Game/windows/Graphics settings for KSP

[Talavar]

Not sure if this is in the right place, but I was wondering if someone could give light on if I should be running with CUDA, or if the processors should be handling most physics calculations.. Don't get me wrong.. I have an impressive rig, But I tend to build ridiculously massive part-count ships. and it can drop me down to the red sometimes.. I have a gtx 660, and a self clocking 3.9 - 4.1 Quad-core processor. Should I be using CUDA to get the maximum capabilities, or should I be letting the processor do all the calculation work, and let the GPU just handle graphics? Just wondering what is the best set-up. If anyone has some insight simply from experience or trial and error, any input would be appreciated.

[Guest]

You won't be able to change anything at all, i don't know how you plan on doing that.

All physics calculations are done on the processor, and graphics are rendered using the GPU. This can't be changed.

Also, "Quad Core processor" is not enough, there are some awful quad cores.

If you'd like some performance tips, list your specs accurately =]

[LaytheAerospace]

Not sure where you got the idea that game physics are accelerated by CUDA. At any rate, your GPU doesn't factor into physics performance. That runs on the CPU. Also, keep in mind that the physics will only run on one thread, so more cores counts for nothing.

No matter how good your system is, you'll have a part count ceiling. I run a 3770k at 4.9 GHz, and can comfortably handle ships up to about 1000 parts. An i7-950 might max out closer to 500-600. An old, overclocked AMD Phenom would probably struggle before even that. AMD chips, unfortunately, have extremely low IPC, and single threaded performance is much lower than Intel chips at the same clock speed.

So you really can't get a feel for how fast a chip is just by saying how many cores at what frequency. You need to post the actual model number.

Edited July 31, 2014 by LaytheAerospace
Added CPU-z link

[Pecan]

Answers above - it comes down to the fact that KSP is written using the Unity engine and Squad and you can only optimise within what that provides.

[LaytheAerospace]

edit: wrong thread

[Renegrade]

No matter how good your system is, you'll have a part count ceiling. I run a 3770k at 4.9 GHz, and can comfortably handle ships up to about 1000 parts.

Er..what's your definition of comfortably? If it's > 20 fps, please let me know your settings/optimizing tips.

I find it rather chuggy at 400 parts (uncomfortably so) on an i7-3820 @3.8ghz.

Or I did in 0.23/0.23.5. I haven't made a 400 part ship in 0.24 as that would be.. wasteful of funds. That and BTSM isn't 0.24 ready yet.

[_Aramchek_]

Comfortably likely means 30+ fps.

[LaytheAerospace]

I'd have to benchmark it, not really sure. It's butter smooth with 600 part count ships, which are pretty common for me. My 1k+ ships were probably in the 10-20 FPS range. Noticeable stuttering, but these were heavy launchers that dropped most of their parts very quickly, and smoothed out.

One neat trick for making the game run more smoothly when you're up against the part limit is a speed hack. Use something like Cheat Engine to hack the speed to 50%. Unless you're really, really overloaded, it tends to speed things up, as well as make the game run more smoothly. You will of course be limited to at most half speed, but half speed at 30 FPS is better than full speed at 5 FPS (estimate, don't quote me on framerates).

The best thing you can do to increase your part cap is to overclock. Your 3820 isn't the best overclocker in the world, but it's definitely very good. The multiplier is partially unlocked, and the bclk can go all the way up to 125 with good silicon. I think they'll go as high as 4.7 GHz, if you push them hard. Liquid cooling helps, but isn't necessary. Both my 4.95 GHz 3770k and my 5.33 GHz 2500k use a $30 air cooler. Gets a touch warm (95C) when I load up all the cores, so I tend to not do that much. Within safe thermal limits, but with the voltage I'm pushing through them to reach those clocks, there's definitely electromigration going on, and I'd like to minimize that.

[Renegrade]

One neat trick for making the game run more smoothly when you're up against the part limit is a speed hack. Use something like Cheat Engine to hack the speed to 50%. Unless you're really, really overloaded, it tends to speed things up, as well as make the game run more smoothly. You will of course be limited to at most half speed, but half speed at 30 FPS is better than full speed at 5 FPS (estimate, don't quote me on framerates).

..what? Does that slew the realtime clock or something?

There's also some physics delta setting in KSP's menu, and people have mentioned in passing that it can affect performance, but I've never found out the why or how, so I've left it alone.

The best thing you can do to increase your part cap is to overclock. Your 3820 isn't the best overclocker in the world, but it's definitely very good. The multiplier is partially unlocked, and the bclk can go all the way up to 125 with good silicon. I think they'll go as high as 4.7 GHz, if you push them hard. Liquid cooling helps, but isn't necessary. Both my 4.95 GHz 3770k and my 5.33 GHz 2500k use a $30 air cooler. Gets a touch warm (95C) when I load up all the cores, so I tend to not do that much. Within safe thermal limits, but with the voltage I'm pushing through them to reach those clocks, there's definitely electromigration going on, and I'd like to minimize that.

I did some testing when I initially got it - I tested around 4.2 to 4.3 with very low temperatures and good stability with air cooling (Noctua NH-D14 SE2011), but I returned it to stock speeds afterwards as I generally keep a processor in service for a good fifteen years...and my prior system, built around an i7-920, was alarmingly hot.

Right now at stock speeds it's about 12 degrees above ambient (36c -24c ambient); under absolute maximum load (P95 in-place FFT) about 38c above ambient. I most definitely don't want to go into the 95c absolute temperature ranges ever again..

I'll give it a shot at 4.2/4.3 settings again and see how that helps (I think that was all multiplier work, but don't quote me on that, it's been a while since I've done it). I don't have any need to launch 1000-part monsters, but running more fluidly at 400 parts would be nice

There isn't any way a 670GTX could be a bottleneck, is there? My research all points towards KSP being CPU-bound..