CPU single core load gets bit high at ECA

As you can see from graph below, I'm getting drop at GPU load, that is because CPU thread spikes very quickly (not visible on graph) and GPU has to wait CPU to process data, hence drop in GPU load:


i7-6700 and 1050Ti, CPU should be able to feed that GPU quite easily, however as can be seen, there is this something still around that tends sometimes to lead lower performance than excepted.

Just wanted to bring this to attention.

Oh and compared to 0.8 game version, result was 2.5fps less in current version, that is 72 vs 69.5fps.

 

Attachment not found. How are you detecting the spike in CPU usage?

 

Now what the devil that picture goes to disappear? Well I put it again there, maybe it stays this time.

With 1000ms polling interval it is not possible to actually measure the spike of such fast kind, even with 250ms that is bit hard, however
performanceCounter1.NextValue() each 50ms is showing those fast spikes of 100% single core usage.

That is my own program to show realtime CPU usage. There is direct correlation of needle staying at 100% and GPU load starting to drop.


Also I did test ECA 3 times and each time GPU load drops at same spot, with CPU load getting high. No thermal throttling of any kind as usual.
For more comprehensive description of testing methodology you can check first post of this thread https://www.beamng.com/threads/your-performance-testing-results.37866/

Next week I should get another GPU to test to, it will be interesting to see what difference I will find.

Update: Just to elaborate bit more how my program works, it checks performanceCounter1.NextValue() with 50ms intervals stores that in array, then each 250ms finds highest value from that array and updates the needle, so it will be very good at spotting peak core usage, rapidly spiking CPU load is what task manager etc. are not showing at all.

Really old video and Beam has actually improved a lot from 0.8 in terms of performance, but this mysterious fast spiking CPU load still occurs:


Testing different GPU's will be interesting as we should see if it is something in this GPU that affects CPU load, or if this CPU load will be actually unrelated to GPU load drop (what else it really could be?).

 

That makes sense. I haven't read your link yet (performance testing results thread). What happens if you slow the physics down?

 

That really depends a lot of things, but what I remember (might be wrong as it is really long time I examined these things more in depth), is that even pausing the physics did not cause huge change in there as it is graphics thread that goes highest at CPU usage.

My system is funny in there that graphics stays on core it starts to run, there was one bios version where code was swapping from core to another, but current BIOS version like all of them are staying put, which make it quite easy to spot if it is graphics or physics on specific core by pausing the simulation.

Now to really answer your question, I need to test ECA again and while having that low GPU load, I need to do pausing and see what happens with CPU load department.
--- Post updated ---
This is kinda weird, I don't remember pausing physics having so much effect in older versions of BeamNG:


I7-6700 being CPU bottle neck with 1050Ti? Rather unlikely, but how else would you interpret this?

I did just quick test with MSI Afterburner and I have some faint memory that it might of had some unreliability in regards of GPU usage levels, I would need to do more proper test to verify this, but for me result looks odd.

Anyway physics are not affecting #6 which is where I believe graphical aspects were running, but how to really know?

Update Oh yes, running over 60fps beam got bit hard on CPU, see how great my memory is, it can drop any piece of information at any time, which makes pretty much everything a chore and struggle.
Anyway, that just shows how 60fps is maximum I can dream of
Need to have my proper settings so GPU remains limiting factor...

Update2: More I test, more it seems to be that in 0.13 physics are responsible of that CPU 'overloading' or how one might want to call it.

This is very different from 0.8 where pausing physics certainly had only little effect.

Without car I'm getting really really good FPS with highest detail level at ECA lots more FPS that I remember seeing in older versions of the game.

Certainly lot has been improved in graphics department, but I still wonder if faster GPU will change much of FPS if physics affect so much FPS currently.
Anyway I get to borrow gtx 1080 next week as friend goes on vacation, so that should show what effect GPU can have.

 

It's pretty normal to be CPU limited in BeamNG, even with reasonably fast processors. GPU usage goes up in your second screenshot because framerate is up... yes, clearly CPU limited. (It doesn't sound like you need me to tell you that though.)

I fired up ECA and did some testing... I didn't do anything in depth (left other things running in the background etc). I also left my game in windowed mode. My initial findings were similar to yours... maybe 60fps-65fps w/ physics enabled and maybe 80fps-85fps w/ physics disabled in a 1920x1080 sized window on an i5-4670k w/ a 1060 at the Town Industrial Area spawn point.

