CPU Performance Database

[Maeyanie]

It should run fine on a LiveUSB, so long as you have the proper video drivers installed and don't mind a horrifically long initial game load.

BuGLe seems to do the job pretty well, though it's a little more complex than FRAPS. I compiled it from source, installed the example filters/statistics files, and ran it using BUGLE_CHAIN="logfps". The logfile it spits out isn't in CSV, but it's fairly easy to convert with awk or sed or whatever.

[_Aramchek_]

Thanks for running all those tests. That 10-20% increase in performance is definitely interesting (the much bigger jump later on is probably less meaningful than the early performance difference).

The difference at the begining gets my attention any way, , later on I mean 100 vs. 80 fps (just generalizing) makes zero difference to me...but it seems to handle the heavier loads a bit better.

I've never really dealt with linux, but the whole "steamOS" thing, which I'm not fully sure of yet, has my interest. I'll probably be buying an ssd in teh next week or two and if I do I'll throw a linux install on it first and run the bench again on my pc, just to give a basis for comparison between architectures.

[xcq]

So I finally get around to register at KSP Forums to contribute to the general knowledge.

First off: Great idea, especially for people who want to upgrade their system and estimate the improvements on KSP. One suggestion though: At your second post the "Cinebench and Passmark" graph is missing the individual names of the CPUs. It would be really helpful, if they were right next to the data point, and you didn't need to look the values up and make a guess which ones they could be.

Also did the benchmark with my old and new system. I tried my normal save with tons of mods but that was horrendously slow. Started with a brand new KSP 21.1 install and save, nothing else there. Only thing I changed from default settings was the resolution and physics delta-time. V-Sync was apparently forced by something else and I was too lazy to fix that. I stopped each run after exactly 6:00 in-game flight time had passed.

FPS data: https://docs.google.com/file/d/0B0PzQSzrNEYnNmZVU2lTTm42WDg/edit?usp=sharing

PM/CB:

Q9300 @ 2.5GHz: Passmark 3,588; Cinebench 2.96

Q9300 @ 3.1GHz: Passmark 4,267; Cinebench 3.62

4770K @ stock: Passmark 10,970; Cinebench 8.55

[DMagic]

Thanks for collecting all of those runs. I'm going to have to make a new graph to split up those C2Q and 920s since there are so many of them.

I've considered adding labels to that first plot, but I didn't want to make it too cluttered. I think what I'll do is make a high resolution version that you can click on to make it easier to zoom in. I should do that for all of the graphs actually.

[Justy]

The System: Acer Aspire M3970

CPU: Intel Core i5-2320 @ 3.00 GHz

(CPU-Z: core speed 3193 MHz under load, bus speed 99.77 MHz)

RAM: 6 GB DDR3 667 MHz

GPU: ASUS ENGT430 (NVIDIA GeForce GT430), 1 GB

OS: Windows 7 Home Premium 64-bit, Service Pack 1

Passmark: 5837

Cinebench: 4.93

The CSVs are in this Dropbox folder. I did several tests; the most relevant are in the folder called "New Drivers," which was a run I did after all the rest, when I noticed nVidia posted new graphics drivers last week.

The rest I did to compare a few things to see if they'd affect performance:

• "Baseline Condition" has MS Security Essentials running.
  
• "No Antivirus" has MSSE's Real Time Protection switched off.
  
• "USB3 Flash Drive" moves the KSP folder from the local hard drive to a Kingston 16GB USB 3.0 "DataTraveler 100 G3" flash drive.
  
• "With Readyboost" uses two of those same flash drives dedicated to Readyboost disk cache.
  
• "New Drivers" is, as I said, Baseline Condition re-done with the most recent nVidia update.

In each case, I tilted the camera until I couldn't see the water tower anymore, throttled up, enabled SAS, hid the navball, started FRAPS monitoring, and waited until FRAPS' green square vanished before launching. All runs were in the default 1024x768 window on a 1920x1080 monitor, with the required tweaks to physics delta-time and V-sync.

