use of the extra 4 threads an i7 can offer that any CPU upgrade you consider is going to offer minimal value for money, even if you don't replace the mobo.
If you're not already overclocking your i5 that's worth a look. Once you do so (e.g. 4.5 GHz ought to be fairly easy), even the very newest i7s (the ones that won't fit your mobo) are only just going to win in single-threaded code, and the 7700K will be so close as to be indistinguishable without a stopwatch.

 

[Deleted member 197115]

https://na.alienwarearena.com/ucf/s...0-game-mode-can-cause-lag-stuttering-in-games

Not sure if it actually was fixed since, but definitely was an issue. MS admitted it.

 

[Deleted member 197115]

Good article analyzing Game Mode performance
https://www.pcworld.com/article/318...akes-unplayable-games-playable-sometimes.html
Seems like if you don't have bulkload of crap running in background, it makes no difference, but on the other hand can make really needed apps (RHM, SimCommander) starve for resource

 

[Case_]

The problem is, both of these articles are over a year old, so not really relevant anymore. There's been too many changes that could've changed a lot of stuff (in many ways).

 

[Deleted member 197115]

The problem is, both of these articles are over a year old, so not really relevant anymore. There's been too many changes that could've changed a lot of stuff (in many ways).

Martin, it does not change the fundamental principle on which game mode works, which second article perfectly demonstrated. It didn't show the loss of performance mind you. but didn't show any gain unless you load system with a bunch of other background tasks.
From the article

Finally, it’s worth keeping in mind that Game Mode essentially blocks resources from any tasks running in the background, which can noticeably affect their performance.

If I want these background tasks to run, e.g. RHM, SC4, why would I want system impede them. And I saw some stuttering in Game mode even with recent Windows 10 public builds.
Anyway, what is your gain in AC with Game Mode on?

One more article for before bed time reading.
https://www.pcgamer.com/windows-10-game-mode-tested-good-for-minimum-fps-bad-for-multitasking/

Maybe there's some specific combination of hardware and software that will show greatly improved results with Game Mode. I doubt it, and a better solution is to shut down unneeded background tasks when playing games—at least if your PC doesn't have enough RAM and processor performance to handle things.

 

[Case_]

The article is a year and a half old. That's not my opinion, that's a fact.
A lot of things have changed in Windows that could've noticeably changed the end result of the things the article is describing. That's also not my opinion, that's another fact.

If you don't see any issues with that in relation to the relevancy of their findings, I can't really help you. I do and I have pointed it out. Nothing more, nothing less. It's as simple as that.

 

[Deleted member 197115]

Article is one year old, I agree. Does it change the fact it describes how Game Mode operates, or performance results?
I'd love to see the prove of that.
And you avoided my question, what is your performance gain in AC with Game Mode on (without background tasks running)

 

[Metla]

The general concept and function of Game Mode hasn't changed since it was launched. Results then as now are much the same,That it is pointless to the point of negative if the software environment is already configured to run efficiently.

So yes, The findings from a year ago can still be considered relevant.

 

[Case_]

Article is one year old, I agree. Does it change the fact it describes how Game Mode operates, or performance results?

No, it absolutely changes nothing about how the Game mode operated or performed a year and a half ago. It just has no relevancy to how it might operate or perform right now, because it is a year and a half old article about things that have changed in the meantime. I really don't get why that's so hard to understand or even agree on.

I'd love to see the prove of that.

Exactly my point, yes. That makes two of us. I would also love to see some up to date proof of that, because, again, a year and a half old article about something that has changed in the meantime (for better or worse) is useless. I don't, however, care about it so much to run and do the tests myself, which for some reason you seem to be suggesting I should do simply because I pointed out the fact the article is old and things might've changed in the meantime (or rather things absolutely HAVE changed, while the results might or might not have), which, for some weird reason, you seem to take as if I'm saying "but the Game Mode has changed and it works much better now", which I'm absolutely not.

And you avoided my question, what is your performance gain in AC with Game Mode on (without background tasks running)

I didn't avoid it, I just didn't find answering it relevant at all to what we were talking about, given that I stated right at the very beginning that I didn't really see much improvement. And I really can't give you much more than that, even if I found it relevant and wanted to answer you, because I didn't bother making any exact notes about it, precisely for the reason that it didn't seem worth it.

 

[Deleted member 197115]

The lack of recent benchmarks or any announcements from MS regarding improvements in Game Mode may only suggests that there is none.
Again, my personal testing showed no difference in games I play, but I had stuttering in some, and that was recent after MS patch.
I guess this is more relevant to me than some non existent up to date article.
And this is fact, as the fact that article is year old, which I agreed on.

 

[whip]

or like any normal person you can just use msi afterburner to check what usage you are getting on your gpu, if it's less than 99% with unlocked frame rates then you are cpu bottlenecked

 

[RasmusP]

or like any normal person you can just use msi afterburner to check what usage you are getting on your gpu, if it's less than 99% with unlocked frame rates then you are cpu bottlenecked

Yeah that's what I said in post 3

 

[Justin]

Game mode is seemingly getting some improvements in the October build of Windows 10, at least feature wise. Wonder if it will be any different in actual performance though.

_____

Anyway, got the i7-3770K in the rig and it seems to be a good chip, got it to the 4.4 GHz I wanted for this test without needing to do anything crazy.

