P4c800E-D Performance tweak tips

[mikead_99]

Droop mod, got it. Will install it if necessary but I'd like to avoid it if possible. The board is going under the prom and don't like the idea of seal string sitting on my mod. No problem until I remove the mounting bracket, but I'll see how it goes first.

Vmem, gonna happen, np.

My real question is regarding DRAM burst length, DRAM Idle, and DRAM Refresh. I think I have a good handle on the rest of this boards config options. From what I've gathered, DRAM burst > = better, Idle > = better, Refresh < = better? Not sure on those yet though. Oh, and yup, will bump up vAGP, probably to 1.7V. I'd also be interested in hearing any other "rules of thumb" for this board, kind of like the A-N-A-D-D "rule" for Abit 875 boards. Funny, but that config never impacted my o/c'ability with Abit, I got away with SR-E-A-A-A, but I know it affected many. Any words of wisdom for me?

Thanks, Mike.

[The Stilt]

My race settings for P4C800:

DRAM Burst Length: 4 (some benches seem to like 8 faster :confused: )
DRAM Idle Timer: Infinite
DRAM Refresh Rate: 7.8µsec
Spread Spectrum: Disabled
MPS Revision: 1.4

Then set Performance Mode to Turbo, and boot up to windows with 200FSB. Then raise fsb to max and run benches ;)
This way you get additional turbos, equal to 70MHz boost.

Too bad it only works with 1:1 divider :(

[Malves]

DRAM Burst: 8 - Longer = more data to be read and/or written.
DRAM Idle: Infinite - reduces the number of refreshes = increased bandwith.
DRAM Refresh: 64usec - > is better.

[mikead_99]

OK, thanks for the response guys. I'll of course play a little to optimize for my own config., but it's always nice to have a starting point beyond the default settings.

edit: I was going to "opimize", fixed spelling. :rolleyes:

[pkrew]

I have everything set the same as Malves, except I have the Dram Idle set at 0T. I tried Infinite, but it seemed to hurt my results in 2001se. I was under the impression that Infinite would be better also, but it just didn't happen. I also have spread spectrum enabled. It didn't seem to affect the overclock or the speed so I just left it alone.

[brwmogazos]

hi guys...been playing with mine last night by the way and used

DRAM Burst Length: 8 from 4

DRAM Idle Timer: Infinite from 8T

DRAM Refresh Rate: 64usec from 7.8µsec

that gave me a second less in super pi 1MB to 36secs from 37 at my current fsb i am testing of 275fsb-220mhz ram @ 2 2 2 5 :)

am getting 5921
5920 in sandra pro 2004...in buffered and

3113
3175 with unbuffered provided i have selected the correcti options :p

does those benches look right?

[Sir Random]

Great thread :D
I've often wondered about these settings but never touched them. I'm gonna try all the settings suggested here, and test with Super-Pi and Sandra. I'll post the results here.

EDIT:

Ok,
I've tried a few settings and these are best.

DRAM Burst Length: 8 (I already had it at 8 so I left it)
DRAM Idle Timer: Infinite (from Auto)
DRAM Refresh Rate: 64µsec (from Auto) I also tried 7.4u but Pi was slower.

Running at 220fsb, Super-Pi was still 38sec, but the 1 second split happened earlier (loop 8 - 17s) with 64u.

Drystone: 11410 from 11084
Whetstone: 4674/8273 from 3794/8289

Damn, I forgot to check if I could get down to 36secs at 3.9GHz :eek: :D

[Soulburner]

Originally posted by Sir Random
Drystone: 11410 from 11084
Whetstone: 4674/8273 from 3794/8289
Are those accurate?

Those numbers can fluctuate a lot. I have seen anywhere from 9,800 to 10,600 MIPS just from hitting refresh a few times...try a different benchmark?

[cpulloverclock]

for me

DRAM Burst Length: 8
DRAM Idle Timer: 64T
DRAM Refresh Rate: 64µsec
Spread Spectrum: Disabled
MPS Revision: 1.4


I'll try again with infinite and dram burst at 4

[pkrew]

I have to agree with Soulburner on that. I can't see how you could really tell if you have any improvement using Sandra. Maybe the memory score, but even that would be suspect. I'd use superPi and 2001se But I did change to infinite and shave 1s off my superpi time from 37 to 36s

[Sir Lancelot]

OK, have anybody reached a level of agreement here?

After reading this, I still haven't met the optimised settings...

[mikead_99]

DRAM Burst Length: 8
DRAM Idle Timer: Inf
DRAM Refresh Rate: 64µsec
Spread Spectrum: Enabled

Back away from those if stablilty is an issue, probably starting with Spread Spectrum.

[[AK]Zip]

Under MPS Revision what is the difference between 1.1 and 1.4? Any input would be great.

Also is everyone runnning the stock cooling on the northbridge? If not what are you running and what are your temps either way.

-Alex-

[mikead_99]

Originally posted by [AK]Zip
Under MPS Revision what is the difference between 1.1 and 1.4? Any input would be great.

