Terrible GPU utilisation?

Like I said there are a lot of variables that go into the performance I have noticed with my system. I never said the results will be the same across the board. I simply took results from what I saw and attempted to help someone figure out an issue. If you don't agree with the concept you can do the research and if you don't think my results are true then I can't help you.

 

It is kinda complicated, but in theory, if code A is what causes CPU to hit limits of SC performance, while code B is what is keeping GPU fed, then limit is reached earlier and FPS gets cut off because of code A has maxed out the core. Both codes run on same core, competing from available limited computing power, we can call this single core headroom or something.

With higher graphics code B takes more of computing times from core and if there is GPU power available, in theory it can allow more stuff being processed to GPU despite it being more work for the core, but code A this time gets less work done for given time frame.

Just a theory though, haven't tested in practice.

Now how lowering GPU clocks would then cause that is beyond me, I would think that there is some other variable in play which has not been realized.

Thing is though that you don't want to end up CPU limited ever really, that causes stutter and rapid drop of fps etc. Vsync limited it best, GPU limited is 2nd best. Sadly FPS limiters seem to cause a lot of stuttering in this game, so those don't work so well.

That single core load in this game fluctuates rapidly, it is spiking up and down in 20ms or something like that, really many times in second, which then can cause waits which are reason why GPU is not fed which is then causing fps to drop rapidly.

So adjusting settings etc. so that such spiking is not touching 100% too often within period of one second is way to get better smoothness and fps. Hence vsync works wonderfully as it kinda limits GPU to lower level, so while I can't quite understand how lower GPU clock would help to get more fps, I can tell how limiting fps by GPU is getting game smoother.

Going much over 60fps tends to also cause LUA etc. to take more CPU time, so something in there might make his situation possible with certain conditions, in theory at least, but CPU would need to get easier time because of GPU limit and that does not quite fit in to what he describes.

 

Interesting:


I wonder what kind of difference would be with this scene, Italy Norte spawn:


I have I7-6700 non K, Ryzen 2700x has something like twice the computing power, but with stock clocks weaker single core. (according to user benchmark it is 119 points and I get 121 points, so I guess I should really say it is same, variation is bigger difference than such point difference.)

 

Have high graphics on, ssao on, dynamic reflections on, like you had in Utah, spawn at Italy map Norte spawn point, turn orbit camera to position, there is not really anything more to it.

 

You notice how your GPU is not utilized so much?

On some maps, like Utah, CPU has easy time, on some other maps CPU has to work harder, ssao affects a lot how much GPU gets load, but it also increases CPU single core load, which then can cause drop in GPU utilization, dynamic reflections even more so.

So as you mentioned how your CPU claims to differ, it really does not, only in some situations your CPU is more limited by single core and in some situations less. With 0.15 update there has been massive improvement to this and there was mentioning that further optimizations might come, so it is getting probably better so using ssao and dynamic reflections becomes more possible without dropping under 60fps.

Also using clone of same vehicle tends to be easier for single core than using many different types of vehicles as there is less batching.

That kinda was whole point, even best CPUs are facing this limit and with really many games, but luckily it does not become much of an issue anymore like it used to become, after all while fps is staying above 60 even a little using vsync this game runs butter smooth.

 

Bottlenecked CPU is something I've taken into consideration. But even if its not technically bottlenecked I bet its possible because its used more in game than the GPU that lowering the GPU's clocks will increase fps. There are a lot of variables but my system is fairly balanced so why I see this is could just be a number of different things causing this.

 

I am not sure if you are thinking of someone else but my cpu is clocked 4.4 across the board with +.108 on the vcore in the bios. I believe because the game is more cpu intensive you would technically need a computer GPU bottlenecked to get best possible performance but that is just a theory and I can't test that. The game has been getting better but it begs the question with modern GPUs why they are not used to offload calculations to it instead of the already stressed CPU. Obliviously not everything can be done on the GPU but the line of what has to be done on CPU compared to GPU has been getting blurred for a while. It'd be nice to see optimizations that could allow the GPU be used for more than it is now. I could be wrong, they might already do this, but the difference in usage between the two says differently. Look at games like fortnite, it generally will uses similar GPU power as the CPU for my system and it shows in fps. Now mind you, they are very different games but it really asks the question if BeamNG Drive is as optimized as it could be since the rule of thumb has been equal usage generally shows an attempt at optimizations, at the very least.

 

I doubt that Fortnite runs incredible complex physics calculations at 2000 times per second.

 

Yeah but I also doubt it runs the calculations it has to only on the CPU and doesn't split the load with the GPU when it makes sense to do so.

