DMagic

CPU Performance Database

367 posts in this topic

Smaller polling interval? What do you mean by that? The tool runs on fraps output - every frametime fraps recorded is shown as data point in the diagram

Now that I'm looking at your data I see how it works. I meant what would FRAPS give you if you recorded FPS instead of frametime. I'm guessing something close to the average from the frametime data.

Thanks _Aramchek_, but do you have your CPU info?

Apparently my FPS dropped as I lost parts. I don't...what...*mind explodes*

Camera Angle: Perpendicular to ground (may explain that ^)

That is a bit weird. When did you start the benchmark, right after the scene loaded? I usually throttle up then hit F11 just before launch. That seems unlikely to cause the results you're seeing though.

Can you send me the data file? Any filesharing service works fine, Google docs, dropbox, mediafire, etc... I want to collect everything so I can show it in a consistent form.

I updated the second post with the data I have so far. I'll keep adding everything there; I want to keep the first post intact so that people can see what I want first.

For the CPU speed, unless otherwise noted I'll put the dual core Turboboost clockspeed. Despite what many forum posts would have you think, KSP and Unity are not single-threaded. KSP CPU usage is regularly over 30% (on a quad-core CPU) so it is obviously using more than one thread. The PhysX implementation used by Unity, however, is limited to a single thread, and this is what causes much of the bottleneck. So most of the time the CPU should be running at the dual core clockspeed (if it has Turboboost or Turbocore).

And let me know if that skydrive file works. The spreadsheet is too big to open online, so it will kick you out to Excel, which might give your browser conniptions.

For the frametime data, because I want to keep everything consistent, I converted everything to frame rate. I divided 1000 by the time between frames to get frame rate, converted the time to seconds, and took a five-frame rolling average of the data to make the plot more manageable.

Edited by DMagic

Share this post


Link to post
Share on other sites

Thanks _Aramchek_, but do you have your CPU info?

lol Oops.

I5 3570k at 4.5ghz

Share this post


Link to post
Share on other sites

DMagic,

I've finished four of the tests, I'll wrap up the other four tomorrow. I think that there is a typo on the graphs that you posted of my data, so far I've tested at

1) 3.2GHz, 1x 6870, 1x 1920*1080,

2) 3.2GHz, 2x 6870, 1x 1920*1080,

3) 3.2GHz, 1x 6870, 2x 1920*1080, and

4) 3.2GHz, 2x 6870, 2x 1920*1080.

No tests at 3.6GHz, those will come tomorrow in a mirrored fashion.

Digger412

Share this post


Link to post
Share on other sites
DMagic,

I've finished four of the tests, I'll wrap up the other four tomorrow. I think that there is a typo on the graphs that you posted of my data, so far I've tested at

1) 3.2GHz, 1x 6870, 1x 1920*1080,

2) 3.2GHz, 2x 6870, 1x 1920*1080,

3) 3.2GHz, 1x 6870, 2x 1920*1080, and

4) 3.2GHz, 2x 6870, 2x 1920*1080.

No tests at 3.6GHz, those will come tomorrow in a mirrored fashion.

Digger412

Fixed that. Thanks for running these; it's interesting that crossfire seems to have so little effect. Does KSP, or Unity, just have lousy Crossfire implementation? I would expect the increased resolution to have an effect too, at least on the later stages.

Share this post


Link to post
Share on other sites

DMagic,

It surprised me as well. Based on the results of others, I would assume that I'm hitting a CPU bottleneck long before I hit a GPU bottleneck (antialised tests, anyone?). I'm curious to try running the test on my integrated graphics now as well.

Digger412

Share this post


Link to post
Share on other sites

KSP just doesn't require a lot of gpu power.

Share this post


Link to post
Share on other sites

Saw the Macbook Air 2011 (which is what i have) I cried a bit because I know the 2013 is cheaper and faster and has a longer battery life. Now that I see the performance difference I want one now.

Share this post


Link to post
Share on other sites

So I ran the test, staged all the way up to five, then checked the log and here is what I got:

2013-07-29 19:29:21 - KSP

Frames: 11068 - Time: 1143456ms - Avg: 9.679 - Min: 1 - Max: 24

Can anyone tell me what settings to have to the benchmark on?

Share this post


Link to post
Share on other sites

I will add my data this evening, I am using an older CPU (e6660 quad core @ 2.4gig) but it handles the game very well, up to usually 500 parts+

Share this post


Link to post
Share on other sites

Can anyone tell me what settings to have to the benchmark on?

