HEADS_UP INQUIRY OF P4-PRESCOTT (socket478) USERS

[BonzaiDuck]

My system -- probably summarized in other posts on these forums (and I should make an effort to get it into my "profile" so it appears with every post I make) -- is:

P4 3.0C Northwood @ (variously ) 3.6, 3.67 and 3.7 Ghz
ASUS P4P800 (standard) with BIOS 1019.500 (November, '04)
OCZ EL "Gold" DDR500 / PC4000 - 1GB (2x512)
dual Hitachi 7K250 RAID-0 320GB-array
Creative SB Audigy 2ZS
Hauppauge PVR 250 tuner-capture card
ASUS V9980 -- (nVidia) FX5950 Ultra 256 AGP
Sony DRU700A and TDK CDRW
ThermalRight XP-120 w SUNON KDE1212PMB1-6A (108 CFM)

The overclock settings use external frequencies of 240 Mhz, 245 Mhz and 247 Mhz respectively, and the voltage settings of VCore, VDIMM and VDDQ are {"Auto", 2.75V, 1.5V}, {1.525V, 2.85V, 1.5V} and {1.5327V, 2.85V, 1.6V} for the respective speeds (3.6, etc. etc.) Memory timings are 2.5, 4, 4, 6 @ 3.6 Ghz, and otherwise the stock settings of 2.5, 4, 4, 7 for the higher speeds.

I was reviewing the ASUS web-site in search of any BIOS revisions. Apparently my current revision, in the "notes" posted by ASUS under the link "More . . . " -- tersely states that the BIOS 1019 "adds the CPU lock-free feature".

I had read the review at THG about the ASUS P5AD2 (and -E) motherboard, in which this "lockfree" feature was explained, but I didn't pay enough attention to it, and frankly, I didn't think it was relevant to my P4P800 motherboard.

But it IS!!! . . . and I verified by further scouring the ASUS web-site, to find a table of newest and older motherboard features, and especially, a succinct explanation of "CPU lock-free".

With the lock-free feature and a socket-478 Prescott "E" processor, you can depress the CPU frequency multiple as low as 14x. I had earlier understood that you might want to RAISE the multiplier for a faster CPU speed with a limited FSB setting, however, as ASUS explains, depressing the multiple allows you to run the CPU at its "stock" speed and a higher-than-800-Mhz FSB setting for use with DDR500, DDR533 etc. memory modules.

It dawned on me that I had been reticent about getting an "E" processor because the Prescott has a "hot" reputation, and the Northwoods ran cooler. Also, with the faster memory and FSB, I had reasoned that I didn't need the extra L2 cache provided by the Prescott to buffer slower RAM in the the storage pyramid.

Now it occurs to me that if you could depress the multiple of a Prescott 3.(x)E while raising the FSB speed, you could run that Prescott at its stock speed while availing yourself of the faster memory, and with double the L2 cache the entire system would benchmark even better. AND . . .. AANNNDDD . . . it seems logical that running the Prescott at its stock speed with lower multiple and higher FSB ought to generate about the same CPU temperature at idle or load that you would observe if the Prescott was run at its stock multiple and 800 Mhz FSB.

Has anyone explored this? Does anyone have any firsthand experience with it?

All comments welcome and encouraged. Even if you are running a 2.8E or 3.0E, your observations might make it worthwhile for me to consider a 3.4E at the going market price. This might actually be more desirable or equally desirable to running my 3.0C at any of the three speeds mentioned above.

. . . . And you can comment on that remark, also

 

[bacillus]

afaik the lock free option only works with 3.2GHz and above Prescotts.

 

[BonzaiDuck]

I don't know if the "14x" lower limit is a hard and fast one.

But it would seem -- given the thermal problems -- that this would be a great way to open some bottlenecks by over-clocking to a "stock" result.

I just want to confirm it. I've spent so much time improving the thermal performance of my "MOJO", that I'm actually rather confident there would be an advantage to having a Prescott that could have its CPU frequency multiplier changed.

