ABXZone Computer  Forums

ABXZone Computer Forums (http://www.abxzone.com/forums/index.php)
-   Sticky Graveyard (http://www.abxzone.com/forums/forumdisplay.php?f=85)
-   -   My overclocking experience with Prescott 2.8GHz (http://www.abxzone.com/forums/showthread.php?t=65294)

Kawaiitenshi 02-05-2004 02:18 PM

My overclocking experience with Prescott 2.8GHz
 
At this thread I will post, as my time allows but hopefully daily, my overclocking experiences with retail (multiplier locked) Prescott 2.8E and Epox 4PCA3+ 875-based motherboard.

If you, my audience, have any questions please feel free to ask.

English is my third language so please forgive me any stupid errors.


Chapter 1 - Setup


I had read a few reviews talking about how the good old Northwood was still superior to regular folk but still I had no second thoughts as I removed my trusty 2.6HT Northwood, threw my new precious in, applied regular silicon paste thinly and bolted my huge copper water cooling block tightly to the board.

These additional components are installed:

-Enermax 520 watt PSU modded with beefier caps and variable resistor on +5v feedback

-4x512MB PC3200 memory, cheapest Apacer, works at 2.5/3/2/6 with all modules installed

-16x160GB Samsung harddisks (RAID0 + 2xRAID5)

-Radeon 9700 Pro +25% voltage modded (thanks to Xbitlabs!)

-Philips 7.1 sound card and digital TV tuner PCI cards

-additional gigabit ethernet card, cheap brand

-Highpoint RAID card for second RAID-5 array

-OS: Windows 2003, Standard Edition all patches applied from WU


Chapter 2 - First Impressions


Pumps on..monitor on.. hit the power switch. Lovely degaussing sound...phew! Nothing broken yet :)

MB recognizes the CPU as "Intel Prescott", future BIOS revisions will take care of this minor problem. Core voltage is at 1.500v, I dropped it back to 1.400v. +12v rail has dropped 0,04v, I will compensate it tomorror by adjusting the +5v feedback resistor on my PSU.

Oh my. The temperature is climbing, are all hoses fine?
Few seconds of anxiety pass but the climbing slows down before panic sets in. Temperatures are +4C above those I got from 3.0C Northwood and I haven't started overclocking yet!

I don't know how diode calibration works on Intel CPUs so maybe this is a BIOS glitch.

Next I will run dual Prime95s overnight to establish baseline stability and get myself an infrared thermometer for reference measurements and if/when tests pass OK...tomorrow we will see how far this beast can be pushed to - at 1.400v!

To be continued.. =)




Edit: added my OS

ps24eva 02-05-2004 02:21 PM

prescotts are supposed to be hotter. this is normal.=)

Kawaiitenshi 02-05-2004 02:31 PM

Well, actually they _shouldn't_ be (process shrink, lower core voltage) but rumor says Intel got bitten in the a-s-s by the strained silicon technology -> faster transistors but huge leak currents so at least this revision of Prescott won't be the new C300/P4-2.4 for air cooler masses :(

However, my journey won't even begin with an air cooler and I have a few aces in my sleeves if things get HOT ;)

OC_NEWBEE 02-05-2004 02:51 PM

Great News!
 
I have the same kind of motherboard, awaiting anxiously for your experience with the new 2.8E Prescott. Which revision of the 4PCA3+ do you have? What bios are you running for your board?

k0NG0 02-05-2004 03:04 PM

Quote:

Originally posted by Kawaiitenshi
Well, actually they _shouldn't_ be (process shrink, lower core voltage) but rumor says Intel got bitten in the a-s-s by the strained silicon technology -> faster transistors but huge leak currents
They've added cache, more pipeline stages, more logic for SSE3 and there's also an extra integer unit in there (for Yamhill no less, even though it may not use power).

There's also a new low-k dielectric that should decrease static power dissipation as compared to before. (I also believe they're using a new silicide on the polysilicon, but I'm not sure, anyway that would decrease resistance and thus some of the static power draw). So I would say it could be the added transistors rather than the process that's playing tricks on the power usage numbers. Or at least, that one can't attribute the increase to only one reason.