Some additional testing showed that at least in my case the slowdown seems to be from using the "orbit" camera. Pressing Shift+C to switch to the "Free camera" instantly added about 15 FPS. Using other cameras such as onboard, external, or chase also had a similar effect (eg the FPS was much higher than the orbit camera's FPS).

 

Going downtown in ECUSA causes me to get 20fps at high settings on a fx-6300/gtx1060.

Upgrading to a 9590 is more tempting with beam getting more physics intensive.

 

Removing a car completely gives huge FPS boost for me, like close to 120fps, but this effect seems to be stronger at ECA city, maybe collisions, but wouldn't they be running without car too?

Steering wheel guide mentions how for best FFB one should have at least 60fps, but 120fps being better, considering experiments, I doubt that is happening.

However some graphics settings then help go past 60fps, if it truly would be physics limitation, there would be no change to limit by graphics settings.

That is unless some physics and some graphics runs on same core, if that is the case, getting part of them to run on some other core would help, but that might not be possible as threading works such way that certain stuff has to be run on same thread or things start breaking fast.

While performance has improved in so many ways, I think limits of what can be put on single core are challenge that dev team might wish could be elevated as it kinda appears to be greatest limitation of FPS in BeamNG right now.

It is challenging also when deciding to upgrades, it seems single core performance is becoming / is even more important, so more cores might not help much, when it is that one which is not doing enough.

6700K would be good 20% more in single core performance I believe, probably should be noticeable difference in FPS on these city blocks, but as 6700 is not exactly slow at single core performance, it is something that might get optimized later, which again makes challenges for anyone planning upgrades.

Age old dilemma, have more cores or more single core performance, we don't know the future so it is hard to tell, but at least for me it looks like that having more cores would not change much for me, I rarely have so many vehicles, but everyone's use case is different.

Certainly worth to keep eye on GPU usage drops as those can be because of single core performance is what is lacking.

 

I'm afraid that upgrading to a 9590 would do very little for you... At a glance it seems that it would add a 25% clock speed bump and 1 physical core (2 imaginary cores). If you're only using a single vehicle then 3 physical cores should be enough I think, I doubt that the extra core will help. The clock speed bump will get you no more than what it says on the box: 25% higher performance per core. Going from 20fps to 25fps isn't worth much from my perspective.... http://www.cpu-world.com/Compare/311/AMD_FX-Series_FX-6300_vs_AMD_FX-Series_FX-9590.html

As far as BeamNG getting "more physics intensive", I'm not sure what you mean by that. Performance should have been steadily improving over the past few years. Some aspects of the game still perform quite poorly, but I'd say that physics isn't what's holding most of us back - especially compared with a few years ago.

I'm not sure what AMD is supposed to be able to do, but I'd look into one of two things:

1. figuring out what's wrong w/ your current setup, if anything
2. switch to Intel or a new AMD platform.

My memory is very foggy, does disabling AMD's CMT or something help? https://www.beamng.com/threads/how-many-cores-is-beamng-optimised-for.37985/ Now I remember - I think that the heart of the matter is that every 2-core package shares a single FPU and Windows's scheduler is not equipped for that. I think that you either need to disable every other 'core' in BIOS or just disable CMP if possible (?)...

I also don't think that middle-aged AMD processors getting destroyed in BeamNG really has anything to do with the topic of this thread. We could start a new thread to discuss the matter if you'd like, although I think that if you look around for threads on this forum about AMD's Bulldozer or Piledriver architectures you'll probably find that something already exists. Feel free to @ me in another thread for more discussion.

 

No, this is relevant. This is about high single core cpu load under circumstances, and, since I am running a 3 core, I notice this more than the people running modern quad/hexa core i7s.

 

Number of cores is irrelevant. High single core usage hits same let it be 2 core or 8 core CPU, because one of the cores has to work really hard, so upgrading from i3 to i7 is not changing much of anything and if CPU is becoming bottleneck no matter what monster GPU you get will change FPS much.

There is of course some variation to that, but you running 3 core has nothing to do with it, it is how much work single of your cores can do.

Only when you saturated all your cores, more cores will change that and with single car even i3 can do pretty much same as i7.

 

Thanks for correcting me guys. I was wrong.
But what is causing the performance drop in town?

I somehow remember it getting worse after they rearranged the center of town and added the church.

But that seems graphical and not processor related.

