This game's performance...

[KITTYONFYRE]

What's wrong with this game? I make a 600 part monstrosity and I get 7 FPS. Okay, understandable. I look at my GPU usage - 30% at most. Okay. I look at my CPU usage. 30%? What the hell? Why am I getting 8 FPS when my CPU is at 30% (accross all four cores)? It's not even one core getting slammed - all four cores are hovering around 30% usage. When I'm in the spaceplane hanger, it's 60% usage and smooth sailing. I then start playing and usage goes DOWN? What the hell is this game's problem? It's happening to me and my brother so it's not my computer.

i5-4690k @ a modest 4.3ghz

r9 290

8 gigabytes ram

win10 64-bit

my brother has a 280x and a... 3570k I believe? at 4.6 ghz. having same problem.

[Majorjim!]

What's wrong with this game? I make a 600 part monstrosity and I get 7 FPS. Okay, understandable. I look at my GPU usage - 30% at most. Okay. I look at my CPU usage. 30%? What the hell? Why am I getting 8 FPS when my CPU is at 30% (accross all four cores)? It's not even one core getting slammed - all four cores are hovering around 30% usage. When I'm in the spaceplane hanger, it's 60% usage and smooth sailing. I then start playing and usage goes DOWN? What the hell is this game's problem? It's happening to me and my brother so it's not my computer.

i5-4690k @ a modest 4.3ghz

r9 290

8 gigabytes ram

win10 64-bit

my brother has a 280x and a... 3570k I believe? at 4.6 ghz. having same problem.

I will look into this, are you using cpuz?

[mythbusters844]

Because it's unity, not us :^)

But seriously, the Unity 5 update will greatly help KSP's terrible optimization, unless Squad screws up. But even without the upgrade, the game seriously needs tons of performance fixes. God, it's 1.0 and the game runs even worse than when I first bought it in .20.

Also, getting a faster processor or graphics card won't necessarily help with your problem. Even with the best of the best, KSP is too unoptimized and a large craft will lag your game.

[KITTYONFYRE]

Because it's unity, not us :^)

But seriously, the Unity 5 update will greatly help KSP's terrible optimization, unless Squad screws up. But even without the upgrade, the game seriously needs tons of performance fixes. God, it's 1.0 and the game runs even worse than when I first bought it in .20.

Also, getting a faster processor or graphics card won't necessarily help with your problem. Even with the best of the best, KSP is too unoptimized and a large craft will lag your game.

It's not even that, it's that it just doesn't take the power available to me. I have PLENTY MORE POWER, and it doesn't take it.

Damn thing still sounds like a leafblower :/

I will look into this, are you using cpuz?

My brother was using HWID, I was using afterburner.

[*Aqua*]

Why am I getting 8 FPS when my CPU is at 30% (accross all four cores)? It's not even one core getting slammed - all four cores are hovering around 30% usage.

It's just the one thread that really does a lot of calculations: the physics thread.

And then there's Windows which pushes a thread from one core to the next. It's the game programmer's responsibility to decide between the default Windows behavior or instead locking a thread to a specific CPU core. Locking might look like a good idea but on the other hand it also limits the ability of the OS to react on varying workloads. (Instead of pushing a single performance-eating thread to an unused core Windows then has to move a dozen of processes.)

[KITTYONFYRE]

It's just the one thread that really does a lot of calculations: the physics thread.

And then there's Windows which pushes a thread from one core to the next. It's the game programmer's responsibility to decide between the default Windows behavior or instead locking a thread to a specific CPU core. Locking might look like a good idea but on the other hand it also limits the ability of the OS to react on varying workloads. (Instead of pushing a single performance-eating thread to an unused core Windows then has to move a dozen of processes.)

No, it doesn't. It doesn't do jack .... with my cores. ALL of them are at 30% - there's not 1 thread at 100%, it's all at 30%. There's no bottleneck in single core performance. all of them are running the same. It's stupid.

[DMagic]

No, it doesn't. It doesn't do jack .... with my cores. ALL of them are at 30% - there's not 1 thread at 100%, it's all at 30%. There's no bottleneck in single core performance. all of them are running the same. It's stupid.

That's not how it works.

Windows, or any other OS, constantly balances the load across all available cores. During normal use you should never see one core at 100% with the others at 0%.

And 1 core is not the same thing as 1 thread, but that's a different discussion.

[marach]

No, it doesn't. It doesn't do jack .... with my cores. ALL of them are at 30% - there's not 1 thread at 100%, it's all at 30%. There's no bottleneck in single core performance. all of them are running the same. It's stupid.

Windows is moving KSP around to load balance your cores (potenially thousands of times a second) keeping usage down overall

[KITTYONFYRE]

Windows is moving KSP around to load balance your cores (potenially thousands of times a second) keeping usage down overall

So it's just KSP being stupid. I see.

