NS2 and CPU Core scaling test

Cr4zyb4st4rdCr4zyb4st4rd United Kingdom Join Date: 2012-08-09 Member: 155200Members, NS2 Map Tester, Reinforced - Supporter, Reinforced - Silver, Reinforced - Gold, Reinforced - Diamond, Reinforced - Shadow
edited April 2015 in NS2 General Discussion
So, I get a fair amount of people asking me how well my PC runs NS2. I stream a fair bit of NS2 and with my specs on my twitch page people wonder why I have what I do when I play NS2 and it won't use it. Reality is I have my PC for work and it does a damn fine job at that, playing NS2 on it is for fun!

However, because people ask and I have the ability to do so, I took a 2min run around Summit with the 'scenload xl' loaded in and compared how NS2 actually scales across CPU cores, both physical and logical (hyper-threads).

For those who don't know my spec is as follows;
CPU: i7-5960x @ 4.25GHz
GPU: SLI GTX 980
RAM: 16GB DDR4 @ 2666Mhz
OS+Game SSD: 512GB Samsung SM951
OS: Windows 8.1 64bit

So the following is NS2 performance with the NS2.exe affinity changed, this does mean that the loads on the lesser core counts are probably going to be exaggerated in comparrision to what an actual CPU of the same architecture and clock speed would see, because I still have many other free cores doing everything in the background that windows and other software needs.

The game was run on maximum settings at 2560x1440, resolution doesn't impact my performance as NS2 really is CPU limited to hell.

All core counts:

5912552539.jpg

That's a bit messy so here is 1+0 to 4+4 (pretty much most peoples maximum) and 8+8

ef237e6c58.png

As you can see NS2 scales pretty well upto 4 physical cores having hyper threading on makes a minimal framerate difference at this point, so if you're debating between an i5 or an i7 and you're only gaming in NS2 the i5 is your most cost efficient choice.

With the super messy graphs it's hard to see which really performs best when it comes to a 4-core hyper-threaded cpu vs a 8-core hyper-threaded one. So here is yet another graph showing the benefits of spending 3x the price on a 5960x over a 4790k.

b76b80645d.png

Ahhh, 10fps more on avg, totally worth it :p
So no my CPU doesnt play ns2 any better than a standard i7 or i5 would really. Especially when you consider that most of the more "consumer" cpus will boost to higher core speeds anyway and can overclock much more than my poor 5960x, which is another way you'll gain a fair amount of frames in NS2.

Comments

  • d4rkAlfd4rkAlf Sweden Join Date: 2013-11-12 Member: 189309Members, Squad Five Blue, Reinforced - Shadow
    Good informative post with nice graphs to boot. Thank you!
  • UncleCrunchUncleCrunch Mayonnaise land Join Date: 2005-02-16 Member: 41365Members, Reinforced - Onos
    Makes me think we do not have a "benchmark demo" like in other games.
  • FrozenFrozen New York, NY Join Date: 2010-07-02 Member: 72228Members, Constellation
    I need one, 115fps with G-sync isn't enoughhhhhh
  • Cr4zyb4st4rdCr4zyb4st4rd United Kingdom Join Date: 2012-08-09 Member: 155200Members, NS2 Map Tester, Reinforced - Supporter, Reinforced - Silver, Reinforced - Gold, Reinforced - Diamond, Reinforced - Shadow
    Makes me think we do not have a "benchmark demo" like in other games.

    Ye it's a shame, im sure it could be moddable, thankfully the scenload xl on summit helps dump a ton of stuff into summit to atleats impact the framerate.

    If you want to run this yourself and compare, I just re-ran it and recorded the loop I take
    (will be done soon if it isnt when posted)

    Fairly simple, start training on summit
    in console enter the following:
    scenload xl
    damage 0.01

    If you don't set the damage multi you will die.

    Then I started all my benchmarks the second I start moving from the center of marine base, walk with axe to reactor and then running with rifle to the end, should take 2mins pretty much everytime, not 100% the same every time but closer than pub testing and better than an empty map.

    I test with fraps FPS benchmark.
Sign In or Register to comment.