Help me Tweak XP/2003 server to hit 20k 3dmark2001

[Radelon]

Well, I'm about spent, here's a little background information:

Mobile P4 1.6m @ 3.06ghz (12x255)
Abit IC7 /w 025 beta
2x256 OCZ PC3700 Gold running 1:1 (2.5-6-4-3)
Powercolor Radeon 9700Pro /w Zalman HP80A (430/350)
2x40gb IBM 60GXP in Raid0 on Adaptec PCI Raid
Windows 2003 Server

First off I started with the 1.3 official bios and the Radeon at stock cooling at 400/350. I use Rage3dTweak to tweak the 3.4 cats to fastest performance and such. At this point I was using 2x256 GeIL PC3200 Ultra in a 5:4 (257fsb on the proc) config running 2-6-3-3 timings Performance Mode set to Fast. This is the score I achieved: 18,233 marks (http://service.futuremark.com/compare?2k1=6581407).

I knew I could do better than that, and when I got the OCZ PC3700 Gold I made a run at 254fsb 1:1 on the same config, Performance Mode Disabled. I achieved 18,316 marks (http://service.futuremark.com/compare?2k1=6609620).

Dissappointed I went for a new bios on the Radeon9700Pro. After the flash I made a pass with 255fsb 1:1 (2.5-6-4-3) and got 18,869 marks (http://service.futuremark.com/compare?2k1=6612376).

At this point Abit produced the 025 beta bios and I was on it like stink on ****. This helped me out a ton. 255fsb 1:1 (2.5-6-4-3) Game Acceleration Turbo, CPC Disabled got me to 19,303 marks (http://service.futuremark.com/compare?2k1=6617138).

I then decided to go to a Zalman HP80A and a volt mod and now I'm currently at 254fsb (2.5-6-4-3 timings) Game Accel Turbo, CPC Disabled, Radeon at 430/350 and I scored 19,843 marks (http://service.futuremark.com/compare?2k1=6647756).

Granted there were a few other runs with different scores along the way, but these were the significant ones. But now I'm stuck on my run for 20k. I've tried closing all background programs, including explorer.exe and putting 3dmark in realtime priority. I have FDC, USB, Audio, Firewire all disabled in the bios. Nothing seems to push me any higher. I want 20k so bad I can taste it now.

Somebody please help me be the first Mobile to hit 20k. I've come too far not to reach my goal.

[Ardy]

How about a Radeon 9800Pro :yummy:

[ThugsRook]

which NorthBridge Strap are you using?

533 is the fastest
667 is next
800 is the slowest

awesome scores from that rig :rock:

:wave:

[pointreyes]

How many services do you have running in the Task Manager and do you have any of the server tasks on that may not need to be left on?

BTW: I get 13010 with a 9500 Pro on my non-oc'ed 2.8C proc on a Canterwood board. Have 56 processes running in the task manager, including Oracle 9i EE. I'm using w2k3 Enterprise Edition.

[Radelon]

Originally posted by ThugsRook
which NorthBridge Strap are you using?

533 is the fastest
667 is next
800 is the slowest

awesome scores from that rig :rock:

:wave:

using the 800 strap, i'm not sure if the other straps work with the ocz, but I'll try it.

[ThugsRook]

here's a better example of the performance difference....

3dmark2001

NB/s 800 = 15749
NB/s 667 = 16006
NB/s 533 = 16088

[Radelon]

only PSB800 will work.

[ThugsRook]

did you relax your other timings/settings?

if you can hold 533 or 667 and then tweak from there, you may end up with a better result.

[Radelon]

flat out won't run there now, used to, but not now

[ThugsRook]

was worth a try ;)

[Radelon]

Just pulled a 19,835 at only 250fsb....so I'm gonna try a few more things

[Radelon]

Just did it tonight here she is 20,003 marks (http://service.futuremark.com/compare?2k1=6665467) :rock:

[ThugsRook]

well ~ what did you do to get there?

:beer:

[pastorjay]

Congrats! That is just insane... especially with a p4m!

[ThugsRook]

ya know what i like about this thread?

it shows mem bandwidth @ 1:1 is the way to go for best gaming performance.
(even @ only 3ghz)

[oldfart]

The change from 1:1 from 5:4 did show a small increase with the addition of some very expensive ram that is only available in 256 Meg modules. What do you do if you want to run 1 Gig of ram?

In this config, he is comparing 257 5:4 and 255 1:1. Obviously, the latter will be faster. In most cases, the FSB difference between a 5:4 and 1:1 config is much greater.

Another point is the 1:1 Vs 5:4 test were done with 13 BIOS. In this version, 1:1 had PAT and 5:4 did not. With the 14 BIOS, they both have PAT. That would likely even them out if the 5:4 setup was running tight timings (which the OCZ can not do) Vs the slow timings that 1:1 has to run.

[ThugsRook]

at this point ~ i wouldnt be surprised if it isnt even a C1!

3ghz and he's pushing 20k in 3dmark?
he's showing that there is more power in ram/board then there is in cpu <period>

:wave:

[oldfart]

I use Rage3dTweak to tweak the 3.4 cats to fastest performance and such


Radeon at 430/350
This system is highly tweaked for 3DMark including an insane 9700P overclock. That is a big reason for how he got there. My score increases dramatically when I increase the 9700 clock speed. More than any other tweaking. Mine wont go anywhere near his speed. That makes the biggest difference in scores.