Now the results for my testing in ACC are quite interesting indeed. Each test is the opening lap of a race with 20 opponents starting in last place using Afterburner's benchmark utility:

Time: 16:00, Conditions: Clear

i5 2500K at 4.4 GHz:
Average framerate : 69.1 FPS
Minimum framerate : 58.9 FPS
Maximum framerate : 82.9 FPS
1% low framerate : 50.3 FPS
0.1% low framerate : 40.1 FPS

i7 3770K at 4.4 GHz:
Average framerate : 73.3 FPS
Minimum framerate : 64.8 FPS
Maximum framerate : 82.8 FPS
1% low framerate : 59.6 FPS
0.1% low framerate : 49.8 FPS

Time: 16:00, Conditions: Heavy Rain

i5 2500K at 4.4 GHz:
Average framerate : 60.1 FPS
Minimum framerate : 47.2 FPS
Maximum framerate : 71.3 FPS
1% low framerate : 41.6 FPS
0.1% low framerate : 27.9 FPS (this result here probably should be closer to the 1% figure, there was a major stutter in the middle and I should have retested thinking back)

i7 3770K at 4.4 GHz:
Average framerate : 63.8 FPS
Minimum framerate : 49.4 FPS
Maximum framerate : 78.3 FPS
1% low framerate : 50.6 FPS
0.1% low framerate : 43.5 FPS

Time: 00:00, Conditions: Clear

i5 2500K at 4.4 GHz:
Average framerate : 63.3 FPS
Minimum framerate : 47.7 FPS
Maximum framerate : 81.3 FPS
1% low framerate : 46.0 FPS
0.1% low framerate : 41.9 FPS

i7 3770K at 4.4 GHz:
Average framerate : 63.2 FPS
Minimum framerate : 49.6 FPS
Maximum framerate : 82.5 FPS
1% low framerate : 48.3 FPS
0.1% low framerate : 40.9 FPS

Time: 00:00, Conditions: Heavy Rain

i5 2500K at 4.4 GHz:
Average framerate : 56.4 FPS
Minimum framerate : 40.8 FPS
Maximum framerate : 74.0 FPS
1% low framerate : 41.0 FPS
0.1% low framerate : 37.5 FPS

i7 3770K at 4.4 GHz:
Average framerate : 55.3 FPS
Minimum framerate : 42.7 FPS
Maximum framerate : 74.6 FPS
1% low framerate : 41.3 FPS
0.1% low framerate : 38.4 FPS

The best improvements seem to be occurring where I'd expect them, in minimums and frametimes, especially in the daytime clear conditions with some OK improvements to the day/wet race. 20 car races in daytime conditions are very close to keeping close to a locked 60 which I'm quite happy with, and considering I'm now rocking a probably stable 4.6 GHz overclock so I think I'll be OK here (4.7 requires way too much voltage and heat to stay stable for me).

However the races in the night-time show close to zero improvement and to me sort of confirms their lack of optimisation status. My new CPU still had loads of headroom on most threads, operating at frequencies that at least put it close to the newer 4c/8t i7s at stock, while at the same time never hitting 100% GPU usage, and with UE4 being an engine that is quite nicely multi-threaded, there surely has to be some improvements coming here as even GPU optimisation is far from optimal.

Still, I'm happy with the cheap upgrade that puts off a full new build for a little while at least and looking forward to see how things improve from 0.1 to 1.0.

Just a disclaimer: Even though I'm positive most of the results here are CPU limited, I really can't guarantee that my GPU wasn't at play here, even if I'm positive it never hit close 100% GPU usage in any test across the i5 and i7. I tested at 1080p using high settings which should well within range for a GTX 1070, but looking back I should have probably tested at 720p using low settings despite it not really being a real-world playing scenario. I'm in no mood to swap over the CPU again so this will have to do, but will update the thread my findings with the i7 at 720p low at 4.6 GHz and see if there's any significant improvement.

 

[natedogg1867]

Making FFB/physics depend on framerate would be going backwards years, if not decades. No serious developer would do that nowadays IMO (he says, thinking of iRacing's 60 Hz FFB...)

Well, there is at least 1 current sim where framerate greatly impacts FFB (if you cant maintain at least 60 fps). RaceRoom

You are certainly right that tying ffb to framerate in some way would be quite backwards in 2018... but with all of the complaints about the ffb feeling weird in ACC at sub 60 fps, I definitely think it is at least plausible that there is more funky stuff going on with ACC's ffb than just the major dampening issue that has been discussed to death already.

 

[manu68]

(I won't post them again, but if you want to see my own benchmark with a gtx1070ti and a i7-8700K for you to compare, you can go to the thread "PC ACC requirements).
https://www.racedepartment.com/threads/assetto-corsa-competizione-system-requirements.159262/page-2

 

[Justin]

(I won't post them again, but if you want to see my own benchmark with a gtx1070ti and a i7-8700K for you to compare, you can go to the thread "PC ACC requirements).
https://www.racedepartment.com/threads/assetto-corsa-competizione-system-requirements.159262/page-2

Interesting results. Next time I test I may do the same tests as you at a slightly lower resolution to equalise the difference between a 1070 Ti and a 1070, and see what the difference is between my 3770K at 4.6 GHz and your 8700K, which is definitely strong enough to have your GPU as the likely bottleneck in all situations.

Just a question, in Afterburner are you monitoring your GPU usage? If you do, does it constantly stay at close to 100% use in all these tests? If it drops below like 90 per cent for significant periods of time especially in night tests, then it's almost certain that poor performance is out of our hands for now, as there's no way any game should be bottlenecked by an 8700K right now.

___

In unrelated news, the one game that caused me the most issues with CPU bottlenecking on the i5, Battlefield 1, now runs very smoothly on the i7 with loads of headroom, so there's no doubt some games have shown huge improvement by the CPU change.