It looks to me like disabling MSSE's Real Time Protection increased performance slightly, as did using the latest video drivers; running KSP from the USB drive had little effect, and Readyboost actually hurt performance. The differences were all pretty small, though.

Edit: I did full-to-burnout runs with antivirus disabled at various resolutions, in both fullscreen & windows, as well.

Edited September 28, 2013 by Justy

[DMagic]

I updated the front page with all of the new results. I also added high-resolution version of all of the framerate graphs. You can just click on them to open up a new page, you might have to click on the zoom button a few times to get to the actual full-resolution images (they are about 4000x2000). This makes it much easier to see the small differences in the early stages. I'll update those full-resolution images again at some point to make them a bit easier to read, and I'll make a labelled, high-resolution image for the Passmark/Cinebench graph as well.

I've only added the "New Drivers" run from your data sets Justy. I'll take a look at the others at some point. I may add a few graphs with comparisons at different resolutions and settings. I have results from a few different people, so that could be useful.

It's not too surprising that running KSP from a USB drive or using Readyboost didn't help. Even USB 3 drives have pretty crappy flash memory compared to any decent SSD, but even using a slow HDD shouldn't really affect performance once everything is loaded. And I think Readyboost only really helps with really low RAM situations. I wouldn't have expected MSSE to have much of an effect on a quad core system (it wouldn't be surprising on a dual core system though), but I guess it's not too surprising that background processes could affect KSP, even with lots of CPU headroom.

[Justy]

My reason for trying Readyboost was that I thought the rhythmic micro-pause hiccups was being caused by disk access... I've since read up on it more, and I see that's not the case. But hey, it was worth a shot. Thing is, though, I thought the worst that would happen would be that I'd have wasted a couple bucks on a thumb drive; I was surprised that it actually seemed to retard performance.

The effect of enabling/disabling MSSE is small, but it's there, especially later in the flight. Disabling MSSE only adds a few FPS, but you can see it's sped physics up quite a bit, I'm better than ten seconds of game-time ahead after ~470 seconds of real-time. By this point I'd done several practice runs and I think I had my staging pretty smooth, I don't think that ten second gap is human error, but I guess to be certain I'd need to do a few identical runs and average them. Comparison below is with old video drivers.

I am kind of surprised how well my i5 2320 3.0 GHz keeps up with the i5 2500k 3.3 GHz, especially in the first half. Of course, the advantage of the 2500k is that, as shown, it can be cranked up to speeds that blow my system out of the water, while mine is stuck. Replacing my CPU for more speed would cost nearly as much as I paid for the whole system, even before additional cooling.

Edited September 30, 2013 by Justy

[DMagic]

There does seem to be a bit of a difference with MSSE off. And a small difference in performance can make a big difference in total simulation time. Although, if I understand this correctly, anything above 33FPS should run in real time (a max of 30ms of game time can pass for each frame, and 33FPS is about 30ms/frame), so that larger difference in the later stages shouldn't affect simulation time. That slight increase, especially around 200-300 seconds could be enough to account for the difference though.

I would have expected the 2500k to run a little bit faster than the 2320 at stock speeds, but there are a lot of factors that could account for a 10% or so performance difference, so it's not too surprising to see them running the test at similar levels. The 2320 is really just a slightly slower, locked version of the 2500k; I don't think there is any other difference between the two CPUs. And the right side of the graph probably isn't a valid comparison, I did some tinkering around with those frametime datasets to avoid making the graphs too messy, so the FPS and frametime graphs don't match up exactly.

[DMagic]

I updated the first page with better high-resolution graphs for the framerate plots, a partially labeled CPU comparison chart and a fully labeled high-resolution version, and a short discussion about the results at the top of the third post.

And thanks for the suggestion about labeling that chart xcq. Once I did that a few trends became clear. First, AMD and mobile CPUs all tend to underperform based on their benchmark scores, which isn't surprising. And the Core 2 Quad CPUs produce very consistent results, with performance scaling very well with clockspeed, that was nice to see. It's also clear that the recent Core series CPUs from Intel tend to perform very well and cluster together. And the overclocked CPUs fill out the top right corner of the chart, though there does seem to be a limit to how much can be gained from this.