That's not how it works.

Windows, or any other OS, constantly balances the load across all available cores. During normal use you should never see one core at 100% with the others at 0%.

And 1 core is not the same thing as 1 thread, but that's a different discussion.

Yeah I meant to say core, not thread. Woops. I know the difference.

[marach]

So it's just KSP being stupid. I see.

No rather PhysX being dumb as a post

[Rocket Farmer]

i5 is your issue. I'm running i7, 16gb of ram and a solid state hard drive on a laptop and 1,000 part monstrosities purr along at 30 fps.

[RocketPilot573]

i5 is your issue. I'm running i7, 16gb of ram and a solid state hard drive on a laptop and 1,000 part monstrosities purr along at 30 fps.

Just my 2 cents, I don't know everything, but I'm fairly certain that some i7s would perform the same as an i5 with KSP... An i7 with 8 threads (4 cores?) can still only use 1 thread for physics, as with an i5. GHz (and other factors) would make more difference. Also, how FAST is your 1000 part ship running? Remember due to the delta-time slider setting, you can easily be at 30 FPS, but maybe running 4 or 5 times slower than real time.

Because of this, I would recommend that OP should make sure his/her delta-time slider (found in the settings) is all the way to the right (because I just have a i5 2.8 GHz and still get 30 FPS with 700 parts, slow motion of course)..

[DMagic]

i5 is your issue. I'm running i7, 16gb of ram and a solid state hard drive on a laptop and 1,000 part monstrosities purr along at 30 fps.

There is little difference in performance between otherwise similar i7 and i5 CPUs. Those CPU's don't have any significant difference in single-threaded performance, which is what limits KSP.

[Majorjim!]

i5 is your issue. I'm running i7, 16gb of ram and a solid state hard drive on a laptop and 1,000 part monstrosities purr along at 30 fps.

Err.. nope. Video or it didn't happen. I have an i7 16gig and KSP on a speedy SSD. No chance a 1000 part ship runs at 30fps. Not even in version 0.90 and slider all the way to the right.

[Laie]

I will look into this, are you using cpuz?

For what it's worth, I notice the same (low performance but no core at max). Linux. Looking at gkrellm for CPU stats.

Incidentally, has the application any say in how local a thread is? If I start just any random task (like compressing random data), it's switching cores perhaps two or three times a minute. KSP doing the round-robin so quickly that all cores appear mostly idle would be very unusual... doing it like this would also impair performance, I guess.

[Columbia]

i5 is your issue. I'm running i7, 16gb of ram and a solid state hard drive on a laptop and 1,000 part monstrosities purr along at 30 fps.

>1,000 parts

>30 fps

tfw 100 parts and 25 fps T_T

In all seriousness, how.. is that even possible?!

[KITTYONFYRE]

i5 is your issue. I'm running i7, 16gb of ram and a solid state hard drive on a laptop and 1,000 part monstrosities purr along at 30 fps.

This is wrong. You're probably not familiar but a laptop mobile i7 isn't meant to be a gaming powerhouse. It's FAR weaker than my i5. My i5, being desktop variant and clocked at 4.3 ghz, will easily outperform your i7 (per core) which is probably clocked under 3 ghz. Considering KSP apparently uses 1 core only, your extra 4 cores won't help at all. I'm also a hundred percent sure that the last part of your sentence is wrong - I'm running the default delta time settings, which I forgot about. You've probably edited them.

[CalculusWarrior]

>1,000 parts

>30 fps

tfw 100 parts and 25 fps T_T

In all seriousness, how.. is that even possible?!

If you have pleased The Gods of KSP Optimization, they will bless you with High Framerates and Large Crafts.

In all seriousness, it's all a matter of the hardware you have (higher per-core performance is better than a whole bunch of cores), as well as Unity being weird. We can only put our faith in the Unity 5 upgrade to smooth things out, though from what I've heard, there's quite a bit to be excited about!

[SAI Peregrinus]

Did some testing. Fresh KSP install via Steam. Added GCMonitor mod only, to have a handy FPS counter. Changed settings: Turned off Vsync, set resolution to 1920x1080, -popupwindow.

CPU is an Intel Core i7 4790k @ 4.4GHz. GPU is an AMD Radeon R9 290X.

Max CPU temperature during run: 65C. Mean temperature: 56C.

SPL during run: 38dBA @ 1m, though this is difficult to measure because I'm next door to a convenience store and their air conditioning is louder than my PC under load. 4 cores/8 threads @ 100% runs to about 41.5dBA.

Affinity limited to a single core.

Javascript is disabled. View full album

Clearly it's nearly maxing out the one thread. 100/8 = 12.5. Remember that hyperthreaded processors will list each possible thread as a full core, so 100% use of one core by a single thread is 12.5%. I'd not limited all other processes to avoid that core, so it couldn't quite reach 12.5%.

