BeamNG and CPUs

What is better for the game: fewer, higher clocked cores or more, slower cocked cores?

Not sure if this has been answered before.

 

Quad core with as high as possible clock speed is optimal, such as a 7700k overclocked to 4.8ghz

 

The short answer is "It depends".

Do you usually just have one car spawned? If so, then higher clockspeeds.

Or do you like to load up a car hauler full being pulled by the T75? If that's the case, then more cores.

I don't remember the exact numbers of threads initally used by the game with only one car loaded, but each additional loaded item, like a car or trailer, cargo, etc. will use an additional thread if it is available to use, if I'm not mistaken. For example, an overclocked G3258 will probably run great with just one vehicle, but will fall flat on it's face once you start adding more, while a stock clocked 5960x may not do so hot with just one car, but it will take a lot more to bring it to it's knees than the overclocked G3258.

Of course, having more higher clocked cores is always a plus... Hopefully that answers your question.

 

Alright, thanks.

 

I think that you have it put very well there.

Single core performance is something many mod maps need in current version, because of shadows and number of mesh objects in map.

 

Really? Wouldn't 6 or 8 cores run better, especially with 4+ vehicles active with AI?
My next build I was considering a 6 or 8 core intel because of this. So tired of being jelly from watching all these 60fps videos with 8 vehicles all with AI

 

As has been said earlier, it depends.

For multiple vehicles cores are a must, but for shadows and some other things you need each core to be strongest possible so that you can run any map with shadows on.

I'm hope that I could move to 6 to 8 cores workstation build soon, but my i7-6700 is often limited with single core performance, for that I think that changes in BeamNG might be providing help in future, so while currently with single vehicle it is more important to have high single core speed, it might change in future so that workstation build would be worthwhile as it would allow more vehicles.

6-7 vehicles is where my fps still remain at constant 60fps, but it starts to drop quickly after that, my cpu has 4 cores and total of 8 threads, CPU with 6 cores and 12 threads would certainly help with massive crash scenarios, while with fps drops with single vehicle it probably won't do anything good.

 

With 4+ vehicles with AI yes, in other situations no.

 

go for AMD Zen, since it will be cheaper, you could also get an octacore too

 

Quad core is perfectly fine, I have a nonoverclocked i7-4790 @3.6 GHz and can run 5-7 ai cars at ~50 fps.

 

While I agree a quad core is well up for the job, the 4790 hyperthreads so you technically have 8 cores.

 

Hyperthreading doesn't just double the amount of cores, that's not how it works

 

Try 720p windowed, that is what I use most of the time, with 27" 1080p monitor it is not too small, but much nicer when doing stuff like using Notepad++ while beam is running.

Most of the time I'm CPU limited with i7-6700, because of single core performance, with your overclock, limit might get shifted quite a bit higher, I would say improvement is roughly the same what cpu load I get from shadows in East Coast USA.


For anyone interested about effect of resolution, I did some testing yesterday until my shoulder decided I have to stop again before finished.
85fps 1080p on high gpu limited and 116fps 720p CPU limited, GTX 1050 Ti on East Coast highway spawn D15, UI off, hood view, no eye candy boxes checked.

Between windowed and full screen there seem to be ~2fps difference.

Some maps require more CPU, some more GPU, Glass view mountain mod map requires both and at places is very heavy on CPU, not sure why, I suspect number of objects does something, usually buildings, constructions, that contain many objects is what seem to require really a lot of single core performance, while lot of trees like in East Coast require more of GPU.

With many vehicles my GPU starts to wait CPU, MSI afterburner shows GPU usage to fall down to 16% or so and I don't need to have 100% CPU load on any of the cores for that to happen, CPU just takes too long to give information to GPU and GPU usage falls down.

GPU would not be hitting 100% without being able to render at 60fps, if CPU would compute enough fast, but what happens is that CPU chokes, most vehicles I have tested was 11 I think, there just is too much to compute for CPU so that GPU could do more, but has to wait CPU, with removing vehicles, fps might not then increase, but gpu usage increases, things have quite interesting relationships when more is investigated.

I might have something around 30 hours of performance test now done on this BeamNG version, there is still more to test and of course I'm limited to only hardware which I have, but it is quite obvious that adjusting gfx level just a little bit down, it is quite easy to get BeamNG to run fine what comes to limited GPU power, but with more complex and bigger mod maps there is no substitute of running without shadows or high overclock.
--- Post updated ---
Too many cars that I can count to, I have only barely 10 fingers, GPU usage 58%, well true it is just a pure grid but wanted to test maximum number of cars, I added one more and BeamNG crashed after hitting Ctrl+E without showing the menu

 

My apologies. Thank you for letting me know.

 

Four threads have around the same amount of computational power as one physical core. So having a four core CPU with hyper-threading is like having five physical cores essentially. Very simplified explanation however it's rather accurate.

 

I've never heard this before, could you provide a source?

HT improves the efficiency of the available cores, and the efficiency improvement depends heavily on the workload. Unthreaded and poorly threaded workloads (same number of threads (or fewer) as physical cores) will never take advantage of the extra threads, because they will never use the extra threads. Heavily threaded workloads will take full advantage of the extra threads.

 

Something I heard Jim from AdoredTV say in one of his more recent videos though I can't specifically say which one. It's highly variable so the statement is not particularly precise but it's a suitable baseline measurement. Something that's easily substantiated by looking at a few synthetic benchmarks (best case scenario) and performing some extrapolations. For Intel it tends to be more similar to 1.5 physical cores while AMD currently suffers in that regard.

 

A source is a specific video, preferably with a timestamp. Not a vague statement that gives no relevant information.

Synthetic benchmarks cannot be compared to real world results.

What??? Your numbers don't match up at all.

 

Efficiency of intel HT in *most* benchmarks is 53%. Namely making 1 hyperthreaded core equivelant to 2 cores at 53% of the clock speed.
Having a 4 core hyperthreaded CPU is most akin to having an 8 core CPU running at half the clock speed.


Certain tasks can reach more efficiency, some actually receive 0 efficiency.