From watching last week's Squad Cast it seems that the 0.22 update will come with some significant performance improvements. This is, of course, awesome, and I really want to see some comparisons between 0.21 and 0.22. This also means that I won't be able to compare performance between versions for the database. I don't think it will invalidate any of the testing I've already done because I was never interested in the performance of any specific CPU, only in the comparisons between many different CPUs.

[briansun1]

I will try this some time this weekend though do we have to use fraps? or can we also use MSI afterburner

[DMagic]

I think MSI Afterburner should work. It has a framerate monitoring function and a logging function, I've just never actually tried using them together. I use the onscreen overlay a lot to monitor framerates and GPU usage (I really only use FRAPS for collecting framerates from runs), I disable it when collecting FRAPS data though.

[minerman30]

Wow, very cool, thanks Psuedo. The last three on your list are mine by the way (w/AMD 7850 GPU), they are all stock, I'm just reporting the turboboost speed.

Passmark has numbers for single threaded benchmarks too. I'm wondering if those match up even better. I'll try to take a look at them later.

Edit: The single threaded numbers don't seem to match up very well, the 4650U and 3470 are almost the same, which doesn't seem to reflect KSP performance at all.

And I'll add this to the 2nd post, if you don't mind.

Well basically, yes, but you can't just compare clockspeed and say one CPU is better than the other. There is a huge range of performance for different CPU's at the same speed.

And I'm not sure what you mean by K number. Those are just the CPU model numbers, and K means they are unlocked and can be OC'd. But generally the higher the model number the better.

Thanks, I got an Intel i5 4670K. Is it worth the risk to overclock it?

[briansun1]

unless you find performance lacking don't

[DMagic]

Thanks, I got an Intel i5 4670K. Is it worth the risk to overclock it?

As long as you have the right kind of motherboard (Z87 chipset) I don't see why not. There is little risk in moderately overclocking Intel CPUs because all you can really change is the clockspeed multiplier. I guess if you cranked up the voltage too high without sufficient testing or cooling you could fry your CPU or MB, so don't do that.

If you are really worried about it you could try increasing the multiplier without changing the voltage. You might be able to get a decent increase doing that, just go slowly and test between each step.

[DMagic]

With 0.22 coming out soon and supposedly bringing some performance improvements I figured I would go back and look at some of the earlier versions. I went back to 0.18, before that there is no docking and I figured that would be too much of a change for my CPU rocket to still work. I had to add a few parts from 0.21 and 0.20, but otherwise everything was tested in the same way, at the same settings values.

0.21 and 0.20 show just about the same performance, which agrees with what I recorded earlier when I was first designing the rocket. 0.19 and 0.18 show a huge drop in performance though. I can't be sure that this drop is real and isn't due to some incompatibility with my rocket or the parts I added. I made sure to edit the part .cfg files so that they were in the correct format, and I changed the version number of the CPU rocket in the .craft file. There could be some other problem, or this difference may be legitimate. I wonder if anyone else has recorded such a big difference in performance between these versions.

[DMagic]

It should run fine on a LiveUSB, so long as you have the proper video drivers installed and don't mind a horrifically long initial game load.

BuGLe seems to do the job pretty well, though it's a little more complex than FRAPS. I compiled it from source, installed the example filters/statistics files, and ran it using BUGLE_CHAIN="logfps". The logfile it spits out isn't in CSV, but it's fairly easy to convert with awk or sed or whatever.

So I finally got a proper installation of Ubuntu 12.04 setup on a partition of my second SSD. I managed to get KSP and Steam working, in x86 and x64, and I managed to get BUGLE running (that was fun), but my performance looks terrible. The only thing I can think of is AMD's drivers. It's hard to be certain, but I think it's the GPU that limiting here. Reducing the settings to almost the lowest values improves my performance a bit, but it's still much lower than in Windows.