I've tested this enough and my benches are posted. 1:1 slow timings and 5:4 fast timings make litttle if any difference.

[ThugsRook]

i know i know, and i hate to argue with ya ~ but you arent using good enough ram :rolleyes: ;)

:wave:

[Radelon]

Originally posted by ThugsRook
i know i know, and i hate to argue with ya ~ but you arent using good enough ram :rolleyes: ;)

:wave:

Thugs, what put me over the top was reading your Abit IC7 post. I shut off the "disable unused pci clocks", I enabled "Video Bios Cache". Those 2 things gave me enough points to put me over 20k. I'm so excited about this.

Btw, this mobile I have is a C1m, and I got it off ebay for $60. Having good ram to match the peak performance of your processor is the way to go. If you have a processor that hits 300fsb, you want ram that you could run 5:4 for 250mhz.

I'll match this P4m up against a simular 2.4C any day. If we took and made the identical clone of my system, but with a 2.4C, both clocked at 250fsb, the P4M will wax it. Heck, it's not until 3.25ghz that someone can beat my 3dmark2001 score, and from that point I stay in 2nd up to 3.3ghz.

Also in 3DMark2003 (http://service.futuremark.com/compare?2k3=945286) I'm first up to 3.3ghz. I think there's a little more tweaking to do, but I'm very satisfied with my scores.

Thanks guys for your help and support......... I feel like :beer: :rock:

[oldfart]

Originally posted by ThugsRook
i know i know, and i hate to argue with ya ~ but you arent using good enough ram :rolleyes: ;)

:wave:
And just what ram SHOULD I be using?? Please LMK what 512 modules I can buy that will allow me to run 1:1 @ 257 FSB with tight timings. If I have to run cr@ppy slow timings, I'm not interested since it will provide no performance increase over what I have now.

[oldfart]

I'll match this P4m up against a simular 2.4C any day. If we took and made the identical clone of my system, but with a 2.4C, both clocked at 250fsb, the P4M will wax it
I'm not saying that is true or untrue. I'm just curious as to why it would be. What is different about the P4M that would make it faster than a P4C?

Both setups are 12 x 250...so nothing different there. Both have the same L1, L2. The only difference is the P4C has Hyperthreading.

[ThugsRook]

Radelon ~ glad we could help.
thats an awesome setup youve got there.

$60 for a used p4m c1 eh?
id like to get myself one of those to play with! :deviltail

[ThugsRook]

oldfart ~ im in the same boat you buddy. im looking for some good 512mb sticks myself for some 4:5 455dcddr action :rolleyes:

[Radelon]

Originally posted by ThugsRook
Radelon ~ glad we could help.
thats an awesome setup youve got there.

$60 for a used p4m c1 eh?
id like to get myself one of those to play with! :deviltail

Thx thugs. I'm not quite sure why it's faster than a 2.4C at same clocks, but it is. Some would speculate that the P4m's are faster clock per clock than it's desktop counterparts due to the fact that they have more processing power. Kind of like the P3's. But nobody can justify that completely.

[ThugsRook]

we noticed the same "speed" when we ran a P4m on P4B533 motherboards @ 225fsb 6 months ago.

unfortunately they didnt work on Granite Bay and we sold them off.
its nice to see they work on a IC7 ~ im now on the hunt for a used P4m C1 :deviltail



fun fun fun! Erik!?

[Radelon]

Originally posted by ThugsRook
we noticed the same "speed" when we ran a P4m on P4B533 motherboards @ 225fsb 6 months ago.

unfortunately they didnt work on Granite Bay and we sold them off.
its nice to see they work on a IC7 ~ im now on the hunt for a used P4m C1 :deviltail



fun fun fun! Erik!?

Remember, I did break BSEL5 off (I believe, to make it default to 200fsb).

[ThugsRook]

did you have to do that? or just something you wanted to do?

[Radelon]

Just something I did, I'm sure you don't have 2.

[oldfart]

How the heck did you get your 9700P to 430/350 :eyepop: ?? Did you remove the shim when you installed the ZM80? I cant get anywhere near that speed.

You might also want to check out the Cat 3.5s. They are supposed to have a small increase in performance.

Congrats on hitting your target. Nice job :beer:

[ThugsRook]

Having good ram to match the peak performance of your processor is the way to go.

If you have a processor that hits 300fsb, you want ram that you could run 5:4 for 250mhz.
you dont want a processor that can hit 300fsb unless you have ram that can hit 300fsb.

async is not the way to go. async has never worked the same as a 1:1 be it either possitive or negative.

1:1 is the way to go. especially now with PAT/MAM involved, it all works best @ 1:1.

i keep tryin to get ppl to understand this ~ but w/o actual benchmark proof its kinda difficult. your results sure do help tho :-)

:wave:

[Radelon]

Originally posted by oldfart
How the heck did you get your 9700P to 430/350 :eyepop: ?? Did you remove the shim when you installed the ZM80? I cant get anywhere near that speed.

You might also want to check out the Cat 3.5s. They are supposed to have a small increase in performance.

Congrats on hitting your target. Nice job :beer:

When I removed the stock hs, I did remove the shim. I also did a vgpu mod since it was really simple. I was only able to hit 400 core before the mod.

I will be trying the cat 3.5's tonight.

[SupermanCK]

hey...
i was wonder what kinda of SIMPLE mod did you do to your 9700pro?
if it is simple i might as well do it to mind...:D :rock: :beer: