BF2S Forums my GTX 280 story

CaptainSpaulding71 wrote:

what version of drivers are you guys using?

do you recommend i somehow clean out my existing drivers before reinstalling the new ones rather than relying on the install tool for the drivers to do that?  does that make a difference.

i'll nuke my bios tonight and reset it to defaults.  will post update later

thanks to all for your inputs.  appreciate your time.

Meh, I always just overwrite older drivers when I install new ones, never bother to clean.

Scorpion0x17 wrote:

CaptainSpaulding71 wrote:

what version of drivers are you guys using?

do you recommend i somehow clean out my existing drivers before reinstalling the new ones rather than relying on the install tool for the drivers to do that?  does that make a difference.

i'll nuke my bios tonight and reset it to defaults.  will post update later

thanks to all for your inputs.  appreciate your time.

Also, and more to the point, what CPU mobo and RAM are they using?

To me it looks, as you yourself suspect, that the low FSB on your CPU, even when it's OC'd, is causing the CPU to bottleneck the GPU.

You can have the fastest GPU evar and it'll still score low if the rest of the system can't feed it data fast enough.

It's still a fucking QX6700. He gets 3.3k CPU score on stock, and that's plenty. His graphics scores are quite low, though.

Something has to be wrong. I beat his stock score with a supposedly weaker board and a dual core running slower.

Freezer7Pro wrote:

Scorpion0x17 wrote:

Freezer7Pro wrote:

It's still a fucking QX6700. He gets 3.3k CPU score on stock, and that's plenty. His graphics scores are quite low, though.

The proof is in the pudding:

CaptainSpaulding71 wrote:

qx6700 @ 2.67 w/ gtx280 at stock speeds = 11462
qx6700 @ 3.2 w/ gtx280 at stock speeds = 12480
qx6700 @ 3.2 w/ gtx280 @ 670/1442 core/shader = 12536
qx6700 @ 3.4 w/ gtx280 @ 670/1442 core/shader = 13639

It may not specifically be the CPU that's the problem, but it is definitely related to it.

Yes, the CPU isn't capable of utilizing the true power of the 280, but it shouldn't score this low, either.

Because it's being bottlenecked. As I said before, you can have the fastest ever GPU ever made and it'll still score low if the rest of the system isn't feeding it data fast enough.

Of course, there are several things between the CPU and GPU that could also be hindering things, but the 1000pt increases from over-clocking the CPU suggest it's more CPU related than anything else.

update: 

i ran some spectrum analysis on my bios settings by running various multipliers and FSB settings along with increasing my stock voltages.  also, i downloaded the latest 1.77.41 drivers from nvidia (dated june 26). 

the analysis i found was that if i kept at stock multiplier of 10 i could only get FSB=1300 = 3.25GHz to boot windows.  anything higher than that would either crash in bios or in windows upon boot.  i could run FSB lower (1066 was very reliable as was 1133).  The score here for 325 FSB (1300 MHz) was 13331 with no overclock on the gfx card.

then i bumped the multiplier to 11 and 12 and ran the same bumps in FSB freq.  x11 was a disaster - nothing seemed to work properly.  x12 was impossible to get beyond 1133 = 3.4GHz.  Then i tried x13

x13 turned out to have a sweet spot for me and i got FSB of 1100 = 3.58 GHz which is much better overclock in terms of freq than i was ever able to achieve before.  this enabled me to get 14407 at stock gfx speeds and 15459 when i boosted the gfx core clock to 670 from 602.

cpu temps were ok on idle (45C) and increased to 63C on load which is a bit hot for my taste.  gfx temps hit a high of 83C during the runs.

my changes in bios for 680i board (p32 bios) were:

+ multiplier = 13
+ CPU = 1.45V
+ FSB = 1.4V
+ Mem = 2.2V
+ FSB = 1100MHz = 3.58GHz
+ Mem = 1133MHz
+ unlinked mem and cpu buses

i'll still play around to see if i can get 15k on stock gfx timings.  maybe i can try bumping the fsb up a bit more at 13x multiplier.  will report back soon.

again - thanks for the comments so far they are a big help

CaptainSpaulding71 wrote:

update: 

i ran some spectrum analysis on my bios settings by running various multipliers and FSB settings along with increasing my stock voltages.  also, i downloaded the latest 1.77.41 drivers from nvidia (dated june 26). 

the analysis i found was that if i kept at stock multiplier of 10 i could only get FSB=1300 = 3.25GHz to boot windows.  anything higher than that would either crash in bios or in windows upon boot.  i could run FSB lower (1066 was very reliable as was 1133).  The score here for 325 FSB (1300 MHz) was 13331 with no overclock on the gfx card.

then i bumped the multiplier to 11 and 12 and ran the same bumps in FSB freq.  x11 was a disaster - nothing seemed to work properly.  x12 was impossible to get beyond 1133 = 3.4GHz.  Then i tried x13

x13 turned out to have a sweet spot for me and i got FSB of 1100 = 3.58 GHz which is much better overclock in terms of freq than i was ever able to achieve before.  this enabled me to get 14407 at stock gfx speeds and 15459 when i boosted the gfx core clock to 670 from 602.

cpu temps were ok on idle (45C) and increased to 63C on load which is a bit hot for my taste.  gfx temps hit a high of 83C during the runs.

my changes in bios for 680i board (p32 bios) were:

+ multiplier = 13
+ CPU = 1.45V
+ FSB = 1.4V
+ Mem = 2.2V
+ FSB = 1100MHz = 3.58GHz
+ Mem = 1133MHz
+ unlinked mem and cpu buses

i'll still play around to see if i can get 15k on stock gfx timings.  maybe i can try bumping the fsb up a bit more at 13x multiplier.  will report back soon.

again - thanks for the comments so far they are a big help

AH....