 

I'm afraid there are no answers for that question.

I can only guess it is something with collisions and also partly graphics loading single core/thread which together then are enough high to cause CPU to be bottleneck and that probably is why changing some graphics options that require CPU processing, will help.

Same does happen on some other maps, but there really is not knowing what is causing that.

However I think that devs have some tricks on their sleeve to improve that, when time of optimizing that area comes about.

In 0.13 and perhaps in 0.12 something physics related was perhaps added to be run on that thread, which then makes it easier to run out of single core computing power. Or maybe it is a glitch, or maybe something else, all I know is that it is getting bit high and can make it difficult to keep up 60fps even with good hardware.

Next week I can borrow 1080 and can test how much difference there is between 1080 and 1050Ti. With 1050Ti light quality and shadows have had kinda big effect on this CPU usage, will be interesting to see if that happens with faster GPU too.

At ECA they have put quite large pieces of city to one object like in WCUSA to improve performance, not sure if it is bringing CPU load up or down (I would think it lowers CPU usage as number of objects seems to be part of the high single core CPU load effect), but it will increase GPU load because of less LOD possibilities.

BeamNG has so much different from many typical games, it is not really same even in this CPU load / GPU load aspect, so it takes bit more to really understand how things work and what would be best ways for a map maker for example to balance CPU and GPU loads, also BeamNG evolves all the time, so this CPU load is just something that is worth to keep on eye.

 

1050Ti with 250ms resolution, yellow is CPU thread usage that does most of the work, red is GPU usage percentage and brown up there is FPS, other cores are dimmed and at the bottom:


1080 with 1000ms resolution, same kind of drive, same settings, same map, car etc:


2-4fps difference between lowest
Not much difference of highest fps, but average

Gtx 1050Ti max min avg
92
60
69.9

Gtx 1080 max min avg
110
63
80

FPS is more 'stable' or how one could say, average fps stays bit higher, even max fps is not that much more and there is almost no difference in minimum fps as CPU is already limiting 1050Ti.

So single core performance of CPU is single most important aspect for maintaining good FPS in BeamNG.

Also difference between 1060, 1070 and 1080 is probably miniscule as my understanding is that even 1060 can do high details just fine.

Other thread where users are experiencing most likely same phenomena:
https://www.beamng.com/threads/interesting-low-performance-3770k-1080ti.56906/
https://www.beamng.com/threads/threadripper-bottleneck.56810/

I don't know if it is possible to optimize that one thread to be less loaded, but if it is, then that would open much of performance in BeamNG.

For example, here I have under 60% GPU usage and I can't keep up 60fps with single car at ECA, dynamic reflections are not maxed out, 190m and whatever high detail setting preset sets rest, graphics are high, all checkboxes (ssao etc) are checked:


With i7 6700 and gtx 1080 it is not possible to keep highest details and 60fps, even GPU would be able to, problem is single core performance.

There really is no alternative, at some point this needs to be sorted out in optimization, but it can very well be that not yet, there might be other tasks that needs to be completed first, there might already be branch that addresses this or who knows what, but I'm sure devs will fix it when it is correct time to do so.

For now, certain graphics settings need to be adjusted to reduce CPU load and to be able to play steady 60fps, shadows, reflections and shader quality seems to have something to do with that cpu thread so reducing those will be workaround.


As a conclusion, from 7534 points in Firemark to 22 645 points, I gain 3fps on lowest fps in BeamNG, average fps increases 10 and maximum 18.

That is 5% (min), 14% (average) and 19.5% (max) increase when GPU is about 200% faster.

i5 or i7 K series CPU and 1060 is what is best for BeamNG currently, spend money / performance wise.

 

The 6700 is still an excellent processor in terms of single core performance.... on Passmark it is probably spot number 20. There is only about a 500 point difference between it and 8086k.

Theoretically, you may only be able to get 120 fps with consumer hardware. (i9, Titan V)

And @torsion I would disable CMT but I bricked my BIOS by flashing it with nothing in the computer to flash onto it. It just freezes whenever I open it. Computer still works, though. Don’t know how that’s happening. XD

 

I'm not sure that it would do even that much, but maybe with overclock?

Yeah, BeamNG is currently really demanding on that single core, 144fps gaming monitor would be so much of waste with BeamNG at this moment, however devs are surely doing their best to get that load shared to other cores.

