Asus P5K Premium with Xeon E5472 and 771 to 775 adapter sticker??

[Bob F]

Not exactly an overclock yet, but maybe??

I am considering buying a used E5472 processor to use in this P5K Premium board
with a 771 to 775 adapter sticker. My memory is a pair of 2GB PC2-6400 samsung
M3 78T5663EH3-cf7. This processor has a 1600 MHz FSB. The motherboard seems to
be
speced for 1600 FSB.

Does anyone have any knowledge or opinions about this kind of project? Is it
likely to work? Will it overclock?

[Paul]

Looks like a fun project. Note that it involves damaging the LGA socket
by removing two pins, which may affect resale to others, unless you
sell mobo+CPU as a set. (I have such a hacked board here, where it's probably
worth leaving the current CPU in the socket, and never changing it out :-) )

http://www.delidded.com/lga-771-to-775-adapter/#xeon-fsb-tdp-warning

I don't think "overclocked" is the word. The chipset is FSB1333,
the FSB1600 BCLK is already an overclock for the chipset. Expecting
more than that, isn't going to leave you much room to work with.
Sure, it might overclock a tiny bit above FSB1600, but
how stable will it be ?

You'll be getting something close to Q9650 performance, hopefully
stable at FSB1600, that's a little hard on the VCore circuit. You
have a nice heatpipe on that board, so have some options for cooling
VCore if it is getting a bit warm. This is a problem I have with
a current Asus board, too-small cooler on VCore, means extra
noisy airflow needed. My corrent LGA775 board has a massively
overdesigned cooler, and I wish I could swap it with the new board
that actually needs a cooler.

You might be a tiny bit memory starved with that setup, with the
best RAM available for it (say memory bus is 67% efficient or so).
And those CPUs, which use two silicon dies and cache coherency protocols
that run on the shared FSB, it causes 87% efficiency when all four cores
are running flat out. Four cores gives 4 * 0.87 or ~3.5 cores worth of
horsepower. Really no worse than some other current day designs (6 core
processors starved to 5 core performance). Just a footnote when
benchmarking or something.

If you previously had a dual core, and did a lot of movie editing,
this would be a nice upgrade. If you already have a Q9650, this
would be a pointless project. Maybe you'd get a slightly
higher Stream benchmark or something. But the board will definitely
be worth keeping for a while, as at that performance level, you're
well past any reasonable minimum for the bloated OSes of today.

*******

Only question I have, is Microcode. Will you be getting an
unknown CPU error at startup ? That's the only risky part of
stuff like this. My first motherboard, there was a tool you
could load a Microcode, into the Microcode cache in the BIOS
chip (of the several BIOS makers, like Award, AMI, Phoenix, one
of them had this Microcode cache feature). That's how I stopped
the BIOS from claiming I was "running a Pentium II" or such like.
But if that delidded page claims it works, and you have nothing
to lose, it looks like a good project for a rainy day. Just don't
expect the FSB to go to infinity and beyond :-) FSB1600 is already
technically an overclock, so your headroom is harvested.

I'm sure one of the enthusiast web fora, has records kept for
how far a P35 Northbridge can be pushed. It would probably
take at least a day of thread-reading, to get a good answer
for that.

Paul

[Bob F]

The current processor is a Q8200. Do you think the E5472 would 'feel' much
faster?

[Paul]

You would be going from 2.33Ghz quad to 3Ghz quad.

I have a 2.6Ghz dual and a 3.0GHz dual system here, and in casual
usage, can't tell the difference. Since the processors
had different caches, and different speed RAM, I could
see a significant difference when running 7ZIP compression.
But when surfing, they looked relatively the same.

Expect better movie render times, but surfing, maybe not
an earth shattering experience.

You could even try your hand at overclocking the 8200.

I've done that on some systems here, as an A:B test. To
see whether overclocking was worth it. On a gaming system,
it was worth doing. On a very weak system with a single
core processor, the chipset was hopelessly bottlenecked,
and a faster processor was a waste of time. I turned it back
down to nominal.