Exactly what RAM have you got?

Mek-Stizzle wrote:

Who cares about 3dmark scores, honestly? Do some real world tests.

Finally, someone gets it.

Scorpion0x17 wrote:

Poseidon wrote:

Mek-Stizzle wrote:

Who cares about 3dmark scores, honestly? Do some real world tests.

Finally, someone gets it.

The difference with so called 'real world tests' and benchmarks is that benchmarks quantify the performance - with a 'real world test' - by which I assume you mean "go play some games and tell us what your 'average' fps is" - you're only ever getting a purely subjective indication of performance.

Exactly. I wish more games had benchmark tracks.

CaptainSpaulding71 wrote:

Scorpion0x17 wrote:

AH....

Exactly what RAM have you got?

i'm running the Crucial Ballistix Tracer 1066 sticks.  i have 4 1GB sticks populating all 4 slots.  some people claim that if you remove 2 sticks and just go with 2 slots you might get better scores.  i'm running vista ultimate 32.

That is true. Now the problem with you not hitting good FSB is that you run your RAM:FSB ratio is too much for the ram. Change it to lower (underclock the ram) and then try more.

Slayer wrote:

first off, good buy Same happend to me with a free 9800GTX, a free Raptor 140GB(?) and a free Samsung 226BW at some german shops

13k in 3dmark06 with this setup and already overclocked? Somethings really wrong here.

I benched the following systems on saturday:

Intel C2D E8500, 2GB Corsair TwinMX DDR 800Mhz Cl4, Asus P5ND-SLI, Sparkle 9800GTX, no ocing ----> 12516
Intel C2D E8500, 2GB Corsair TwinMX DDR 800Mhz Cl4, Asus P5ND-SLI, XFX 260GTX, no ocing ----> 13957
Intel C2D E8500, 2GB Corsair TwinMX DDR 800Mhz Cl4, Asus P5ND-SLI, XFX 260GTX, no ocing ----> 15149
Intel C2D E8500, 2GB Corsair TwinMX DDR 800Mhz Cl4, Asus P5ND-SLI, 2 x XFX 260GTX, no ocing ----> 17003
Intel C2D E8500, 2GB Corsair TwinMX DDR 800Mhz Cl4, Asus P5ND-SLI, XFX 280GTX, no ocing ----> 15916
Intel C2D E8500, 2GB Corsair TwinMX DDR 800Mhz Cl4, Asus P5ND-SLI, Saphire HD4850, no ocing ----> 12636
Intel C2D E8500, 2GB Corsair TwinMX DDR 800Mhz Cl4, Asus P5ND-SLI, Saphire HD4870, no ocing ----> 14291

and the same settings but 4GB to 8GB...

edit: why the fuck wont the links work

Thanks again Slayer for the comparison points.  I managed to do some more tests and tweakings.  i found out that my mem was seriously underclocked with linked and synced since FSB was quad data rate and mem is dual data rate resulting in a 2:1 split so FSB of 1200 would mean mem is at 600 when my mem is rated at 1066.  the BIOS wording is really confusing here.  Secondly, my board is noted for having issues with FSB overclocking above 1200 FSB and i am unable to get FSB to be reliably above that on any of the multiplier settings.  Lastly, i have an older quad core QX6700 so it has it's own limitations of 1066 FSB and smaller cache than some of the newer processors.

that said, here are some updated numbers of experiments i had.  maybe this could be of use to others who are in the same boat.

2.66GHz:  10x FSB=1066, mem=1000 @ 5/5/5/15/2T with stock gfx timings = 12812 3dmark06
3GHz:       10x FSB=1200, mem=1000 @ 5/5/5/15/2T with stock gfx            = 14016 3dmark06
3.3GHz:    11x FSB=1200, mem=1000 @ 5/5/5/15/2T with stock gfx            = 14835 3dmark06
3.58GHz:  13x FSB=1100, mem=1000 @ 5/5/5/15/2T with stock gfx             = 15581 3dmark06
3.58GHz:  13x FSB=1100, mem=1000 @ 5/5/5/15/2T with gfx at 650 core clk = 16015 3dmark06

NOTE:  i'm boosting voltages on Vcore to 1.45, FSB=1.5, mem=2.2 and SPP=1.5.  if i pushed FSB more above the values listed, it would result in crash in either bios boot or windows boot.  i've run probably over 100 different settings and only listed some.

additionally when i compare # with ORB, it seems that many of the results there are with pretty high overclocks as well.

so finally i'm over 16k with a pretty hefty overclock though.

bottom line is that i believe my board and processor are bottlenecking this particular card.

CaptainSpaulding71: I was looking back at your original settings:



and, firstly note that I don't know CPUZ, and I've only a smattering of practical experience setting up Core2 systems, (and maybe you've now put this right), but something just doesn't look right to me...

Can anyone else confirm my suspicion that the Core Speed is wrong?

(I primarily want to check my understanding of how it works is correct)

Also, CaptainSpaulding71, would you mind posting CPUZ screenies of your current configuration? (both CPU and Mem tabs, please)

Last edited by Scorpion0x17 (2008-07-04 13:47:33)

The multiplier is wrong, way off. Should be 10, not 6.

Scorpion0x17 wrote:

Poseidon wrote:

Mek-Stizzle wrote:

Who cares about 3dmark scores, honestly? Do some real world tests.

Finally, someone gets it.

The difference with so called 'real world tests' and benchmarks is that benchmarks quantify the performance - with a 'real world test' - by which I assume you mean "go play some games and tell us what your 'average' fps is" - you're only ever getting a purely subjective indication of performance.

So can you convert 3dmark score to fps on any game?