Here is a comparison of Linux and Windows at the same settings. The drop off is huge, around 50%. Bugle was giving me errors when I tried to load the x86 version (some missing 32 bit .so file), but I didn't notice much of a difference in my brief test of the x86 version loaded from Steam.

What kind of GPU do you have Maeyanie? Does anyone know if there is a good set of AMD drivers to use, or are they just all terrible?

[DMagic]

So about those supposed performance updates for 0.22, it doesn't look like they made the cut:

If anything I get slightly worse performance. In last Friday's Squadcast C7 mentioned that some of the performance updates might not make it into 0.22, I guess he was right. If anyone else sees a difference let me know.

[Tollazor]

Damn, I was hoping it would.

I would really like to see ~10-20% improvment.

[DMagic]

So I got a Surface Pro 2 today and what's the first thing I do, try out KSP. Ok maybe it was more like the fifth thing I did. I didn't get a chance to try out Diazo's touchscreen mod, but I do think that with proper support the VAB could be awesome on a touchscreen.

I also ran some more tests on the MacBook Air on 0.21 and 0.22. This further confirmed that 0.22 did nothing for performance. But the Surface did surprisingly well considering that both the CPU and the GPU are a bit weaker. I guess these things are just too constrained by power and temp for small differences in specs to make a big difference in performance. There could also be a difference between OsX and Windows.

You can see that performance is pretty much even between them. I updated the front page too so you can see the comparison to other CPUs.

Edited October 23, 2013 by DMagic

[TUFOM]

Forgot Vsync on & too lazy just now to run again with vsync off. But here is what I got:

AMD 5600K @4.02Ghz (atm because ksp caused some overheating/throtling issues, was 4.4Ghz, even wPrime95 couldn't throttle this O_o)

8GB DDR3 1333Mhz

AMD HD 7850 2GB OC (1000/5200Mhz GDDR5)

WD AAKS 500GB SATAII

Windows 7 Home Premium 64bit

And I did it anyway with little bit fixed clocks & chart more clear:

Edit: for some reason my CPU overclock was not applied so this is just without vsync then & ram 1333->1600. (to be clear cpu clock supposed to be @4.2Ghz)

Edit2: Don't know if 1920x1200 resolution & 2xAA is affecting result.

Edited October 23, 2013 by TUFOM

[DMagic]

Thanks for running these tests. I think with your HD7850 the game should be fine at those settings, that's the same resolution and GPU I used to have and I don't think it was really limiting performance.

Can you send me the log file with the FPS numbers so that I can add this to the front page?

[TUFOM]

http://www.tampereentietokonepalvelu.com/KSP/KSP%202013-10-23%2021-59-29-70%20fps.csv

[DMagic]

Thanks TUFOM. These are interesting results. If you look at the comparison on the first page you'll see that your CPU performs much lower than other AMD CPUs at the beginning of the test, but then jumps ahead of them later on. I would say that being GPU limited could cause this, but that doesn't seem to be the case. The plot of your results has the stair-step pattern typical of CPU limited performance, and your GPU should be more than enough to handle KSP. The others don't seem to be GPU limited either. Very strange.

[Diazo]

So about those supposed performance updates for 0.22, it doesn't look like they made the cut:

Were there serious in-game performance updates for 0.22?

I understood that the major performance focus for 0.22 was to reduce loading times between scenes, not in-game performance.

I'll go and see if I can find the official patch notes.

D.

edit: Actually, where is the official change log? All I can find is a list of changes in the 0.22 released thread in the general forum.

Edited October 28, 2013 by Diazo

[Diazo]

So I got a Surface Pro 2 today and what's the first thing I do, try out KSP. Ok maybe it was more like the fifth thing I did. I didn't get a chance to try out Diazo's touchscreen mod, but I do think that with proper support the VAB could be awesome on a touchscreen.

I've actually been wrestling with how to do the VAB on a touchscreen.

There is no way to drag correct? With a mouse you get the part ghosted and can preview where it goes. A touchscreen does not have this capability correct? All I get are mouse clicks?

D.