Oh and I'm so drooling after that 8086k special model, too bad I can't afford that and new motherboard + new GPU, but I'm betting 6700 will be more than enough in a future with optimizations coming in future of BeamNG, also if I find 6700k on cheap, or even Kaby K, that would give me some 40 points more in cinebench single core bench, which would be already quite close to 8086k with half the cost or less.

 

I am a bit late to the show here.
AMD FX (bulldozer and piledriver core) processors are made of modules. Each module is TWO ALU's* and ONE 256-bit FPU. When FPU is demanded by both cores, the FPU *SPLITS* into two 128-bit wide FPU's. This halves physics performance for BeamNG, on an FPU that isn't that great to begin with - maybe only 50% as good as comparable 2000~4000 series intel chips (I upgraded from an FX6300 @ 4.2ghz with 1600mhz dual channel memory, to a 4.4ghz 4790k with 2400mhz dual channel memory - both units using DDR3, back in 2014, and the 4790k was precisely two-fold more powerful).

I upgrade BECAUSE OF FPU performance differences, and the 4xxx series (from here on out - Haswell and higher) had 256-bit FPUs, double that of previous intel chips at the time.

Going from an i3 to an i5 or i7 will only help you if it's GHZ speeds are higher then the chip you're coming from, and a little boost to cache memory amounts on the chip (which helps keep the chip moving along full-throttle and not waiting on fetch and store commands via the scheduler so much). Secondly, it won't make your performance BETTER with a SINGLE car UNLESS the GHZ speed is higher than your old chip! Know this well.

RAM speed has a big impact on AI pathfinding and Physics performance (and also in other simulations like Cities Skylines, Zip/Unzip file performance - yes even in 2d games if there's enough CPU work). It's not worth tossing 16gb of ram to the way-side to buy another 16gb of faster memory, though. It's only important when you first build or buy the system, and only if there isn't a huge (20~30% or more) price difference between slower and faster memory options.

Don't bother upgrading to the 9590 FX cpu, or the 9370 (or whatever it's named), unless it's under 30~40$ and you really want to. You'll spend that much a year just feeding power to that power hungry CPU even if you only games 2 hours a night - VS your FX 6300 CPU.

Buy a used XEON workstation CPU, or a used business-class full-sized tower from one of the major brands, or whatever you can get on the Intel side of things, or snag a new Ryzen CPU and motherboard - Any RYZEN class CPU should run this game with ample FPS.

CPUs to look into:
Ryzen 2xxx* any number CPU with a B450 motherboard. x470 chipset is an option if you can afford that. CPU is Unlocked, as are all current Ryzen CPU's.
Ryzen likes FAST samsung B-die memory kits (these are code words for the chips used in the RAM sticks).
Intel 8350K i3 CPU. It has more REAL cores than the old i3 did. Unlocked.
Intel 8600K i5 CPU. It has six real cores, but no hyperthreading. Unlocked.

A Ryzen is your best bet, but the intel parts can sometimes be found on sale, and honestly, any intel 4xxx CPU or higher, that's 3.2ghz base-clock or better, will run this game well when paired with the right video card. Any Ryzen 1xxx or 2xxx (better) CPU from AMD should be fine. FX-based CPUs are old and out-dated. It's time to put that horse out to pasture when you can afford to do it, as they were quite underwhelming even when new (not knocking your machine, I had one, for 3 weeks, that was as long as I could tolerate it). A Ryzen 2200G, (and 2400G) are both decent choices, with the 2200G remaining a value king with a half-decent integrated GPU for when push comes to shove and your graphics card has unexpectedly gone to the PC chassis in the sky, at 99$ or less. The 2400G is only a good choice if you don't want to buy a graphics card in the near future, as it's 169$, though you get better GHZ speed (and two threads per core) by a bit and a more capable integrated GPU.

* When building a Ryzen 2xxx-based PC, if you buy a 3xx series chipset motherboard with it, you must update the bios before booting the computer using a 1xxx series CPU. Opting for a 4xx series motherboard will help prevent this issue. Boot-up-Kits are available on loan from AMD, however, to ratify this issue.
*A core in traditional computing terms since the 486 DX days, meant an ALU and FPU pair, being a complete package, but this is only false assurance, as a CPU core itself is one ALU, FPU is not required to advertise as a CPU CORE.