I have a new machine on the kitchen table, in test mode.
(The parts have seven day no-questions return policy.)
Surfing, the box doesn't feel any faster. But when I run
7ZIP on real files, the new system is 6.7x faster. It was
compressing at 23MB/sec. I expect some other content trials
to weigh in at 4.5 to 5x or so. But the first semi-uncontrolled
test was impressive. But so was the power consumption.
12V * 13A at Vcore input, or 156W just for the processor
component. And the Vcore regulator was too hot to touch
(not a good sign if you wanted to overclock some day).
After running Prime95 overnight, the kitchen was warm.

It's one of those "be careful what you wish for" things.
Has its plusses and minuses. If you edit movies and
change movie formats, it could be worth your effort.
Or gaming with some heavy-weight titles and $250 video
cards. Just OCing the 8200 might be as much fun.

Paul

[Bob F]

I got the E5472 and the 771 to 775 converters, and inserted the modified E5472
into an Abit IP35 Pro I happen to have on the bench currently, after updating
the bios to the latest ver 18 from 14. I had found several cases of overclocking
that board to well above FSB 1600, and thought it was worth trying. It does run,
but will not boot to memtest86+ and any memory clock over 272 MHz. Every time I
boot to Bios and set the clock higher than 272( even 275), when the processor
reboots after the bios change, it stops a second time, and ends up back at 272
by the time it gets to memtest.

This confuses me, because the processor should run at 400 by spec.

Is this likely a microcode problem? I read that the Ver 18 Bios added the "E0"
processors, and elsewhere that the E5472 was an E0, so I was hoping this would
work.

Time to research modifying Bios with new microcode, I suspect.

[Bob F]

Further testing:

I built a modified bios with microcode including the zeon 54xx processors. Using
that, no change.

I cannot change the "external" clock or multiplier. Every change to these goes
through the double boot, and ends up at 272 and 7.5, whether I change them
higher or lower. Looking at it after changing the multiplier to 6.0, the bios
shows it at 7.0 afer reboot, with the estimated CPU clock of 1904 (And DRAM
speed of DDR2-652). but when I get to memtest86+, that shows FSB 272 and 2040
MHz at the processor (7.5 multiplier), and memtest86+ shows the memory SPD as
1024 MB DDR2-400 - Corsair CM2X1024-6400C4 for both slots, and the Settings: RAM
: 408MHz (DDR816)/CAS: 5-5-5-18 / Dual Channel.

I am again confused about the RAM naming conventions and ram speed relation to
FSB. Or maybe Memtest86+ is just confused.

Interestingly, when the board boots, it shows the CPUID as 676 rev 0F instead of
the 10676 Rev 10F values I supposedly put into the microcode.

This double boot problem seems to be a regular problem for this and other Abit
and other P35 motherboards.

[Bob F]

I came up with memtest86+ errors and a crash, and gave up on the E5472/IP35 Pro
combo. I put the Q6600 back in it, and the double boot problem is gone, and it
seems to be memtesting fine, currently at overclocked 2.7 GHz. It seems the IP35
Pro just can't handle the E5472. If anyone has any clues why, I'd love to hear
them.

I'll try the P5K Premium later.

[Paul]

Just a word of warning, the kitchen table build has had significant
problems with reported speed. I run tools and they don't indicate
speed properly. As an example, on one run, memtest86+ claimed "DDR3-3100",
which would have roasted the RAM, and might only be possible
with very expensive sticks. Booting to Windows and running
CPUZ, indicated the normal speed I was expecting. Even the UEFI
BIOS has had problems, and has a mind of its own.

(Use the no-install version - for Windows OSes)
http://www.cpuid.com/softwares/cpu-z.html

If you use Windows for testing, make a backup image of C:
using your "stable" processor, just in case. You might need
it if Windows crashes while testing your "exotic" processor.
Maybe the "strap" used by such a processor, is confusing the
BIOS or something. But I thought there was at least one
1600MHz processor in the table - if the CPU Support chart
has one, that's a good sign.

Otherwise, you may have to consider locating a Windows based
overclocker tool. Such tools, work with particular clock generators,
and it's the luck of the draw as to whether your particular
motherboard would work. I think I got lucky once, and managed
to find a tool that would overclock one of my older boards.
It bumps the FSB a megahertz at a time, and it would
take 30 seconds or so to get to the "target" frequency I
specified while sitting in Windows.