For FRAPS? It looks like you only have "MinMaxAvg" checked under benchmark settings. You just need to check the "FPS" box, and if you want results like jfx or _Aramchek_ you can select "frametime". You can check PsuedoMonkey's results below to see what the output should look like when you only have "FPS" checked.

For KSP you just need the physics slider all the way to the right, and v-sync turned off. Everything else shouldn't have a big effect, at least not during the early stages.

AMD Turion X2 mobile ZM-82 @2.2GHz, 4GB DDR2, Radeon HD3850

This is why I stick to smaller vessels!

I made a more forgiving, 450 part rocket, but I found that performance wasn't slow enough with that. I didn't want people with faster CPU's starting out at over 30 FPS. Thanks for testing it though, I know it's a pain to spend 30 minutes on a launch.

Saw the Macbook Air 2011 (which is what i have) I cried a bit because I know the 2013 is cheaper and faster and has a longer battery life. Now that I see the performance difference I want one now.

The 2013 is definitely better. And more than that, I don't have to be so worried about it melting during testing. The 2011 was painfully hot to the touch while I was launching this, and the fan sounded like it was about to come loose.

Edited by DMagic

Share this post


Link to post
Share on other sites
For FRAPS? It looks like you only have "MinMaxAvg" checked under benchmark settings. You just need to check the "FPS" box, and if you want results like jfx or _Aramchek_ you can select "frametime". You can check PsuedoMonkey's results below to see what the output should look like when you only have "FPS" checked.

I had everything set up like you said there, including having only "FPS" checked. I'll do some Googling and check my settings.

Share this post


Link to post
Share on other sites
I had everything set up like you said there, including having only "FPS" checked. I'll do some Googling and check my settings.

Are you talking about the .txt log or the .csv log? I just checked my FRAPS benchmark folder and the FRAPSLOG.txt file shows what you have. But you should also get a separate .csv file for each benchmark, something like "KSP 2013-07-25 19-37-22-14 fps.csv". Maybe it's saved to a different folder?

Share this post


Link to post
Share on other sites
Are you talking about the .txt log or the .csv log? I just checked my FRAPS benchmark folder and the FRAPSLOG.txt file shows what you have. But you should also get a separate .csv file for each benchmark, something like "KSP 2013-07-25 19-37-22-14 fps.csv". Maybe it's saved to a different folder?

I should slap myself... Info coming soon, I'll just edit this post.

CSV:

https://drive.google.com/folderview?id=0B_8iwkSrd4jNTExWMFJJVGtxcjg&usp=sharing

My terribly embarrassing FPS graph @ 1366x768:

Rg3xXni.png

Edited by regex
1 person likes this

Share this post


Link to post
Share on other sites

Test done at 1366x768 with minimum settings (everything off). First stage decoupled early (Made no difference in performance).

Processor: Intel i3 2367M @ 1.4GHz (17 Watt)

Graphics: Intel® HD Graphics 3000

Frames: 6037 - Time: 1020200ms - Avg: 5.917 - Min: 0 - Max: 20

Y5tTg4V.png

Share this post


Link to post
Share on other sites

Here's a graph for a Q6600 @ 2.4, 3.0, 3.3 GHz

WsHaNR4.png

Share this post


Link to post
Share on other sites

I've got an Intel Core 2 Duo E4300, 1.8 Ghz. One above the slowest minimum spec spec.

?idl=213751819770&out=1

Data

What's not immediately obvious is the sheer amount of time; 26 minutes to stage 5. The meaningful values aren't just about framerate, they're about simulation speed. At launch it was running at 8 simulated seconds per real second, 4-1 at stage 14, 3-1 at 10, and 1.5-1 at stage 5. All but the first of these are approximations, but the issue is real.

1 person likes this

Share this post


Link to post
Share on other sites

That is a bit weird. When did you start the benchmark, right after the scene loaded? I usually throttle up then hit F11 just before launch. That seems unlikely to cause the results you're seeing though.

There was at most a couple of seconds delay between hitting F11 and firing up the engines, yet FPS decreased for almost a minute. It's definitely odd. The best explanation I can come up with was that something in the background of my laptop started running shortly after I started the recording. I loaded up KSP for just that flight, so that seems reasonable...possibly.

By "Data" I assume you mean the .csv file:

http://www./?l6mwnn43s3hxvks

Only contains FPS, not minmaxavg or 'the other one'.

