Home   Help Search Login Register  

Author Topic: Benchmark  (Read 1170 times)

0 Members and 2 Guests are viewing this topic.

Offline XCess

  • Former Staff
  • ****
Benchmark
« on: 26 Jan 2006, 10:44:15 »
I know this isn't really on topic with the board, but this is something that's really got me confused.
I just realised I had a backup of OFP on my hardrive under C:\New Folder from a prevous reinstall of windows. Now, lately, my standard installation of Operation Flashpoint, has been getting a benmark of around 2.5k, but the version in this folder is getting 5000+.
I am thoroughly confused. This question does in someway pertain to editing, as benchmark can effect the erformance of a mission and therefore the users enjoyment of a mission. And as mission designers we should strive to make our missions as enjoyable as possible, in every way. (defense for topic lock over :P)

Any ideas as to what the hell is going on?

Offline macguba

  • Former Staff
  • ****
    • macguba's operation flashpoint page
Re:Benchmark
« Reply #1 on: 26 Jan 2006, 11:47:12 »
Don't panic, discussions of benchmark are on topic for the reasons you mention.

As to the question, yes I know exactly what is going on.   Benchmark is a weird fucked up function that provides no more than a very general guide to performance of your CPU.   ;D   If you click the detect button repeatedly the display will cycle through several different benchmarks.    A difference of a factor of 2 is, in benchmark terms, not a big deal.

In previous discussions of benchmark we have settled on 2000 as being a good cutoff level if you want to use it as an eyecandy control.    I believe there next rest up is just under 7000.

In Abandoned Armies THobson used a dialogue to let the player decide whether to turn environmental eyecandy on or off.   For sophisticated missions this is probably a better approach.
Plenty of reviewed ArmA missions for you to play

Offline XCess

  • Former Staff
  • ****
Re:Benchmark
« Reply #2 on: 26 Jan 2006, 12:02:47 »
Thanks for moving the thread to the appropriate section. I've noticed before that the benchmark changes slightly each time you click auto detect, but never more than a hundred either side. But simply opening a different install of the gamein another folder has increased my benmark by almost 3000 (the improvement alone is better than the previous benchmark score)
So how des this relate to computer performance?

The mystery of the benchmark goes on..

Offline macguba

  • Former Staff
  • ****
    • macguba's operation flashpoint page
Re:Benchmark
« Reply #3 on: 26 Jan 2006, 12:06:01 »
Well, yeah, it does that.  Don't worry about it.

Actually you could do some investigations.   Have you changed any hardware between installing the two versions?   What happens if you move the old installation to a new directory?  
Plenty of reviewed ArmA missions for you to play

Offline Mandoble

  • Former Staff
  • ****
    • Grunt ONE and MandoMissile suite
Re:Benchmark
« Reply #4 on: 26 Jan 2006, 12:53:44 »
In my ofp directory I have two executables to setup the textures, memory usage, etc of OFP (dont remember the exact names of these programs). Both are similar in purpose, but different in layout and options, probably one was installed with OFP CWC and the other with Resistance. The first one gives me about 2k in performance, but, if used, I get weird graphic effects and very poor frame rate. The second exec gives me 5k in performance and, when used, everything works fine. I would say that the "older" exec is obsolete and should not be used (not even executed).

Offline XCess

  • Former Staff
  • ****
Re:Benchmark
« Reply #5 on: 26 Jan 2006, 13:31:09 »
I believe I have since upgraded my computer since the first install of OFP (the 5k version), before I was using a 700mhz AMD Duron, a GeForce MX 440 and 256mb SDRAM,  the second, lower performing version was installed on an Athlong 2400+, with GeForce FX5500 and 512mb DDR RAM.

The old install was transferred to the new hardrive by making the old one a slave (which has since been disconnected, and accidentally formatted  ::)).

 Another difference is that the first version was installed one Windows XP Home Edition SP1 and the second (the 2k bench) was installed on Windows XP Office.

The hardware, quite obviously, has been improved immensely, making the drop in benchmark very confusing to say the least. I'm not sure which version of windows is more efficient all i really know about windows performance is that all versions suck (except ME which never ever crashed on me, shame modern games won't run on it).

I did however, used to think running auto detect would asses your hardware at that point in time (when I upgraded I did notice an improvement in benchmark while using the same hardrive).

I'm not so sure now ???