And 272MHz is a weird value. Not a canonical value. I wonder
where that is coming from ?

Paul

[Bob F]

Interestingly, that is the same as the Q6600 that I had in it previously, and
that is back in the board working great now. I can't imagine any way that value
could have stuck through the processor change/bios update/cmos reset/bios to
default sequence.

I haven't gotten as far as windows on the IP35 Pro yet, so a windows version of
CPUID is out on that PC. I've basically just been trying to get the new
processor to run memtest at a reasonable speed. I does run it at the 272.
although the last test I left it on had 10 or so errors, then crashed.
Interestingly, the errors were complete bit inversions.

One thing that really has me wondering is why I cannot change the E5472 speed
either up or down. This probably is associated with the double boot, which
always sets it to 272. No idea why that processor makes it always double boot.

[Bob F]

Is it possible that the bios is detecting a problem and restoring the settings
to the previous fully working settings of the previous processor, even after
resetting the bios? But then, why can't I change the speed slightly, even down,
from where it does work?

If it resets to the processor default, and doesn't like that, it resets to the
previous working settings? I'd be really impressed if it was that smart, but
then again, it could be dangerous with some processor swaps.

Maybe I'll try putting the E5472 in after running the Q6600 at 300 instead of
272, and see what happens.

[Bob F]

Bob F wrote:
.......

Interesting.

I tested Q6600 on this board for awhile with a 320MHz clock got 2880MHz
processor speed. I then shut it down, swapped in the E5472, reset CMOS, and
booted it to MemTest86+ V5.1. (My previous tests were on 4.1) The E5472 is now
running at a clock of 320MHz, showing a speed of 2400 MHz. So it is remembering
the clock settings of the previous working processor. I wonder if it is
remembering the voltage and other settings of the Q6600 also.

ESCing from Memtest86+, it does soft boot without the double boot , and going
into the BIOS, I see the 320 MHz clock, 7.5 multiplier in the active "User
Defined" settings. The CPU core voltage is shown as 1.2250V and CPU VTT is
1.10V, MCH 1.25V voltage is Auto, ICH 1.05 is 1.05, and ICHIO 1.5V is 1.5V. CPU
temp is 47C, System temp 30C, and PWM 38C.

It does seem to be running Memtest86+ just fine, with no errors. Memtest86+
shows the RAM running at 480 MHz (DDR2-960) - BCLK: 320m timings 5-5-5-18 @
128bit mode. The Bios shows DRAM speed as "Default (DDR2-768)"

Looking at the double boot in detail: Powering the board up with the E5472 after
powering down, and pushing the start button, it boots, displaying the numbers
83, several quickly I can't catch, then 90 and 99, then powers down and
re-boots. The on board display then displays 84,83,84,83,84,82, then hangs
there, with the green on board LED flickering off every half minute or so.
Powering down and back up and pushing the start button, It then starts up
correctly, but at the same settings each time. Number sequences are
abbreviated - some are too fast to catch.

Changing the speed from 320 to 300 in the BIOS and saving the change results in
the standard reboot, but letting it boot to MemTest86+ shows the speed has not
changed from 320.

Changing the speed

[Paul]

Most boards in that situation, not knowing what you've plugged in,
wouldn't attempt to start. They would have an appropriate beep code
and you'd be stuck.

To me, it seems a bit dangerous to be using the previous
settings for the new processor.

What I can't understand, is why your board doesn't apply
canonical values. Clock generator at 100Mhz, 133Mhz,
166MHz, 200MHz, or 200, 266, 333, 400. As selected by
the BSEL straps on the processor. Similarly, the VID
voltage can be specified by the processor, as it drives
out bit values on the VID pins, to set the voltage it
wants. And the board really should not be attempting
to program the clock generator, if it doesn't know the
correct value. It is supposed to rely on the processor
strap values to determine that.

As for your Port 80 codes, some motherboard manuals have
a table of values for those. They're "progress" codes rather
than error codes, so even if you decoded them, you might
not be any further along in understanding what is
going on. The last code on the display should be
"Booting OS", and after that point the BIOS won't be
writing any new values to the display.