Personally, I'm of the opinion that an exclusively fast processor doth not alone a fast screamer make. And I think I'd swap out my Northwood for a sufficiently fast Prescott if I could confirm my thermal suspicions and theories.

 

[LTC8K6]

14X is the only other multiplier you can choose. It only works on the latest LGA775 chips with the new throttle down feature. The 3.6 and 3.8, I believe. Those chips drop their multiplier to 14X and lower their voltage when they get too hot. The lock free feature is just taking advantage of the 14X option.

 

[LTC8K6]

It is part of Thermal Monitor 2, aka TM2. I think it is only active on E0 stepping Prescotts and newer Xeons.

 

[Paratus]

I've read post somewhere that it does indeed work on the 3.2E and up. Whether it requires a certain stepping I don't know.

Bonzai I noticed the same thing for my P4P800 E, latest BIOS has the lock free feature. Since I have a 3.2E and Corsair TCCD memory I probably could knock the multi back to 14X and raise the FSB to get me back to 3.2 while keep memory 1:1.

Anyone have some thoughts on whether I ethically violate my warranty by running it "stock" at a higher FSB?

 

[ntrights]

Yes lock free works with s478 also. Here is a breakdown of which ones work.

3.2 SL7B8 C0 stepping. works
3.2 D0 stepping does not work
3.2 E0 To few reports to draw any conclusion..most likely Dell got a bunch of s478 E0's as they are very had to find in a retailshop....

3.4 D0 some works some don't it has to do with loadline A and loadline B. I can't say which s478 3.4Ghz D0's work or not. My guess here is that those with high TG will work regardless of stepping...

As far as LGA775 all the 3.4 works regardless if they are D0's or E0's

Hope this helped.

 

[LTC8K6]

Man there is a lot of vague info out about this on those S478s.

I wonder if Intel is even sure when they turned this feature on.

 

[BonzaiDuck]

I appreciate your comments, and keep them coming.

Some will argue that there is chaos in circuit-board-design-land, and they will say my assumptions and conclusions are "inconclusive". But why would ASUS include the "lock-free" feature in these socket-478 motherboards unless the Prescotts that fit the sockets had it?

Paratus said:

Anyone have some thoughts on whether I ethically violate my warranty by running it "stock" at a higher FSB?

Interesting question, especially if we manage to prove that thermal performance is roughly the same. And think about it -- if the features that allow depressing the multiplier are part of the throttling capabilities that are designed to address thermal issues, that is, if the processor would ordinarily knock down its multiplier to cope with a "hot" situation, there would be no ethical reason for Intel to void your warranty.

That's simply "logic". But when it comes to corporate decisions about legally enforcible warranty agreements, the correspondence between what is logic and what is corporate whim may also be "inconclusive."

You would think, however, that the only warranty you would avoid is the warranty on your motherboard, which is made in turn by a company that implements over-clocking features in the BIOS and in spite of -- well -- whatever their warranty says.

 

[LTC8K6]

You do violate the warranty by running the chip faster than it's rated speed, and/or at a higher voltage.

Not ethically, actually.

Now if you are only running the FSB higher and keeping the chip itself at the same rated speed, then I don't think you are voiding the warranty.

 

[Paratus]

Originally posted by: BonzaiDuck
I appreciate your comments, and keep them coming.

Some will argue that there is chaos in circuit-board-design-land, and they will say my assumptions and conclusions are "inconclusive". But why would ASUS include the "lock-free" feature in these socket-478 motherboards unless the Prescotts that fit the sockets had it?

Paratus said:

Anyone have some thoughts on whether I ethically violate my warranty by running it "stock" at a higher FSB?

Interesting question, especially if we manage to prove that thermal performance is roughly the same. And think about it -- if the features that allow depressing the multiplier are part of the throttling capabilities that are designed to address thermal issues, that is, if the processor would ordinarily knock down its multiplier to cope with a "hot" situation, there would be no ethical reason for Intel to void your warranty.

That's simply "logic". But when it comes to corporate decisions about legally enforcible warranty agreements, the correspondence between what is logic and what is corporate whim may also be "inconclusive."