Offline macguba

  • Former Staff
  • ****
    • macguba's operation flashpoint page
Re:Benchmark
« Reply #6 on: 26 Jan 2006, 14:11:19 »
I do know that RAM is irrelevant and nominal CPU speed is important.  But I also know that when I changed from an Athlon to an equivalent Sempron the benchmark dropped from 6000 to 3000 and the performance on OFP was, overall, very slightly better.  (In some other areas it was very slightly worse, but i most things it was pretty much unchanged.)

It is well known to be a great mystery.
Plenty of reviewed ArmA missions for you to play

Offline Mikero

  • Former Staff
  • ****
  • ook?
    • Linux Step by Step
Re:Benchmark
« Reply #7 on: 26 Jan 2006, 15:53:18 »
Following the introduction of anything past an Intel celeron, yes, ofp goes _that_ far back, benchmark numbers are irrelevent. It is a measure of cpu execution speed which is fairy floss territory because

athlons / pentium 4's and celerons all treat cpu cache differently

introduction of Doube Data Rate ram fudges the figures making internal cache *sometimes* worse than using external ram.

cpu speed is so blindingly fast anyway that you spend most of your time waiting around quickly.

The real issues for any sort of benchmark that involves video graphics is how fast things can be rendered and <> transferred. Transfers involve direct memory transfers from the hard disk, vs, fast access to dedicated memory areas. None of which has a single thing to do with the OFP benchmark. At the very best, it is a highly misleading figiure of relative machine performance. But, as most of us know, a machine running a benchmark of 2000 (athlon), *often* beats a pentium 4 at 5000. It doesn't make the athlon better, it makes the benchmark, ridiculous.

In terms of OFP, memory is not an issue either. The sweet spot for ofp, regardless of the quantity of addons used, regardless of whether your mother in law had cornflakes for breakfast on tuesday, is around 312 meg of _dedicated_ ram. The reason why MORE memory is irrelevant is because ofp uses earlier versions of Direct X whose functions dont' exploit larger values. There is nothing to be gained running ofp in 1meg, versus 300. The direct x rendering will pull from hard disk cache irregardless.

However,  8). The more memory you have on your cpu, the better it is for OFP because all those other tasks being run in background are being run in memory, not, in virtual memory.

Whenever and wherever you start getting poor performance out of the ofp engine, the last place to look, is ofp!. It's because you have added something nasty. And typical among these nasties is Norton Antivirus. It serves as an excellent example of, unknown to you, some other task running in the background. You can extend this to all those other shifty little buggers like adobe acrobat upater, even Windows auto update. They kick in, you, kick out.

If you hit performance issues, alt tab out of ofp and hit the three fingered salute (alt-ctl-del) it will bring up your performance dialog. Look at ALL the processes running under YOUR name. You'll see things in there that will surprose you. (Oh DUH, i never knew ICQ was chatting)

---
If you want a much, much closer realistic figure of ofp performance, use Kegety's DxDll. And select the fps option (framerate). This will give you a highly accurate and realistic figure of how well frames are being displayed as you play. Figures beyond 25 are excellent, figures around 10 are playable. Anything less is sludge territory.

Finally, there's a wrinkle that no-one can measure. The bis engine will sacrifice AI intelligence to keep the video performance up and smooth. It means that while everything remains oily slick and responsive, the actual AI, the actual nasties trying to shoot you are dumbed down dead, along with your own troops that can't drive a straight line.

The good news is, if this one hit's you randomly, as in , sometimes mission is fierce, soemtimes not. It is a dead giveaway that you have some background task swallowing cpu crunch.

« Last Edit: 26 Jan 2006, 15:53:59 by Mikero »
Just say no to bugz

Offline macguba

  • Former Staff
  • ****
    • macguba's operation flashpoint page
Re:Benchmark
« Reply #8 on: 26 Jan 2006, 17:03:19 »
Interesting point on the RAM.  I noticed an immediate improvement in performance when I increased RAM from 512 to 1.5g.   (Well, it certainly meant I could use higher viewdistances.)   However, assuming all the nasties are switched off and you are just running a normal setup, then the optimal amount of RAM to have is probably 756 or thereabouts.     That gives enough for the max on OFP, plus all the normal background gubbins.

Intel chips do indeed give much higher benchmark results than AMD ones for the same performance.

The summary is, if you want to chuck a benchmark test into your mission for a bit of eyecandy, then do it.   However, if you have lots of sophisticated effects, then you need something, well, more sophisticated.
Plenty of reviewed ArmA missions for you to play