The "progress codes" are of more value to the developers,
than to us. I haven't run into a scenario yet, where
knowledge of "progress" helped debug an error
condition.

If you wanted to capture more of the codes, you could point
a movie camera at the display. It's still going to miss
stuff, but at least you don't have to write them down
"on the fly".

Paul

[Bob F]

I finally got around to trying this processor on the P5K Premium after my
problems on the Abit IP35 Pro.

I modded the ASUS AMI bios using the following:
http://www.delidded.com/how-to-update-cpu-microcode-in-ami-bios/

Booted it up to bios, reset to default, rebooted to windows. Windows runs for
awhile, then crashes. That's way further than I got with the Abit board.

Speedfan seems to display temps about 15C below the real temp displayed by
coretemp on the E5472. It was right on with my Q8200.

Temps, still using the stock Intel heatsink, seem to be higher than they should
be, so maybe that is why it crashes. The E5472 seems to have a max temp spec of
75C. Speedfan was showing temps in the mid 60's, so with the 15C offset, maybe
too high. The E5472 is supposedly 80W max compared to the Q8200 at 95, so I'm a
bit surprised. They seem to both idle around 50C, compensating for the speedfan
errors on the E5472, with speedfan controlling the CPU fan speed according to
temp.

I thought these newer processors were suppost to slow down if they get too hot,
rather than just crashing. Could it be a memory error? The DDR2-800 memory
should be able to handle this speed, right? ASUS claims this board is good to
FSB 1600, which I understand matches the DDR2-800 memory

I've got the Q8200 back in now, as this is my HTPC machine, so I don't want it
crashing, but will do more playing with the E5472 when I have the opportunity.

I'd appreciate opinions and suggestions.

[Paul]

The processor should throttle if it gets too hot. And if it gets
20C above the throttle temp, the whole computer shuts off (known
as THERMTRIP).

*******

So, we need an excuse for why it would get hot...

1) Heatsink isn't sitting flat on the CPU.

Or the TIM underneath the lid is defective - highly unlikely on
the LGA775 ones, as they used low temperature solder and the lid
s soldered to the silicon die. It's one of the best TIMs there
is (in terms of not being defective). Modern processors, like the
very latest Haswell, have switched back to "cookie dough" style
internal thermal interface material. Some flavors of cookie dough,
are not uniform from one section to another. Or even have voids in
them.

Intel switched from the low temperature solder (so they claim),
because the low temperature solder is a "conflict mineral".
I think they stopped using it, to boost the profit margin :-)
Solder is more expensive than the cookie dough they use.

2) Not enough thermal paste or too much thermal paste between CPU and HSF.

3) VCore higher than intended (shouldn't happen on modern CPUs).

My LGA775 processor, I think there is a register inside the
CPU which is "range checked". The register is loaded with a
number, that drives the VID pins. The only way to overvolt, is by
using the offset pin on the VCore switching regulator. On my
Asrock board, that's the pin I had to do a mod on, to get some
VCore boost for overclocking. The LGA775 I'm typing this on, has
proper BIOS-inspired offset pin overvolting, if I wanted to overclock.
No mod is necessary on this motherboard.

There was a time, when the BIOS could mis-interpret the CPU type,
and by using VID override, pump the wrong voltage into it. Your
board is modern enough to not be doing that. Only if you entered
the BIOS and intentionally boosted above the range-checked value,
could you warm up that CPU. (Some BIOS, display the VCore value
field in RED, if it's set really high.)

4) The CPU can be overclocked by increasing BCLK. You would use CPU-Z
or similar, to verify stuff like that. If Intel EIST (SpeedStep) is
disabled, the multiplier stays at the high value all the time, which
makes things a bit warmer.

There is certainly a history of this sort of funny business,
but it's the era of the hardware that suggests nothing
nefarious is going on. So about all you can do, is take a lap
through the BIOS settings, later, review with CPU-Z, to see
if anything is dialed too high.