Also is everyone runnning the stock cooling on the northbridge? If not what are you running and what are your temps either way.

-Alex-

MPS has to do with power save standards. I don't think it has anything to do with stability while running normal, though I could see different versions causing problems when trying to come back from standby or whatever. I just left mine at whatever the default is.

For nb cooling I use the Swifty nb hsf. Can't tell you the temps, no real way to check 'em, but I can tell you the board has let me go 265 fsb 1:1 2-2-2-5 with all the other settings as aggressive as possible. Might have helped, might not have, never tried with the passive stock hs.

[[AK]Zip]

Originally posted by mikead_99
MPS has to do with power save standards. I don't think it has anything to do with stability while running normal, though I could see different versions causing problems when trying to come back from standby or whatever. I just left mine at whatever the default is.

For nb cooling I use the Swifty nb hsf. Can't tell you the temps, no real way to check 'em, but I can tell you the board has let me go 265 fsb 1:1 2-2-2-5 with all the other settings as aggressive as possible. Might have helped, might not have, never tried with the passive stock hs.

Probably does help since I can actually test my temps. with a temp reader with a small sensor on the end which I can get all the way under the HS and temps read 46*c idle and almost 57*c full load. I would add about 3*C for the temp of the core itself. I will be adding a waterblock to it soon since that will cool better then any air cooled HSF combo and will cost me less since I already have a water cooling system in the computer. Will let everyone know how it goes.

-Alex-

[SPL15]

My setup gives best performance with:
Burst length ~ 8
Idler Timer ~ 60T
Refresh ~ 64u
Spread Spectrum ~ OFF
PCI latency timer ~ 96 (I have no PCI cards so i set it sorta high for better bandwidth for other devices using that bus)
Graphics app size ~ 256MB
ICH Delayed Trans ~ Disabled
DRAM Freq ~ 400 (1:1)
I makes sure all devices/ports/features i dont use are disbled in my BIOS including IDE slave ports, Promise controller, MIDI, Serial's, Parrellel, Audio, APM, etc.
I also only enable 2 USB ports and disble all legacy devices.
Also manually typing in what you want is quicker than hoping the board autodetects correctly (CDRoms, PS2 mouse support, etc.)

Automatic settings are for people who dont know what they are doing, use manual settings!! I just wish this board had Abit's BIOSes, That would be THE board to own!!

MPS I think stands for Multiple Processor Support if I remember right, but I also remember that people didnt see jack :banana: :banana: :banana: :banana: :banana: of difference between the two.

Here's something odd on my system, I get much lower and unstable benches with Vagp set high (1.7V, 1.8V). I get the best stability, and benches with it set to 1.5Volts! I'm not sure if it's the stock HSF on the northbridge or what, but I've tested and retested it to make sure cause I didn't believe it myself.

Also, I have my AGP/PCI busses OCd to 80.00:40.00 and noticed quicker responses in windows (Clicking menus, etc) and also gained a few hundred in Aqua and both 3dMarks and 2 Mhz in my FSB.

I got better scores with performance mode set to standard with the FSB clocked 2Mhz higher than with turbo enabled, but with 2 Mhz lower.

I'm using corsair DDR550 pair, so I go tmy mem timings 3448 and run 2.65Volts to them. Since my FSB isnt high enough to need the 2.75volts, i have them at 2.65.

My system runs 2.6C @ 3.38Ghz with 260FSB. My clocks are purely CPU limited (SL6WS suck,4th one best OCer of the bunch) My CPU voltage is 1.73 average (1.68-1.78). I did droop mod (Helped a lot) I have it set to 1.7V in BIOS but as you know the droop mod.....

I had a great 2.4C OCer to test board limits and I got 291FSB 100% stable running benches outdoors middle of Janauary with large sqaure fan blowing into side of my computer while open. I had Vagp at 1.5Volts, Vmem at 2.85Volts and the rest of the settings the same as above. This corsair mem is expensive, but it works good for me. most people i hear get around 285 max with this memory.

Also the settings I listed are 100% stable 100% of the time and give me the best benches in all of the benches i run.

****ONE TIP: VERY IMPORTANT FOR RELIABILITY****

There is a mosfet below the DIMM slots, to the right of the AGP slot that gets dangerously hot when this board is OC'd. Find some way to heatsink that sucker. You'll be able to find it pretty easy by feeling them. There are three in a little cluster and the lower right one has way too much current flowing through it to be safe, so it needs to be heatsinked well.

Best thermal epoxy I've found is regular old JB weld costs 5 bucks for a lifetime supply pretty much.

Hope this helps!

Edit:

My benches are:
55000 Aqua
7741 3dMark03
7003 PCmark04
22012 3dmark03
No benches had any artifacting.
operating system tweaks helped bench scores greatly

[Sir Lancelot]

What does the "ICH Delayed Transaction" do?

I have Intel RAID (southbridge)...so, what setting should I have?...Or doesn't it have nothing to do with RAID..?