 

I think forcing the GPU at a lower clock rate will smooth things out if it is CPU limited, since then the "keeping GPU fed" code doesn't use as much CPU and hence doesn't slow down the physics. But i think the main one is that it will keep the GPU locked onto one powerstate, sometimes it can adjust the powerstates and during the P-state change it can cause a microstutter.

 

It cant be a cpu bottleneck. Even on max settings my Ryzen 7 1700 is only around 15% usage. And this guy has a 2700x, which is a little better than a 1700

 

You have got yourself bit confused. CPU usage tells about use of all cores and thread of CPU.

There are very few games that actually can use all of CPU, usually it is 1 thread on 1 core that creates CPU bottleneck.

More cores you have, more likely it is that you can't use them all in games.

Hence it is more important to have fast single core performance, especially with more cores that becomes very important.

 

go on task manager, then performance, right click, change graph to logical processors. One core might be maxed out causing issues to all the others.

 

To add to that, sometimes 80% can be maxed out, task manager is too slow to register all super quick spikes and if you have like sawtooth spiking really quick, you get only average of those, while still each spike to 100% is adding to wait for CPU.

Pressing ctrl-f ingame and seeing wait for GPU gives good indication if CPU is limiting or if GPU is limiting, numbers above 1 or so are GPU limited.

 

What CPU seems to be the best at single core performance then? I got my new Z mobo today so im going to be overclocking my 6600K to about 4.3 or 4.4...how do u think that will work? And I got a 1070 coming in the mail this week as well, aGigabyte 1070 Windforce 8gb. Im hoping with that CPU/GPU/OC combo the game will STOP MICROSTUTTERING lol I manage about 60fps on most maps, but constant microstuttering.

 

Keep SSAO off, run game from fast SSD, make sure never drop below 60fps and have vsync ON, fps limiters off.

6600K should be better at stock clocks than Ryzen 7 2700X stock clocks, with overclock you could get to be better than Ryzen can be, but when running many vehicles, that is where Ryzen really shines.

You want best? i9-9900K overclocked to oblivion, ignoring chip life and all that, you can get insane single core speeds with that, while still having enough threads to run as many cars you need.

There is of course HEDT intel chips then, you could get ridiculous amount of threads, but single core performance of those is not more than what you currently have.

Making sure CPU never peaks is the key to smoothness, so weak GPU can be lot smoother than faster, despite less fps.

If you have 144Hz monitor, run it at 60Hz for BeamNG.

Of course running full screen and choosing vsync half refresh rate from nvidia control panel will do fine too even it would be only 30fps

Oh and if Windforce has only 2 fans or 1 fan spinning at boot, there is loose cable inside fan unit, easy to fix.

 

I have a SanDisk z410 SSD as my only drive in this computer, do you reckon thats a decent hard drive? (pretty much the only thing I havent touched since I built the computer 2 years ago) I feel like its def fast enough, my computer boots up in about 4 seconds from splash screen to windows login
--- Post updated ---
https://pcpartpicker.com/list/n4D9LJ

thats my PC currently, as is, (when I get around to putting the new MoBo in) and the case is not listed currently, so that and the fact that the Z170 I have is not listed with a price gives you an idea of how much the computer I have is worth...Altogether, I am into this PC for about 700 bucks, if not less lol not bad I'd say

 

I think that your SSD leaves some slight room for improvement:
https://www.tweaktown.com/reviews/7726/sandisk-ssd-plus-z410-sata-iii-review/index.html

M2 Nvme if you want fast performance, but that is easily costing 100 for 256GB, some nice 2.5" SSD like MX500 or Samsung Pro model or similar might do bit better job, however I'm not sure difference in practice are that huge.

I have one of the fastest in practice NVME units, SM951 from Samsung, even today it still outperforms many newer bits. Intel advertises Optane being super fast thing, but SM951 outperforms it, only in 4K random read Optane is faster.

Difference in practice just is not so huge really, some tasks NVME handles lot faster, like huge file transfers, but practical differences are not numerous times faster.

 

hmm, so perhaps ill be upgrading the SSD along with the RAM then...time to stalk EBAY again lol (I have been getting EXTREMELY lucky with prices and reliability thus far on everything I have bought, even making money on each item I sell lol)

 

You brought up something I never took into consideration, storage. I run my games off two WD Blue Drives ran in raid 0. I don't think they affect my game, however, I also noticed that maps are compressed and will not be recognized if they are not. Maps are located in Content>levels. I am not sure, but slow storage could cause micro stuttering as the maps might still be trying to unpack? If anyone knows how to make the game recognize the uncompressed files of maps it would be interesting to test how this affects fps, as I imagine it does.