Bob F wrote:
Bob F wrote:
Bob F wrote:
Paul wrote:
Bob F wrote:
Paul wrote:
Bob F wrote:
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?
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-7...sb-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
The current processor is a Q8200. Do you think the E5472 would
'feel' much faster?


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
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.
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.

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.

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