You would think, however, that the only warranty you would avoid is the warranty on your motherboard, which is made in turn by a company that implements over-clocking features in the BIOS and in spite of -- well -- whatever their warranty says.

Thats about what I thought as well.

Originally posted by: ntrights
Yes lock free works with s478 also. Here is a breakdown of which ones work.

3.2 SL7B8 C0 stepping. works
3.2 D0 stepping does not work
3.2 E0 To few reports to draw any conclusion..most likely Dell got a bunch of s478 E0's as they are very had to find in a retailshop....

3.4 D0 some works some don't it has to do with loadline A and loadline B. I can't say which s478 3.4Ghz D0's work or not. My guess here is that those with high TG will work regardless of stepping...

As far as LGA775 all the 3.4 works regardless if they are D0's or E0's

Hope this helped.

Well CPU Z says I've got a 3.2E Revision C0 Stepping 3 Model 3. How do I know whether this is the SL7B8 version?

 

[Paratus]

Originally posted by: LTC8K6
You do violate the warranty by running the chip faster than it's rated speed, and/or at a higher voltage.

Not ethically, actually.

Now if you are only running the FSB higher and keeping the chip itself at the same rated speed, then I don't think you are voiding the warranty.

Interesting the second half of your quote was missing? Maybe my IE is horking up. I really need to put Firefox on this machine.


Anyway I agree with you.

 

[ntrights]

Originally posted by: LTC8K6
Man there is a lot of vague info out about this on those S478s.

I wonder if Intel is even sure when they turned this feature on.

lol yeah i agree...but i think that the lock free was implemented by intel and not by the mobo makers...they knew that it run hot and s478 was not really made for prescott..so all the mobo makers would have to redo the powerdesign specifikations but it would be to costly so instead they moved directly to implement thenew standard LGA775 boards. To be on the safe side intel made the early prescotts (C0) stepping lock free so that people with weak powerplant mobos still could run them without melting the mosfets

Paratusyou can find the product code on the box or cpu. It's the last five digits of the Product Code.
There are two 3.2Ghz C0´s steppings. One is SL79M the other SL7B8.

Im not 100% sure about SL79M though...But my guess is that its lock free also.(Thermal Guideline 103.0W)

BonzaiDuck One thing is for sure and thats when it comes to lockfree and overclocking it stresses the mobo alot ..the additional stress is casued by the high FSB so you better make sure that you sink the mobo...(all the mosfets) or you might toast it! With prescott its a new ballgame...its not the cpu that you have to be worried about of when giving more voltage its your mobo!!!!

Anyway i higly recommend the 3.2 sl7b8 only if you:
1. sink the mobo all the mosfets.
2. get good cooling. case ventalation air-duc to the cpu etc...
3. invest in a powerfull psu..OCZ or PC Power @ Cooling...or go with dual psu if you dont have a OCZ or PC power cooling. Or modd the whole psu like i did (highly not recommended though)

If you fulfill all 3 points you will have great fun with it. its a blast running multiple sessions High FSB is the thing...i did run 6 tmpgencoding instances simoultaniously converting dvd 1gig files and browsing and chatting on msn..whittout any problem whatsoever with my 3.2 sl7b8 @ 4 285FSB my 3.0@4.0 cant do that ...

its all about fsb the higher the better. the 3.2 is argubly the best P4 bang for the buck imho...

 

[BonzaiDuck]

ntrights --

I guessed "from the gitgo" that I had all the bases covered except one: the mobo and chipset. But THG and other sources were encouraging about running the ASUS P4P800 at FSB 1,000. That is, they promoted the idea -- originally -- of putting a 2.4C Northwood on the board with DDR500 rams (the article was about the OCZ EL "gold") -- and pumping up the external frequency to 250 and the FSB to 1,000.

THG concluded noting the possibility that the user just couldn't quite push the combination of components to FSB 1,000, and suggested "Well, you should probably be happy pushing the 2.4 to 2.8 or 2.9 and getting an FSB of 960 or 980." After I had things running cool as the driven snow with the 2.4 at 3.0 and FSB 1,000, I couldn't contain my curiosity and read through many forum and newegg "buyer" reviews before deciding to spring for the 3.0C.

