Hardware vs. part count

[WhiteWeasel]

The game runs decently for me: AMD 5800K APU 4.2 Ghz.

[ouion]

I got a AMD Athlon X2 Dual core 2.1 Ghz QL-65 and 2 logical Processors

HD 3200

4 GB RAM

It runs perfectly

[Colonel_Panic]

Reporting back, as promised. With a totally new installation (vanilla settings all the way) of 0.20 I am getting about 40 fps on the runway and 25 - 35 in flight (or an attempted flight, as FAR is not installed ) Some rare dips to 20, but those are short lived ). This is the same Windows installation as before, just KSP has changed. Quite a major difference with the 8 fps as before, which confirms my suspicion that the mods were a major contributor to the relatively low frame rate.

However, still no maximally loaded CPU and/or GPU. CPU load is about 35-40% (again, pretty evenly distributed over four cores) and GPU load is about 50%.

KSP does not use SMP. It is limited to one CPU thread, so will only use the equivalent of one CPU core.

[John FX]

I get best performance on my I7 ([email protected]) by giving KSP cores 4-8 and high priority and letting the system run on cores 1-4. I have plenty or RAM (24Gb@1660) and a GTX 570. I also run KSP from a freshly booted system as there seems to be memory issues possibly due to conflicts with other installed programs on my PC

As people have said this game does not scratch the surface of my GPU. I can (with 0.20) launch 400 part ships with hardly any lag (much better than 0.19) and in orbit a 400 part launch runs smooth as silk. Launching 800 part ships is ok but starting to get fairly laggy but even these are ok when in orbit and you have lost some parts and the ground.

I would say, if you want optimum performance from KSP then give your system the RAM it needs to be able to not touch the 4Gb KSP needs and give it its own core to completely control at least (As I say I find giving it four is best for me for some reason)

Hope this helps.

[Camacha]

KSP does not use SMP. It is limited to one CPU thread, so will only use the equivalent of one CPU core.

Yes, well, that has been the mantra over and over. The evidence does not entirely seem to support that statement. At least some parts of the calculations seem to get offloaded to different cores. If your statement were true, KSP would never use more than 25% on a quad core, but that does actually seem to happen.

[jfx]

Your OS switches threads (and yes - a single thread too) to different CPU cores as it sees fitting dozens of times a second (Windows time slice length is about 20ms). There is nothing you can do to prevent that and there is nothing that you should try to prevent that. Because of that you see an 100% Load (limiting) Thread appearing as roughly 25% load on four cores as you can see here in this retarded example:

http://i.imgur.com/frK6kDo.jpg?2

The amount of myths and plain voodoo which is conveyed in this thread is -again- staggering.

Edited May 25, 2013 by jfx

[Colonel_Panic]

Yes, well, that has been the mantra over and over. The evidence does not entirely seem to support that statement. At least some parts of the calculations seem to get offloaded to different cores. If your statement were true, KSP would never use more than 25% on a quad core, but that does actually seem to happen.

You're forgetting there are other background processes besides the main one. Your CPU can thread different processes differently. There's also system overhead in managing the load.

[Camacha]

Your OS switches threads (and yes - a single thread too) to different CPU cores as it sees fitting dozens of times a second (Windows time slice length is about 20ms). There is nothing you can do to prevent that and there is nothing that you should try to prevent that. Because of that you see an 100% Load (limiting) Thread appearing as roughly 25% load on four cores as you can see here in this retarded example:

http://i.imgur.com/frK6kDo.jpg?2

The amount of myths and plain voodoo which is conveyed in this thread is -again- staggering.

That was exactly what I was saying; more than a 25% load means that more than one thread is used. I am seeing a load that is bigger than 25%, so the conclusion should not be a surprise.

Also, remember you can set affinity to cores too. That this actually means a single physical core is used can be proven by looking at the temperatures. One core heats up significantly, its neighbor a bit too (they are not thermally isolated) and the remaining two stay close to idle temperatures. Performance suffers badly when KSP is set to a - and otherwise available - single core. This again points to the game normally running on more than a single core and/or thread.

I have heard the stories over and over, but like I keep saying, the evidence really seems to suggest something a bit different. That has nothing to do with myths or voodoo, but with actually trying to get to the bottom of a story people keep telling each other.

You're forgetting there are other background processes besides the main one. Your CPU can thread different processes differently. There's also system overhead in managing the load.

By background processes you mean Windows processes that have nothing to do with KSP?

That there is a certain amount of overhead is true, that is one of the reasons adding more cores is a game of diminishing returns. There is a certain point where spreading the load is more calculation intensive than actually doing the calculation. Of course there is the problem of some calculations being hard to split up, but that is another matter.