On my new board, I got off to a bad start, when the
fan was spinning at the wrong speed. Turned out the
BIOS defaulted to "really quiet fan settings". I had to
disable fan control entirely, to get decent cooling.
I was shocked at the time, that I was getting "low fan speed"
warnings in the BIOS. When I thought they'd solved those
problems years ago. I used to have older systems here,
that would start whining (and stop booting), fi the RPMs
of the CPU fan were below 1800 RPM. My new board might
have been running the fan at 1200 RPM and I'm getting
a warning. I guess I'm happy I got the warning, because
it did get me to look at the settings. But I'n not happy
with the default behavior of the fan control. "Aggressively slow."

Now, my CPU runs cool as can be, even when (measured value)
it is burning 156 watts. The worrying part, is the VCore
regulator is too hot to touch. I had to fit an extra fan
next to Vcore to bring it into range. VCore will still be
hot, right at each MOSFET, but the puny fins on the Vcore heatsink
are no longer scalding hot. If MOSFETs get too hot, the
channel resistance goes up (potential for thermal runaway).

Paul

[Bob F]

I just re-installed the E5472, and modified Speedfan to compensate for the 15C
temp offset for reading the E5472. With the compensation, SpeedFan reads the
same as CoreTemp.

I ran MemTest86+ with no errors, with my ram at both 6-6-6-18 and then 5-5-5-15.
Funny thing, memtest86+ on the Q8200 showed the RAM speed as 4394 MB/s at
6-6-6-18 setting. The E5472 shows 3969 MB/s at 6-6-6-18 and 4197 at 5-5-5-15. So
the E5472 is displayed as slower on Memtest86+. The time for a whole pass
testing the memory was faster with the E5472 however. It took 33:42 on the
Q8200, and 30:32 with the E5472, both at 6-6-6-18. At 5-5-5-15, that was 29:33
on the E5472. This suggests something funny about the MemTest86+ speed reading.
Just looked, MEmTest86+ has a new version. I'm at 4.1, so I need to get 5.1 and
try that.

I ran the Prime95 torture test for 35 minutes without a problem (at 5-5-5-15).
Core temps maxed at 67C. Interestingly, coretemp shows Tjmax as 100C, although
I've found numbers from 71 to 85 online. Intel lists case temp max of 67C, which
I'm about 10C under.

It looks like this E5472 on my P5K Premium is a GO! I guess it's time to try
overclocking it a bit.

[Bob F]

I am now running at 3.3 GHz (FSB 1761 - DDR2-880)) without problem. At first
when I got to 3.3, the temp under Prime95 was pushing the limits. The default
CPU voltage bounced around 1.25V, and since the spec is .85-1.35, I tried
reducing that voltage, and finally settled at 1.18V setting, which displays on
Speedfan as 1.16V. Running Prime95 at that voltage indicated no problems, and
the core temps now stayed below 70C. CPU case temp got up to about 62C. RAM
voltage remains at the default 1.8V. I'll keep my eyes open for a better
heatsink than the standard Intel assembly, but I'm quite happy with these
results. The PC definately is faster.

[Bob F]

I did more research about the Tjmax value for the E5472, and it seems that 85C
is the consensis. RealTemp uses that value, and CoreTemp uses 100, which I can
to figure out how to change. So, Speedfan was initially displaying the correct
core temps, and my temps were OK (on the second test described in another post.
My idle temp currently is around 32C, and my temps in the hot Prime95 torture
test are in the low 50's, at 3.3GHz overclock. This is a nice processor if you
have a motherboard that will handle 1600 FSB.

[~misfit~]

Once upon a time on usenet Paul wrote:
[snip]

> So, we need an excuse for why it would get hot...
>
> 1) Heatsink isn't sitting flat on the CPU.
>
> Or the TIM underneath the lid is defective - highly unlikely on
> the LGA775 ones, as they used low temperature solder and the lid
> s soldered to the silicon die. It's one of the best TIMs there
> is (in terms of not being defective). Modern processors, like the
> very latest Haswell, have switched back to "cookie dough" style
> internal thermal interface material. Some flavors of cookie dough,
> are not uniform from one section to another. Or even have voids in
> them.
>
> Intel switched from the low temperature solder (so they claim),
> because the low temperature solder is a "conflict mineral".
> I think they stopped using it, to boost the profit margin :-)
> Solder is more expensive than the cookie dough they use.