(I'm not sure how the gate leakage is for the strained silicon trannies. I know both Intel and AMD are thinking of high-k gate dielectric, but that's a few years into the future still.)

-kONGO

Kawaiitenshi 02-05-2004 04:16 PM

Small Update:
Played two hours of Firestarter (nice game) with Prime95 running, no problems whatsoever. Temperature is stable, +4C over 3.0C.

OC_NEWBEE: I have BIOS version 2D00, don't know about MB version (I bought it when it was brand new since I hated my P4P800 with it's crappy 865 so much).

hotcpu: I didn't buy it, I work at a large company and I got it for testing (well, actually begged for it ;)).

OC_NEWBEE 02-05-2004 06:05 PM

Quote:

Originally posted by Kawaiitenshi


OC_NEWBEE: I have BIOS version 2D00, don't know about MB version (I bought it when it was brand new since I hated my P4P800 with it's crappy 865 so much).


Bios version 2D00?
I'm not fimiliar with that version at all. The only bios that epox released, that I know of is the 07/03(4PC33703), 8/19(4PC33819), 12/16(4PC33C16) and the 12/25(4PC33C25). Unless the bios you have is a leak bios from somewhere, other than that what is the different between the version your running and the previous version?

EDIT: Never mind I found which version your talking about. Bios version 2D00 is the same as 12/25(4PC33C25.BIN). Thanks, waiting to hear more from yah on the prescott and Good luck!

Anarchi 02-05-2004 06:34 PM

Anyone know if the Prescott will work on an Abit IC7-G? Im using BIOS 2.0.

Does the Prescott use 0.13 or 0.09 silicon?


[EDIT] BIOS 2.1 is out now, it has prescott support:

BIOS ID:2.1


1. Updated CPU micro code for Prescott.
2. Changed default RTC year to 2004.
3. Updated Intel OSB logo.
4. Fixed an issue where a system with BIOS 20 will hang at post code "AF" with certain memory modules is installed.
5. BIOS compile date: 01/28/2004

Fraoch 02-05-2004 06:56 PM

Quote:

Originally posted by Anarchi
Does the Prescott use 0.13 or 0.09 silicon?

0.09. Which should have decreased the heat, but didn't, due to the chip's architecture.

FearFactory 02-05-2004 07:21 PM

Maybe a stupid question.... but.. Prescott is already out? I have an Epox 4PDA2+ i865PE rev 2.0, hope Prescott will work on this board.

XJ. 02-05-2004 07:28 PM

Quote:

Originally posted by Fraoch
0.09. Which should have decreased the heat, but didn't, due to the chip's architecture.
Actually, it's due more to process limitations (leakage current) than the implementation of the architecture, AFAIK.


Edit: Oops, as others above have already pointed out :blush:

Kawaiitenshi 02-06-2004 08:04 AM

Part two!
 
Chapter 3 - Heat is not my friend


Ten hours of dual Primes passed so it's time to crank up the speed. I rebooted and set core voltage to 1.450v, saved settings and returned to BIOS.

Prescott seems to be very sensitive to core voltage increases, temperature went up one degree from just +0.05v increase, at default speed and water cooling!

I upped FSB to 233MHz and set memory divider to 5:4. I have four double-sided sticks so it's a very heavy load for the memory controller even at default clock speed. Increasing vDimm actually makes things worse since at higher voltage even less current gets through to physical memory chips!

14*233=3260MHz should be easy and it was, I played more Firestarter and had no problems.

+12v rail dropped a little again, this CPU really does suck up juice!

A rumor from Intel says that the 2.8GHz model originally had a default core voltage of only 1.30v but yields were so horrible that even after at least two core revisions they had to use voltages originally *reserved* (should Tejas fail and large 4GHz+ Prescott yields needed) for 4.0-4.2GHz (!) processors.

Let's test this theory a little, shall we?

Speed/voltage/temp after 15min of dual Prime 95s.
Please note that on air cooling the differences would be larger due to much smaller deltaT.

3.26/1.450 42C
3.26/1.500 44C
3.26/1.550 47C

Need I comment? On my old Northwood systems the differences were much smaller.

Time for a bigger pump - old one had a flow of 700L/h, new one has 1200L/h. I will tell which one I used before results.