Terms:
FPU = floating point unit - this is where all non-medically precise (float) numbers are taken care of. AI and Physics data (not just in BeamNG) is handled here.
ALU = Arithmetic Logic Unit - there is where all your exact numbers and such are processed - basically everything but Physics and AI. A cpu with 8 ALU's would be considered an 8 core CPU (not THREADS).
Threads = the amount of simultaneous things your CPU scheduler can task at once.
Hyperthreading or SMT (same thing for arguments sake!) enables the scheduler to read ahead a bit and get the CPU ready for the next task the millisecond that it's done with whatever it's currently working on. This shows up as having 'double' the amount of real cores (on 98% of CPU's with it) to the operating system. This allows the Windows scheduler to (normally) address it specifically, and enables HT or SMT to work as intended under normal circumstances.
Hyperthreading helps reduce hitching and small FPS drops when you're keeping the CPU very busy especially with not 100% sustained, even work loads (like max amount of cars in BeamNG with web browser in the background, or playing Cities Skylines on a HUGE city - with lots of mods - and having Winamp or Windows Media Player playing music at the same time, without HT or SMT, your music might click, stutter and pop or the game might lag more).

Intel chips are a bit faster because most of them clock higher than Ryzen when overclocking manually (if you know what you are doing), however, the intel chips are also quite a bit more expensive. So get what you can afford, as all the CPU's released in the last two years marketed for gaming or content creation should be viable for years to come. This article was written with an emphasis on budget.

I hope this helped!

 

Yes, this sure did. However, I am cheap, and after a decent amount of time on Craigslist, I found a rig I could build for about 400 usd.

CPU: i7-870
RAM: 20gb DDR3-1533
GPU- Nvidia GTX 1070

I don’t know the full specs at this point, but is it worth buying at this price point? Later on down the road I could move to newer haswell (although I have no idea how it plays with low speed ram) or broadwell chips (with a new motherboard, of course). Would this be a good build overall at that price point?

 

Is that socket compatible with Haswell? If it is, then putting 4790K there would make that to be quite fast even by today's standards, 1 year ago 4790K was still at the top what comes to BeamNG CPUs, imo.

It has slow ram, but it has lots of it, which is good, also 1070 is good even for WCUSA.

I don't know about i7-870 though and I am not sure if it is compatible socket with 4790K, but if such is a case, then it would be good path to take, imo.

CPUs for BeamNG, I think that K series 4790K, 7700K and 8th gen, 8600K 8700K etc are best ones, but none is budget solution, not sure of used 4790K, however 8th gen are especially nice for BeamNG as those have fast single core, but yet also numerous of threads.


Now, if money is not limiting...

8086K is best for playing BeamNG at the moment, but only 50 000 made and price is really steep, however not much comes to close of that @ 5Ghz.

If one likes to make videos and use really lot of vehicles, then AMD Threadripper, just make sure it is model with fast single core speed, web browsing still needs single core speed too, JS etc. are not much threaded, sadly.

For memory, G.Skill Trident Z 3200Mhz CL14 is pretty fast and is not overly expensive, 'only' 230 euros for 16GB, but that kit delivers performance, on userbench it is very close to top. There are of course some other memory kits that might be faster, especially when overclocking, but some crazy fast kits cost insane amount of money and there is not much performance to gain over that kit.

Also when choosing motherboard, pay attention to M2 slot location, most Nvme units perform far better with some airflow and heatsink installed, but many motherboards have M2 slot(s) placed so that you can't get good airflow to M2's location.

With 1080p even 1070 is handling high graphics without a sweat, only if you like to do big burnouts, you will like that 1080Ti or even top of the line 2080, I can tell that 1080 likes to lag with big burnouts.

This all is for keeping solid 60fps and never dropping below.

Fast single core is needed, even on empty grid map my 6700 can't feed 1080, even my 6700 is scoring 121 points in SC test of userbenchmark and gets bit over 40Mbeams/s with 1 car in Banana bench (fast 8th gen gets over 50Mbeams/s, I have seen 8600K getting 41Mbeams/s)

Yes, that is over 900fps on high graphics, can't get much over 70 at slow spots of WCUSA though

Most of the time, with single car or with two cars, multicore performance becomes pretty much meaningless in this game. To be able to drive against AI in WCUSA, one needs really fast single core.

I have only 2666Mhz CL15 DDR4, so sure that is bit slower than fastest possible, but it is Hyper-X, which is performing quite well for those numbers.

 

4790 is 1150. 870 is 1156.