Cujo's CPU Dependency Thread

[Cujo]

Alright, I've been talking for a while about doing this and now I finally have. I've got proof that cs:source is in fact, cpu/ram limited.

 

I've done 4 FRAPS runs of about 10 minutes each. For each run I was on the same server (xT pub: 69.65.5.11:27015), on the same map (de_dust2), and on the same team with about the same number of players (14-20). I was running the same apps in the background (MSN, teamspeak, xfire and fraps) and my graphics settings were identical. All in game settings were on low except I had trillinear filtering on and shadow detail on high. My autoexec.cfg for testing is below (normally I cap my fps at 100). Only the first two are relevent to system performance.

 

cl_ragdoll_physics_enable 0 - takes load off the cpu

cl_phys_props_enable 0 - takes load off the cpu

cl_lagcomp_errorcheck 1

cl_interp 0.01

cl_crosshairscale 2000

fps_max 1000

rate 20000

cl_updaterate 100

cl_cmdrate 100

 

I did 2 runs at 1024x768 and 2 runs at 1600x1200. At each resolution I did 1 run at 2.0ghz (200x10) and 1 run at 2.5ghz (250x10). My ram was 1:1 ratio so at 250 it was running at 250. As seen in this link, memory bandwidth has a very large impact on HL2. I will do further testing by setting my cpu htt to 200 and my ram speed to 250 using the dividers on my motherboard to see what affect ram bandwidth has on performance.

 

Here are my findings:

 

As you can see my avg fps at 2.0ghz is EXACTLY the same at each resolution. This indicates that with my x800xl, and my cpu running at 2ghz, I am completely limited by the cpu. In this instance upgrade my cpu would help me to get more out of my video card. I, however, don't upgrade, I overclock. So I bump my brand new x2 3800+ up to 2.5ghz. What do I see then? Well my fps goes up by a massive 26% at 1024x768 with only a 20% cpu overclock. Obviously the ram has some impact on that which is why it's more than a 20% performance increase. What you also notice is that my performance at 1600x1200 only increases by 19%. Still a substantial increase but it indicates that my video card is starting to hold me back at this higher resolution.

 

So what can you take from these findings? I'd say that if you have an x800xl or higher and a cpu that's running 2ghz or lower then your next upgrade, without a doubt, should be your cpu. If you have an athlon xp or an intel rig then your next upgrade should be your cpu anyway regardless of your video card as even a lowly 2ghz a64 is faster in every game then an intel p4 of less than 3.4ghz or any athlon xp.

 

If you already have an a64 of higher than 2ghz and you have an x800xl/6800gt or lower then your upgrade money would be best spent on a new video card or dual core rig.

 

Note: The above recommendations are for HL2/CS or DOD Source only. Other games may or may not benefit from a cpu or video card upgrade. If you'd like to know which games benefit from which upgrades then feel free to pm me and I'll give you answers based on reviews I've read/my experiences. For example, NFS:Most Wanted is almost completely video card limited.

 

Other relevant system specs not mentioned above:

DFI NF4 Ultra-D

2x 512MB OCZ PC3700 EB running at 3-2-2-11 for all tests

21" Dell P1130 @ 100Hz for all tests

 

 

Edited for grammar and spelling.

Edited December 16, 2005 by Cujo

[ScubaDan]

Thanks for posting that up Cujo, a very interesting read.

[Vovik]

Something I knew, but I appreciate the time and effort spent posting anyway.

[anonymo]

Very helpful...nice read too...confirms everthing I've always thought...

[Cujo]

thanks guys. it only took about an hour to do cause each test was only about 10 mins plus reboots in between. i was actually a bit surprised at the results when overclocked. i figured i'd still be totally cpu limited. i will post up more tests with an ati x550 as i've sold my x800xl and won't have it after this evening. i'm borrowing the x550 froma friend temporarily until my x1800xt gets here. i'll post more results with it as well.

[random_n00b]

You can always check that +showbudget graph. On my machine, in all but the most complex scenes, the "client animation" bar is the largest - which is a CPU bound resource. Not surprising considering my now-humbled Athlon XP 2800+ is running the task, but I don't really mind if the framerate isn't always top notch.

 

I wanted to buy a new sound board (you know - to upgrade something that isn't broken for once), but now I have to get a monitor since this one is getting an awful flicker in the bottom-right and top-left corners. Hopefully my college has some they'd like to get rid of for cheap .