Anyway...what setting should I put: Enable ou Disable?


Concerning the timmings above quoted...I don't see any common sense...eheheheheheh...it's crazy...

[SPL15]

I get better slightly better benches with ICH delayed transaction disbled, but it was a slight difference and could have been just natural tolerances.

From what I remember, it improves performance of ISA connected peripherals (Intel's Site) BUT I believe this board has no ISA devices, so I have no clue on what they are talking about.

I really have no idea what it does really. I dont have anything connected to my PCi bus and dont use the promise raid controller. If I remember right, Intel's RAID is connected to the north bridge which makes ICH delayed transaction have no effect on those drives.

I only enable things I know I use, which means I disable anything and everything that I know i dont use or think I dont use, but you still gotta use common sense...

Do some experimenting, maybe you can clear this up for us.

[Sir Lancelot]

The purpose ICH Delay transaction is to "activate" the latest standard for the PCI's. Version 2.1


Bye

[SPL15]

Wow.. looks like i was halfa$$ backwards. Thanks for the info. I'm thinking this still doesnt apply to me since I have no PCI cards installed?? Either way, I enabled it and my benches are still pretty much identical. but it could be a much different scenario with PCI cards installed, like the sound car i'm buyin.

[OC_Newbee]

This is very great info since I just switch over to the P4C800e-Deluxe. I hope this board serve me as well or better than my Epox 4pca3+ rev 2.0.

[SPL15]

Update on the ICH delayed transaction setting.

In my benchmarks I noticed absolutely no difference what so ever, but I ran Folding@home last night the whole night and my times have reduced considerably?!?!

This is the only setting i've changed and the last 5 folds i've done have taken only about 6 hours to do instead of about 8+ hours. I'm 100% confused about this, and still think something else is up.... But so far folding runs much faster.

Can any of you guys confirm this?? I still cannot see a anything that PCI delayed transaction would have to do with processor FPU speed. Maybe i just happened to get 5 small folds in a row???

Whatever the cause, i'm kickin some a$$ in folding with it enabled.

Also, i removed the IHS and reduced CPU temps by 7 degrees celcius. My max temp at 3.45GHz 100% load is 96F (From board) with voltge at 1.725 Air cooled (Modified Zalman 7000Cu with 92mm Vantec Tornado).

[mista_chewey]

Originally posted by SPL15
Also, I have my AGP/PCI busses OCd to 80.00:40.00 and noticed quicker responses in windows (Clicking menus, etc) and also gained a few hundred in Aqua and both 3dMarks and 2 Mhz in my FSB.


how did you do this???
my harddrive corrupts as soon as i use 76:36
do you use SATA HD?

[mikead_99]

Originally posted by mista_chewey
how did you do this???
my harddrive corrupts as soon as i use 76:36
do you use SATA HD?

You really can't get away with this using the Intel SATA controller. Maybe using the onboard Promise, or a PCI add-on controller. SATA drives are pretty sensitive to the bus speeds.

[Wingz]

my mate uses a scsci hard disk to overcome the problem.

[mista_chewey]

oh ok so if i use the onboard promise controller for my ide drives i should be fine

[SPL15]

I have mine on the Intel ICH5R raid controller. I've disabled ALL (yes ALL) devices, ports, features, etc not used. Not sure what your setup is, but I've got some decent harddrives(WD 74GB Raptors).

You may not be able to OC the buss because your hardrives can't handle it. I did compare tests, and I got decent improvements in everything with the busses OC'd. And another thing, I get white speckles if I DON'T have the busses OC'd at my max stable/Artifact free Video card settings (Modified 9800XT). I've got some other MoBo mods though that significantly increase my OCability on my whole system, but I stil could easily with no problems at all OC the PCI/AGP busses to max setting.

I Really see no point in increasing Vagp setting as this only increases chipset temps in my setup and decreases max OCability (For Me).

I personally like the intel sata better then the promise. intels boots MUCH MUCH MUCH faster. I dont even get to see it detecting the drives before my monitor brightens up. I notice no typical raid lag boot time and my computer boots in less than 20 seconds.

I'm not exactly sure how the intels raid is implemented, but it is obviously much less effected by buss ocing than the promise, because I did get errors using the promise at the max PCI buss setting.

My theory is that regular hardrives with the sata bridge are already using pretty cheap interfaces and this added transaltion stage cant handle added frequency because it is already running at it. The raptors are essentially SCSI disks in a IDE package using high end IDE electronics with a good sata bridge which gives it's bus the headroom needed to run at higher buss frequencies. You always get what you pay for in the electronics industry and usually you dont even get that. there are absolutely no free lunches in the electronics industry, no matter how much the marketing department lies. Biggest liars in the world are not lawyers, they are people in marketing departments, and markineers (Marketing specialists who think they are engineers). Just some food for thought:D

[oc-rookie]

I can't get go one blink beyond 33 with my raptors off the intel controller.