LFS Benchmarks

[Roadie]

I decided to do some benchmarks to see how CPU dependant LFS really is. The results I found are pretty impressive.

Test Setup:
CPU: Intel P4 2.4c @3.2Ghz
Memory: OCZ PC4000
Video Card: XFX 6800GS AGP @ 420/1.1 (Unlocked to 16/6)
OS: Windows XP MCE
Sound Card: Creative X-Fi Extreme Music

The Benchmark:
I made a custom replay with 12 XFG AI on Blackwood.

Fraps was used to record the framerate for 60 seconds, benchmark started and stoped at the same time every run.

Each test was ran twice, then the results were averaged.

The Results:
http://gallery.turbogfx.net/d/148-1/Chart+AA.jpg

The first test was run with 8xaa/16xaf.

http://gallery.turbogfx.net/d/150-1/Chart+no+AA.jpg

The second test was run without AA and AF.

Conclusion:
As you can see AA/AF, and resolution has a very little effect on performance. Framerates are almost identical throughout all the resolutions with and without AA/AF. So if you want to squeeze thoes extra frames out of LFS, but have a mid range video card, and a low end cpu, upgrade the cpu first. Most of you already know this, but to bad, I felt like posting something. :thumb:



UPADTE!

Ran the benchmarks again at 2.4ghz. Even more impressive results. ;)

http://gallery.turbogfx.net/d/152-1/LFS+aa.jpg
With AA/AF

http://gallery.turbogfx.net/d/154-1/lfs+no+aa.jpg
Without AA/AF

[Cue-Ball]

You've done plenty of work, but this is really nothing new. There's already an "unofficial" Live for Speed benchmark site with hundreds of results from all sorts of different configurations. It's quite easy to interpolate the results and see that CPU is the biggest factor by far in frame rate. The only other factor that makes a significant difference is using a video card without HVS support.

Also, using AA/AF as a comparison is not really valid since certain video cards take a bigger hit from these factors than others. Enabling 4xAA on a Radeon 9600 is going to make a big difference, where enabling it on an X800 will barely make a difference at all.

[Roadie]

Forgot to add that the LFS settings are all on the highest quality, and the latest drivers.

Yah, I know its nothing new, but i was bored. :P And yah, different cards will take a different performance hit with aa/af on. I should throw my 9500 pro in here and see how it does. :D

[B2B@300]

I decided to do some benchmarks to see how CPU dependant LFS really is.

Just how do those tests determine that? I noticed you have got a 2.4 OC to 3.2, why don't you drop the processor speed back to 2.4 and run the tests again :tilt: those results would be interesting :thumb:

[Jo-Chen]

why don't you drop the processor speed back to 2.4 and run the tests again :tilt: those results would be interesting :thumb:
I agree. Also it would be more interesting to benchmark a multiplayer replay since then the CPU would not have to do the AI's "thinking"*. Then you could show both, the CPU usage (2.4 vs 3 GHz) and the performance impact of AA/AF (at least of your GPU).

*since it is obvious that the CPU impact is huge with 12x AI (which is not representative for an "online simulator")

[AndroidXP]

Just how do those tests determine that?As you can see, changing resolution & AA/AF did virtually nothing to the frames. This means the GFX card has enough power reserves to cope with the craziest GFX settings (1600x1200@8xAA/16xAF) without any problems. This also means, that not the GFX card limits the amount of frames but the CPU does.

But you're right, it only shows that LFS is CPU dependent and not how much it depends on CPU. Anyone willing to underclock/overclock the CPU in 100 MHz steps? :D

[AndroidXP]

I agree. Also it would be more interesting to benchmark a multiplayer replay since then the CPU would not have to do the AI's "thinking".I'm not quite sure if the AI's thinking is really that CPU hogging. It's rather the ongoing complex physics calculations which have to be done for every car, even in multiplayer.

But you have a point, in MP the CPU stress is a bit lower, due to optimized code which "dumbs down" physics of far away cars, whereas the AI one has to be calculated all the time, no matter where on the track they are.

[thd]

http://lfsbench.iron.eu.org/

[bbman]

As you can see, changing resolution & AA/AF did virtually nothing to the frames. This means the GFX card has enough power reserves to cope with the craziest GFX settings (1600x1200@8xAA/16xAF) without any problems. This also means, that not the GFX card limits the amount of frames but the CPU does.

But you're right, it only shows that LFS is CPU dependent and not how much it depends on CPU. Anyone willing to underclock/overclock the CPU in 100 MHz steps? :D
How much AA/AF affects the FPS is more down to the RAM a graphics card has than it's GPU... Also, those graphics cards handle highly detailed textures far better than lower equipped ones...

[Roadie]

Yes, I will drop the cpu to 2.4GHz, and run the tests, then up to it 3.6ghz and run the tests again.

[Roadie]

There we go, check the first post. I posted benchies with my CPU @ 2.4

[B2B@300]

Very interesting results :thumb: it certainly does highlight the importance of cpu speed in LFS...

It is interesting to note that it has the largest impact on the minimum fps where it is for each 1% gain in cpu speed there is a 1.8% gain in fps :tilt: where for the maximum fps it has the least effect (i.e. for each 1% gain in cpu there is a 1.26% gain in fps)

That gives me the impression that adding in a better video card in a system, will likely result in better maximum fps but will help the minimum fps by a much smaller proportion :scratchch

Where as upgrading the cpu will increase the maximum fps a little but will have the most significant impact on the minimum fps. Which is the most important aspect for online racing :)

So it does indeed point to the general assumption of upgrade your cpu before graphics card is correct.

Can you please post the replay you used and what time points you used for the test, I'd like to try it on my system just for curiosity sake :D

[Roadie]

here is the replay. started at the second red light, and ended exactly 60seconds later.

5426

[vari]

The benchmark site uses that name for the replay so could you please rename it to avoid mistakes like running wrong replay for the site, overwriting it and so on.

[Roadie]

cant you rename it when you dl it?

[vari]

I wasn't asking it for myself.