This is really interesting thanks Paul. (I stopped reading tech / hardware
news / sites etc a while back as my financial situation went even more
downhill and reading about the latest stuff made me miss not having even
close to 'cutting edge' tech any more. Hi, my name's Shaun and I'm a
hardware junkie....)

I've got a socket 775 E7300 that's always had one core running around 15º C
hotter than the other. I spent quite some time messing with TIM, bought an
overkill cooler and even lapped the IHS and it made no difference. In the
end I put it down to bad TIM between IHS and die. I tried to remove the IHS
(like I used to with Tualatins when I was O/Cing them) but it didn't move
with as much force as I cared to exert trying. Now I know why.

Cheers,
--
Shaun.

"Humans will have advanced a long, long, way when religious belief has a
cozy little classification in the DSM."
David Melville (in r.a.s.f1)

[Paul]

One of the enthusiast sites showed how the users
there remove them. This is an example.

http://www.overclock.net/t/305443/ihs-removals-how-to-do-it-should-i-do-it-and-the-facts

"The solder Intel uses roughly melts at 80-90c so it could take a while."

If that were true, boiling water would remove it :-)

http://www.legitreviews.com/intel-ihs-lapping-and-removal-for-lga775-cpus_402/2

"the extra solder on the core was scraped off using a plastic credit card"
"The system would power on, but sadly it wouldn’t post."

Just a small failure rate problem. Good if you start
with a bucket full of processors.

I like the lapping job they did on the lid in the second article.
They claim it only takes 20 minutes to do that, but if I was
doing it, I'd need to start with the belt sander, to be
able to finish off with 2500 grit in 20 minutes :-)

Paul

[~misfit~]

Sorry for the late reply. For some reason I've not checked this newsgroup
for ages.

> One of the enthusiast sites showed how the users
> there remove them. This is an example.
>
> http://www.overclock.net/t/305443/ihs-removals-how-to-do-it-should-i-do-it-and-the-facts
>
> "The solder Intel uses roughly melts at 80-90c so it could take a
> while."
> If that were true, boiling water would remove it :-)

Hmmm. From what little research I've done 'In52' solder (Indium 52% Tin 48%)
would probably be best for attaching an IHS to a silicon die if you were
going to use solder. It has a melting point of 118º C and has a good
'wettability' of glass, quartz and ceramics. Also I think 118º C is above
'THERMTRIP'? Even though the IHS is glued on and isn't about to fall off you
wouldn't want the solder 'sagging' due to gravity in a tower case, migrating
away from one side of the die.

The solder with the lowest melting point is 'Cerrolow 117' which melts at
47º C. There are a lot of solders in between those two;
http://en.wikipedia.org/wiki/Solder#Solder_alloys

> http://www.legitreviews.com/intel-ihs-lapping-and-removal-for-lga775-cpus_402/2
>
> "the extra solder on the core was scraped off using a plastic
> credit card" "The system would power on, but sadly it wouldn’t
> post."

Heh!

> Just a small failure rate problem. Good if you start
> with a bucket full of processors.

But if you have the money for a bucketful of processors why not just buy a
faster one? ;)

> I like the lapping job they did on the lid in the second article.
> They claim it only takes 20 minutes to do that, but if I was
> doing it, I'd need to start with the belt sander, to be
> able to finish off with 2500 grit in 20 minutes :-)

I first lapped the IHS on my E4300 a few years back, my last real overclock
to speak of. I managed to get it Prime95 stable with a 50% overclock for a
year. It took quite some time. I didn't clock it but I'd say a couple of
hours (not counting the breaks to rest the tired arm). It was massively
dished with high corners and lapping it resulted in about 5º C temperature
reduction under full load overclocked, to around 58º. Before that it was
over 60º and I wasn't comfortable with it.

After a year I bought a faster 45nm CPU (the E7300 mentioned above). Since
about then I've had no need to overclock as CPU performance isn't the
bottleneck it once was and powerful CPUs are affordable too. However I still
ended up laping the E7300 due to the 15º differential between the two cores.

These days it's I/O that holds things up. Newer CPUs can do all that I'll
ever need (and much much more) so my days of overclocking are behind me.