-----

As someone pointed out, to make this truly representative you need to be taking graphical settings into account. If I were to run the test on lowest settings I'd probably get 'better' results - i.e. higher initial framerate, faster increase. Perhaps you should add "Run on lowest settings" to the things to do, or provide a settings.cfg file for people to run with (only problem then would be that low-spec PCs/laptops might crash :P)

So, here are the settings I was using. It's just a copy of the settings.cfg file from the first line down to just before keybindings start:


FLT_CAMERA_CHASE_USEVELOCITYVECTOR = True
FLT_VESSEL_LABELS = True
SIMULATE_IN_BACKGROUND = True
MAX_PHYSICS_DT_PER_FRAME = 0.03
MAX_VESSELS_BUDGET = 250
CONIC_PATCH_DRAW_MODE = 3
CONIC_PATCH_LIMIT = 3
SHOW_PWARP_WARNING = False
EVA_ROTATE_ON_MOVE = True
USE_STAGING_CONTROLS_ON_DOCKING = False
CALL_HOME = False
DONT_SEND_IP = True
SEND_PROGRESS_DATA = True
CHECK_FOR_UPDATES = True
VERBOSE_DEBUG_LOG = False
SHOW_CONSOLE_ON_ERROR = False
AUTOSAVE_INTERVAL = 300
AUTOSAVE_SHORT_INTERVAL = 5
SHOW_SPACE_CENTER_CREW = True
UI_SIZE = 768
SHIP_VOLUME = 0.3189855
AMBIENCE_VOLUME = 0.3248061
MUSIC_VOLUME = 0.3343491
UI_VOLUME = 0.2957029
VOICE_VOLUME = 0.3131648
SOUND_NORMALIZER_ENABLED = True
SOUND_NORMALIZER_THRESHOLD = 1
SOUND_NORMALIZER_RESPONSIVENESS = 16
SOUND_NORMALIZER_SKIPSAMPLES = 0
SCREEN_RESOLUTION_WIDTH = 1366
SCREEN_RESOLUTION_HEIGHT = 768
FULLSCREEN = False
QUALITY_PRESET = 5
ANTI_ALIASING = 2
TEXTURE_QUALITY = 0
SYNC_VBL = 0
LIGHT_QUALITY = 8
SHADOWS_QUALITY = 4
FRAMERATE_LIMIT = 120
FALLBACK_PART_SHADERS = False
PLANET_FORCE_SHADER_MODEL_2_0 = False
PLANET_SCATTER = False
PLANET_SCATTER_FACTOR = 0.5
AERO_FX_QUALITY = 3

EDIT: Renamed graph on page 2 with processor name instead of "FPS"

Edited by Epthelyn
1 person likes this

Share this post


Link to post
Share on other sites

Here are my result ( csv file from fraps ).

zip file.

http://gupl.dk/699676/

fraps.

Frames: 30604 - Time: 741972ms - Avg: 41.247 - Min: 10 - Max: 63

system..:

windows 7 64bit ultimate

core i7 980x ( 3333mhz )

12 gb ddr3 ram

2xvertex 3 ssd harddrives running in Raid. ( steam/kerbal installed on this drive ).

nvidia gtx 780 from gigabyte ( oc version ).

1 person likes this

Share this post


Link to post
Share on other sites
There was at most a couple of seconds delay between hitting F11 and firing up the engines, yet FPS decreased for almost a minute. It's definitely odd. The best explanation I can come up with was that something in the background of my laptop started running shortly after I started the recording. I loaded up KSP for just that flight, so that seems reasonable...possibly.

As someone pointed out, to make this truly representative you need to be taking graphical settings into account. If I were to run the test on lowest settings I'd probably get 'better' results - i.e. higher initial framerate, faster increase. Perhaps you should add "Run on lowest settings" to the things to do, or provide a settings.cfg file for people to run with (only problem then would be that low-spec PCs/laptops might crash :P)

Since KSP requires something slightly faster than an electrocuted cucumber as GPU to run flawless even at raised settings ... pretty much any settings will do. Especially shortly after launch the framerate is completely cpu bound even on i5-3/4 or i7-3/4 as we've already seen.

Here's a graph for a Q6600 @ 2.4, 3.0, 3.3 GHz

http://i.imgur.com/WsHaNR4.png

You shall be shamed for making a diagram without either axis descriptions nor caption. €i guess its the different frequencies OO.

I'd be interested in a benchmark by someone with an AMD FX cpu. It'd be interesting if they suck as badly as expected.