.craft, though be warned: it has 2337 parts.

The FPS counter lies a bit. Since it only samples at intervals the FPS can dip without showing on the counter. Still, it was reasonably close to accurate for most of the time. If I cut off 1000 parts or so from the thing I could easily see it getting 30 FPS.

OP, if your PC sounds like a leafblower you just need a better designed cooling system. Mine is air cooled, no need for water cooling, pumps tend to be noisy. Just a high-end very quiet case and set of fans with properly placed baffles to direct the airflow and reduce turbulence. (Antec P183 case, all fans replaced with SlipStreams and GlideStreams from Scythe, Scythe Mugen 4 heatsink.) Unless you like the leafblower sound, of course.

Edited August 29, 2015 by SAI Peregrinus

[ainurakne]

Incidentally, has the application any say in how local a thread is? If I start just any random task (like compressing random data), it's switching cores perhaps two or three times a minute. KSP doing the round-robin so quickly that all cores appear mostly idle would be very unusual... doing it like this would also impair performance, I guess.

Compression is usually one continuous process. So, if there aren't many other running tasks that demand lots of processor time, your compression task is probably rescheduled only if the core it runs on starts to heat up. Hence the switching two or three times a minute.

Rendering and physics threads usually run in cycles (for examples 60 cycles per second) with somekind of sleep or yield (or somekind of synchronization) between the cycles. So, they are rescheduled after every cycle anyway. Depending on how much other stuff currently runs on the processor, how many cores there are and how hot each core is, rescheduling will most likely also cause core switching.

[rkman]

Did some testing....

Affinity limited to a single core.

From my experience KSP really does not like running on only one core. Affinity on two cores makes it run a whole lot better.

[SAI Peregrinus]

From my experience KSP really does not like running on only one core. Affinity on two cores makes it run a whole lot better.

Yes, I agree. But the OP wanted to know if KSP could actually max out a single core, and the answer is a definitive "Yes." Even though the physics is single-threaded the game as a whole is multithreaded, so it certainly helps to add an extra core or two. It certainly can't max out all the cores of my CPU, but it can get a full (virtual) core to max.

[montyben101]

The games physics is kinda stupid.

[cantab]

No, it doesn't. It doesn't do jack .... with my cores. ALL of them are at 30% - there's not 1 thread at 100%, it's all at 30%. There's no bottleneck in single core performance. all of them are running the same. It's stupid.

As mentioned, KSP hammers the equivalent of one and a bit cores, thanks to being dominated by a single thread. Windows likes to switch the thread between cores. The last piece of the puzzle is that the CPU usage you see is an average over a period of time long enough that KSP has been switched around the cores, so instead of seeing different cores briefly at 100% you see them all around 30%.

The good news is there's no significantly better processor for KSP than what you have. The bad news is there's no significantly better processor for KSP than what you have.

Do be mindful that mods can impact on game performance. I know one person running a bunch of mods and getting framerates far below what I feel their hardware should manage, based on comparison with my own inferior hardware that performs better.

And one last thing, don't trust the fps meter in the game, it gives wrong values at low framerates. Use a mod or an external program instead.

[Renegrade]

Eh? What's this nonsense about Windows scheduling threads on different cores to optimize load?

Anybody with any SupCom experience knows that Windows schedules threads using the 'worst core possible' algorithm. Basically, it looks for the busiest core and moves all the work there, unless it can detect that there's other cores available that will force a cache flush (like the old Core 2 Quads), in which case it swaps all threads between the two as fast as possible.

Also - anybody not getting why multicore performance is almost universally failing to deliver should spend some time and learn about things like Amdahl's Law and it's consequences.

(the best way to optimize KSP's performance is not multithreading, it's part welding and physics simplification. Reaching for a bigger hammer isn't always the right answer -- it's usually almost always the wrong one in fact -- simplifying the problem is usually much easier, cleaner, and neater.)

OP, if your PC sounds like a leafblower you just need a better designed cooling system. Mine is air cooled, no need for water cooling, pumps tend to be noisy. Just a high-end very quiet case and set of fans with properly placed baffles to direct the airflow and reduce turbulence. (Antec P183 case, all fans replaced with SlipStreams and GlideStreams from Scythe, Scythe Mugen 4 heatsink.) Unless you like the leafblower sound, of course.

The Antec P183 is a beautiful case. I have one for my fileserver - whisper quiet. I have the low-static-pressure case fans from Noctua though (S12Bs) instead of Scythe fans. My only complaint is that it's kinda on the small side of things, it's otherwise a superb case.

(I used to use crappy, $40 cases in the old days, but got tired of the damn things biting my fingers and having crappy, thoughtless layouts and poor airflow)