I HAVE a tape-on thermister measuring temperatures off my chipset heatsink. With motherboard ducting, the temperature stays below 89F, although, to be honest, I know there is a lot of air-turbulence around that thermister and it makes you wonder about the "TRUE" chipset temp. But the system has proven rock-solid so far.

I said earlier that I had three over-clock settings I had found that were "apparently stable." They were 3.6, 3.67 and 3.7. I had hoped to get it up to 3.808 and pass PRIME95, but it didn't work out with the PRIME95, although it benchmarked with PCMark04 around 5,750. I watched the temperatures, noting that 3.6 requires no increase in ANY voltages -- VCORE, VDIMM or AGP-VDDQ. 3.67 requires bumping the memory up to 2.85V, but that is within the OCZ warranty of 3.0. It also requires bumping the VCore to 1.525V -- which always seems to show a range ending in 1.6V in monitoring software under XP. Temperatures rise 3 or 4F at 3.67. 3.7 requires bumping the AGP to 1.6V, and temps rise again another 3F.

I hadn't explored fully the torture-test options of PRIME95, and while the higher over-clocks would allow PRIME95 to run error free in its prime-number search for so many hours you were sure there was no problem, the small FFT torture test fails after an hour or so with the higher settings. Getting them to work certifiably under PRIME95 would require bumping up the VCORE again, so I have set everything back to 3.6 -- which passes "small FFT" and "Blend" tests.

It may be that I'll "blow" the motherboard, but at 3.6 Ghz, all the voltage settings are ASUS defaults, and the temperatures are at most a couple degrees higher than at the stock settings for 3.0C.

I'm eager to try a Prescott 3.2E or even the 3.4E. I can bide my time and save some pennies first. The worst thing that will happen is that I won't be happy with it, and I'll have to sell it to someone for less than I purchase it for. The second worst thing is that I actually do blow the mobo chipset, but I can always replace that for a little over $100. And the third worst thing is turning over the 3.0C Northwood for too little in resale.

 

[ntrights]

Not sure what is holding your northy back but 3.6 is pretty good though! With vapo i've seen 3.0C@4.0...If its worth the upgrade is hard to say...I noticed major performance
increase going from northwood to prescott mainly in heavy-duty multitasking sessions (increased multitasking abilities) + the potentiall for higher overclocks.

It's defenitly worth a try..your board can handle very high overclocks also..couple of guys on another forum have there p4p800 (rev2.0) with a 3.2E@ 4.45 range with vapo.

common overclock range with3.2E:
air 3.8-4.0
water 4.0-4.2
vapo 4.4-4.5+

 

[BonzaiDuck]

It seems there are more people who are happy with their 3.0E and 3.2E processors using the P4P800 motherboard than you would imagine generally given the thermal myths that prevail about the Prescott.

 

[VirtualLarry]

Originally posted by: LTC8K6
14X is the only other multiplier you can choose. It only works on the latest LGA775 chips with the new throttle down feature. The 3.6 and 3.8, I believe. Those chips drop their multiplier to 14X and lower their voltage when they get too hot. The lock free feature is just taking advantage of the 14X option.

That's vaguely interesting. How different is that to the SpeedStep support that the S478 P4 mobile chips have? I've read that at reset, if the chipset doesn't support SpeedStep properly (like most desktops), the multi defaults to 12X. The reason that it's vaguely interesting, is that I have a 2.0Ghz P4 Northwood 'C' mobile CPU here, that I'm putting together to work in an 845PE desktop board that I recently picked up. Since the chip has a default FSB of 533, I figure that if it drops down to 12X, but I OC the FSB to 800, I'll end up with a 2.4Ghz Northwood, which should be a viable overclock for this chip. Any suggestions, possibly for the replacement for the IHS, since the mobile chips lack the IHS?

 

[TStep]

Originally posted by: VirtualLarry

Originally posted by: LTC8K6
14X is the only other multiplier you can choose. It only works on the latest LGA775 chips with the new throttle down feature. The 3.6 and 3.8, I believe. Those chips drop their multiplier to 14X and lower their voltage when they get too hot. The lock free feature is just taking advantage of the 14X option.