Edited by jfx

Share this post


Link to post
Share on other sites
Since KSP requires something slightly faster than an electrocuted cucumber as GPU to run flawless even at raised settings ... pretty much any settings will do. Especially shortly after launch the framerate is completely cpu bound in this launch even on i5-3/4 or i7-3/4 as we've already seen.

My assumption is that anyone running a high end CPU will also have a decent GPU (and don't dismiss the power of electrified gourds and tubers, we all know what happens when you put a

). I can imagine that someone with an OC'd i5 3570k running on integrated graphics might be GPU limited, but except for those edge cases I'm not very worried about it.

Just look at Digger's results. Doubling the resolution has almost no impact on performance (Crossfire/SLI might have issues of its own, so I'm not so sure about those results).

I'll run some tests later today with min and max graphics settings. I've seen an impact based on changes to graphics settings, but only on the later stages where I'm running above 60 FPS and am likely GPU bound.

Slugy sent me his data, so I'll add that to the first page later. But his graph really nicely illustrates the effects of CPU clockspeed. The increase and frame rate and the decrease in time for each stage is obvious just from the plots (well, assuming you've seen these graphs a thousand times and can discern which stage it's at just by looking at the frame rate).

What's not immediately obvious is the sheer amount of time; 26 minutes to stage 5. The meaningful values aren't just about framerate, they're about simulation speed. At launch it was running at 8 simulated seconds per real second, 4-1 at stage 14, 3-1 at 10, and 1.5-1 at stage 5. All but the first of these are approximations, but the issue is real.

Oh definitely. I really spent a lot of time making sure the rocket was stable because of this. The first two stages can take over ten minutes to finish. I didn't want the rocket falling apart somewhere in the middle just because it would take so long to try it again.

And let me know if anyone has launch failures. I had a few rare cases where the 2.5m radial boosters on the lowest section were breaking off. I made a few minor changes and haven't seen the issue again (I've launched probably 15 to 20 times in its current configuration without issues), but I'd like to know if it happens so that I can see about fixing it.

Edited by DMagic

Share this post


Link to post
Share on other sites

I'd be interested in a benchmark by someone with an AMD FX cpu. It'd be interesting if they suck as badly as expected.

If I still had mine I'd run it, it's pretty much even with the 1090t's results iirc.

The idea that the FX's suck is a bit misleading though, for day to day use and heavy multitasking my BD @3.9ghz (bad chip) was better than my i5 is, gaming however, aside from a few games. In KSP though, like I said, it was fairly even with the 1090t's results.

Share this post


Link to post
Share on other sites

jfk and _Aramchek_,

According to a cpu benchmark website, here are the scores for the processors tested so far (and the lowest- and highest-end AMD FX processors):

AMD 1090T Six-Core   Passmark score: 5709, Single-Thread Passmark score: 1218
AMD FX-6100 Six-Core Passmark score: 5405, Single-Thread Passmark score: 1194
AMD FX-6350 Six-Core Passmark score: 6980, Single-Thread Passmark score: 1493
AMD Turion II M640 Passmark score: 1723, Single-Thread Passmark score: 0929
Intel Core i5-3570K Passmark score: 7119, Single-Thread Passmark score: 2011
Intel Core i5-2500K Passmark score: 6386, Single-Thread Passmark score: 1864
Intel Core i5-3470 Passmark score: 6609, Single-Thread Passmark score: 1907
Intel Core i5-2467M Passmark score: 2293, Single-Thread Passmark score: 0934
(note to self: aligning columns without tables is a pain.)

It appears that the FX's do indeed perform at about the same level as the 1090T, with up to an approximately 20% speed increase at most. And then Intel comes along and shows AMD the door.

Digger412

Edited by Digger412

Share this post


Link to post
Share on other sites
If I still had mine I'd run it, it's pretty much even with the 1090t's results iirc.

The idea that the FX's suck is a bit misleading though, for day to day use and heavy multitasking my BD @3.9ghz (bad chip) was better than my i5 is, gaming however, aside from a few games. In KSP though, like I said, it was fairly even with the 1090t's results.

Ask and ye shall recieve stock 6350 fps only

1 person likes this

Share this post


Link to post
Share on other sites

marach,

Wonderful! As far as FPS goes, it does seem to average about 20% faster than my 1090T. It completes the simulation in half the time though D:

Digger412

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!


Register a new account

Sign in

Already have an account? Sign in here.


Sign In Now