[*BiGBonES**]

In a couple of days I'll have my new upgrade results posted up. I'm going to do tests for:

 

Current rig (i'll give specs then)

Current rig with new graphics card

Current rig with new graphics card and processor

Current rig with new graphics card, processor and sound.

 

Can't wait to see the results, as I'm expecting them to boost the fps by a reasonable amount. First up is taking the computer home,buying a wireless internet card and then downloading source again... which could take a while

[Vovik]

just something to throw in here

 

I tested my CSS in Video Stress Test. I used all the different settings: resolutions (from 640x480 to 1600x1200), detail levels (lowest to highest, 6AA til no AA, 16AF til bilinear) EVERYTHING. Anyways so my fps are the SAME for all of this stuff (ranges only vary 6fps but inconsistent with loads).

 

Only, when I overclock my cpu to 3.3Ghz I see the difference compared to non overclocked cpu.

[Cujo]

odd that you barely saw a difference. the stress test tends to be very gpu bound, especially compared to in game play. at least with my x800xl there were very big differences between 800x600 and 1600x1200 with everything on highest. maybe your vid card is just that good. also, do you have any configs running? most of them take away a lot of eye candy and would completely skew your results no matter what video settings you enter.

[Vovik]

odd that you barely saw a difference.  the stress test tends to be very gpu bound, especially compared to in game play.  at least with my x800xl there were very big differences between 800x600 and 1600x1200 with everything on highest.  maybe your vid card is just that good.  also, do you have any configs running?  most of them take away a lot of eye candy and would completely skew your results no matter what video settings you enter.

<{POST_SNAPBACK}>

i used to take away eye candy with my 9500pro, with this one i don't, all i have in config is crosshairscale 2000, errorcheck, and update rates

[Cujo]

i guess your card is just that good then.

[Vovik]

i guess your card is just that good then. 

<{POST_SNAPBACK}>

While we at it. I went back to oc'ing my comp to 3.3Ghz. I get improvement in frames but the studdering remains. I'm am almost certainly positive it's because of memory. Because I OC my comp it turns it from PC3200 to PC2500 and it runs not at 400Mhz but at 266Mhz. I checked the latency with EVEREST Home Edition and it claims 3-4-4-something. Now, I'm NO expert at memory but that just seems a little slow. Since Cujo you say that HL2 is CPU and RAM dependend this gotta be the case. I know you said in my previous case that new faster memory won't give me much of results it's probably not much results in generic testing of fps. But, just maybe, it will give me more stability. It's obvious that I don't have fps problem but studdering is bugging me

[Cujo]

if you have only 512mb of ram then you'll get pagefile access stuttering. your ram is probably not running slower it's probably just auto ratioing. that's a good thing cause it most likely can not handle the fsb like your cpu. dl a program called cpuz to verify this.

 

btw, if you're overclocking then your fsb is going up. your fsb should not be at 200 if you're overclocking. hmm, actually, do you have a 133fsb cpu or a 200fsb cpu?

[Vovik]

if you have only 512mb of ram then you'll get pagefile access stuttering.  your ram is probably not running slower it's probably just auto ratioing.  that's a good thing cause it most likely can not handle the fsb like your cpu.  dl a program called cpuz to verify this.

 

btw, if you're overclocking then your fsb is going up.  your fsb should not be at 200 if you're overclocking.  hmm, actually, do you have a 133fsb cpu or a 200fsb cpu?

<{POST_SNAPBACK}>

I have 1Gb, Cuj. Mine's a 200fsb CPU. I OC fsb to 235. BIOS, however automatically lowers memory clock to 266 (or 133x2). If I manually keep it at 400 (or 200x2) I'll get huge amount of artifacts in my programs especially games. I can force it to 400 up to 3.0Ghz, however now I run at 3.3Ghz and forcing RAM to 400Mhz would give me those artifacts. There's no option to clock my RAM higher in BIOS btw.

[Cujo]

forcing it to 400 is actually forcing it to remain in sync with your fsb 235. when it drops to 133 it's actually dropping to a ratio of your fsb to ram. if you have a ratio of 5:4 somewhere then try setting that.

 

i've never heard of oc'ed ram causing artifacts... it can cause file corruption and can prevent you from booting to windows.

 

what mobo do you have? can you take some pics of some of your bios pages with your overclocking options? run cpuz and check memory tab where it says frequency. below that it also says fsb:dram. take a screenie of that and the cpu tab and post it here as well if you can. actually on second thought maybe just pm me as this is off topic.

Edited December 21, 2005 by Cujo