Back to BIOS, new FSB=250 divider stays at 5:4.
Boots fine! :D
Old pump: 3.50/1.550 50C (ouch!)
New pump: 3.50/1.550 47C

One hour of Firestarter+Prime95 passed with no problems.
I hoped for a better improvement, maybe my hoses are too small or my radiator isn't up to task anymore?
Time for a few beers, I need more courage for the next chapter..


Chapter 4 - Real men fear no +20% overvolt, but from what?


I ghosted my boot partition just in case and went back to BIOS.
New voltage: errr what should I use..

+20% from 1.3v? (1.550v)
or
+20% from 1.4v? (1.675v)

Will I be the first to witness SPDS (Sudden Prescott Death Syndrome)?

I'll make a compromise for now and use 1.60v for now, somebody else can find out wheter SPDS exists/will exist ;)

New voltage: 1.60v, new FSB: 270 save&exit..
Message tells me BIOS is corrupt.
Oh f-uck.

Holding insert and power-cycling brings sanity back to the chips of the beast but dents my courage.. I back down the voltage to 1.575v and memory divider to 3:2.

It lives!! :D
I'm currently running 14*270=3780MHz at 1.575v, I will let it to "burn in" for 4-6 hours by looping Firestarter demos and running Prime95 at the background.

Temps are as follows:
Old pump: 3.78/1.575 54-55C
New pump: 3.78/1.575 50-51C

+12v rail has dropped from 12.18v to 11.94v, I will adjust after burn-in period.

If I can keep this furnace cool I will try to hit the magic 4GHz barrier, I have never run over 280+FSB so if any of you have tips for very high stable FSB speed I'd be glad for help.


To be continued.. (unless it melts before)

trader88 02-06-2004 08:46 AM

Nice work so far. Keep it comming.

I think Prescott can go far without adding vcore. Why don't you try leave the vcore at lowest possible and test its OC ability first.

Phalanx28 02-06-2004 08:50 AM

Quote:

Originally posted by Kawaiitenshi
Well, actually they _shouldn't_ be (process shrink, lower core voltage)
This is an incorrect assumption. That's a goal, not necessarily a given for process shrinks. Remember, with any die shrink, you're increasing the overall density of the core (also a reason for the added layer, me thinks). And, that means you have a smaller area with which to dissipate heat.

Current leakage is the second factor that plays into the increased heat output. Before the launch, it was rumored that up to 30% of Prescott's heat output was due to leakage. Only Intel knows what the real percentage is here, but Prescott is certainly a step in the wrong direction as far as heat output goes.

Plus, when you factor in all of the elements that kONGO were added to Prescott, they might have been fighting an uphill battle. Though they probably had Prescott on the drawing board for a few years, it's hard not to say that adding 512KB L2 cache to Northwood and a shrink wouldn't have been a not-so-bad idea.

Of couse, there's probably a reason I don't work for Intel. ;)

Quote:

but rumor says Intel got bitten in the a-s-s by the strained silicon technology -> faster transistors but huge leak currents so at least this revision of Prescott won't be the new C300/P4-2.4 for air cooler masses :(
SS alone isn't going to help Intel, here. The quicker they move to SOI (FD or SSOI), the better off they're going to be in the heat battle. Leakage will become an even bigger problem down the line.

Quote:

However, my journey won't even begin with an air cooler and I have a few aces in my sleeves if things get HOT ;)
Water cooling seems to be the minimum requirement for a successful Prescott overclock at the moment. Still, there is a large portion of the enthusiast community (like yourself) that is well-equipped to make the 2.8E a nice overclocking prospect, IMO.

I hope you break 4GHz.

XJ. 02-06-2004 09:54 AM

Quote:

Originally posted by Phalanx28
SS alone isn't going to help Intel, here. The quicker they move to SOI (FD or SSOI), the better off they're going to be in the heat battle. Leakage will become an even bigger problem down the line.
Well, the quip I've heard, regarding the shink to 90nm and below "quantum physics is real" :D


All times are GMT -5. The time now is 03:33 PM.

Powered by vBulletin® Copyright ©2000 - 2017, Jelsoft Enterprises Ltd.