That's vaguely interesting. How different is that to the SpeedStep support that the S478 P4 mobile chips have? I've read that at reset, if the chipset doesn't support SpeedStep properly (like most desktops), the multi defaults to 12X. The reason that it's vaguely interesting, is that I have a 2.0Ghz P4 Northwood 'C' mobile CPU here, that I'm putting together to work in an 845PE desktop board that I recently picked up. Since the chip has a default FSB of 533, I figure that if it drops down to 12X, but I OC the FSB to 800, I'll end up with a 2.4Ghz Northwood, which should be a viable overclock for this chip. Any suggestions, possibly for the replacement for the IHS, since the mobile chips lack the IHS?

I've had, in the past, several Northwood P4m's though my Abit AI7 and Asus P4S800 mobos. Both boards booted right up with the 12x multi. Some boards will not boot, so it's trial and error. I've only used an SP94 with the AI7 and P4 mobiles, so no IHS was needed. I've also found that the core and heatsink make sufficient contact on the Asus to not need the addition of the IHS. But I always would check this first with a little dab of paste and lightly install the hsf, then pull it and inspect the contact patch.

 

[VirtualLarry]

Originally posted by: TStep
I've had, in the past, several Northwood P4m's though my Abit AI7 and Asus P4S800 mobos. Both boards booted right up with the 12x multi.

This board is QDI PlatinX 2PE/800. (It was cheap, and I don't even know if the P4 chip works, so I didn't want to go all out on the mobo. But the pages I read about using a P4 mobile chip in a desktop were using 854PE/ICH4 systems too, so I figured I should try to get an 854PE-based one to be safe, in case Intel locked-out mobile chips from running on newer chipsets, and this one explicitly supports an 800Mhz FSB.)

Originally posted by: TStep
But I always would check this first with a little dab of paste and lightly install the hsf, then pull it and inspect the contact patch.

Yes, already tried, no contact at all. Not sure if the heatsink is being held up by the plastic tops of the nubs holding the heatsink bracket to the mobo, or if it's the bent-up part of the level in the CPU socket itself. I'm not sure what is the best way to go here: 1) Find a replacement (factory-spec) IHS, off of a dead CPU, 2) use a different, more-expensive heatsink, one that has a center that protrudes a bit to make contact with the CPU core, or 3) find some sort of metal shim to put in-between, and just be careful applying two layers of paste, and mashing everything back down when attaching the heatsink. I've heard someone suggest sanded-down pennies. I just want to make sure that I don't accidentally crack the core, whatever I do.

 

[TStep]

VirtualLarry: chances are it will probably work. All the original P4 800fsb black art was done on 845PE boards using P4 mobiles around the various forums. I believe SIS uses the desktop chipsets in their mobile solutions, so I believe they are a sure thing. Since the IHS is nothing more than a piece of metal, I would just go to a local machine shop and ask them for a scrap piece thin copper or aluminum and do the sandwich thing. A craft store may also have it. An Intel data sheet should be able to provide the IHS dimensions.

 

[VirtualLarry]

I found this link showing the stock Intel P4 HSF assembly, and noticed that it is different from my OEM P4 HSF assembly. Specifically, the stock Intel one seems to have two recessed troughs along each side, which would allow the sink to make closer/lower contact with the CPU in the center region. The HSF that I have, the bottom is all one flat level, and appears to be touching/bumping the plastic clips holding the HSF retention-bracket to the mobo. (No wonder it was so difficult to clip on.) I wonder if I obtain a swap for a stock P4 heatsink if I can simply just use that one. I'm going to check around. If not, I'll use a "shim sandwich", although I'm wondering if I shouldn't use a small piece of felt around the core, to prevent un-even pressure from the shim and HSF from crushing the exposed core too. Guess I'll try that, thanks for your replies.

 

[TStep]

I've used the stock intel with the round copper plug core that seems to have come into around the 2.8 models. The copper core slightly protrudes from the rest of the base and makes good contact. I find no need for felt pads.