Prescott tests

  • Thread starter Thread starter Yousuf Khan
  • Start date Start date
Derek Baker said:
"Yousuf Khan" <[email protected]> wrote in message

I posted these and more, more than 11 hours before you, but my ISP is
having trouble with its newgroup feed, so noone else can see them.
Bastards! :)

This posted via Google Groups.
Click to expand...

I'm sorry, but it's not like as if my ISP is the model of high quality and
efficiency either. :-)

Anandtech seems to think that the Prescott should only be purchased if
you're going to try to overclock to 4Ghz, otherwise go for Northwood. This
despite the fact that the Prescott uses up far more power than the Northwood
at the same frequency.

Ace's hardware seems to think that at 125 million transistors, Prescott is
hiding some circuitry inside it, which Intel hasn't told us about yet. They
come to this conclusion by adding up the extra circuitry required by the
512K larger L2 cache, and the 8K larger L1 instruction cache, and estimate
that altogether this should take Prescott only up 80 million transistors
from the Northwood's 52 million. So there's still 45 million transistors
unaccounted for inside Prescott.

Tom's just seems annoyed by Prescott for some reason. :-)

Yousuf Khan
 
Ace's hardware seems to think that at 125 million transistors,
Prescott is hiding some circuitry inside it, which Intel hasn't
told us about yet. They come to this conclusion by adding up the
extra circuitry required by the 512K larger L2 cache, and the 8K
larger L1 instruction cache, and estimate that altogether this
should take Prescott only up 80 million transistors from the
Northwood's 52 million. So there's still 45 million transistors
unaccounted for inside Prescott.
Click to expand...

64 bit ? naaaa



Pozdrawiam.
 
Yousuf Khan said:
http://www.aceshardware.com/read.jsp?id=60000315

http://www.tomshardware.com/cpu/20040201/index.html

http://anandtech.com/cpu/showdoc.html?i=1956
Click to expand...


I thought one of the more interseting comments from Anandtech is that
going from 2.8gz up to 3.2ghz Prescott narrows the gap (or increases
the lead) with Northwood at each successive higher frequency. They
extrapolate that Prescott at 3.6ghz and above might be a better
indicator of why the internal changes were made from the Northwood
core.
 
(e-mail address removed) (G) wrote in
I thought one of the more interseting comments from Anandtech is
that going from 2.8gz up to 3.2ghz Prescott narrows the gap (or
increases the lead) with Northwood at each successive higher
frequency. They extrapolate that Prescott at 3.6ghz and above
might be a better indicator of why the internal changes were
made from the Northwood core.
Click to expand...

Am I the only one that sees the FSB impact here ? OCing from 2.8 to
3.2 = 230Hz FSB - thers your speed


Pozdrawiam.
 
(e-mail address removed) (G) wrote in


Am I the only one that sees the FSB impact here ? OCing from 2.8 to
3.2 = 230Hz FSB - thers your speed
Click to expand...

Except he wasn't overclocking the FSB, but using a chip with a different
multiplier. It seems likely it's the cache that makes the most of the
difference. The bigger the difference between main memory speed and cache
speed, the more important it is to have a large cache.
 
Except he wasn't overclocking the FSB, but using a chip with a
different multiplier. It seems likely it's the cache that makes
the most of the difference. The bigger the difference between
main memory speed and cache speed, the more important it is to
have a large cache.
Click to expand...

prove me wrong, dont yust say it :)
http://images.anandtech.com/reviews/cpu/intel/pentium4/prescott/cpuz3
700.gif

fixed 14x multiplier. Its normal processor, not an engeenering sample
with open multiplier.


Pozdrawiam.
 
Yousuf Khan said:
I'm sorry, but it's not like as if my ISP is the model of high quality and
efficiency either. :-)

Anandtech seems to think that the Prescott should only be purchased if
you're going to try to overclock to 4Ghz, otherwise go for Northwood. This
despite the fact that the Prescott uses up far more power than the Northwood
at the same frequency.

Ace's hardware seems to think that at 125 million transistors, Prescott is
hiding some circuitry inside it, which Intel hasn't told us about yet. They
come to this conclusion by adding up the extra circuitry required by the
512K larger L2 cache, and the 8K larger L1 instruction cache, and estimate
that altogether this should take Prescott only up 80 million transistors
from the Northwood's 52 million. So there's still 45 million transistors
unaccounted for inside Prescott.

Tom's just seems annoyed by Prescott for some reason. :-)

Yousuf Khan
Click to expand...

Never mind the performance, feel the heat:

http://www.aceshardware.com/read.jsp?id=60000317

http://www.sudhian.com/showdocs.cfm?aid=489&pid=1815

Shame on sites that didn't do temperature readings.
 
I thought one of the more interseting comments from Anandtech is that
going from 2.8gz up to 3.2ghz Prescott narrows the gap (or increases
the lead) with Northwood at each successive higher frequency. They
extrapolate that Prescott at 3.6ghz and above might be a better
indicator of why the internal changes were made from the Northwood
core.
Click to expand...

It was interesting enough that one of my friends went to
compile the results from Anand and did a Prescott vs Prescott
comparison instead of P vs NW.

His conclusion was the scaling was all withing +/- 1%. I.e.
from 2.8 to 3.0 to 3.2, each % Mhz gives more or less the same %
performance increase.

Quite a few went below 1:1 so nothing as interesting as Anand
make it out to be. After all, why can't it be the NW core stapering
off and seeing no more benefits from upping the clock?

His estimated (512 vs 1MB cache can't directly compare as Mhz
vs Mhz) figures for the A64 bode better for the A64 scaling than the
Prescott :PpPpP

--
L.Angel: I'm looking for web design work.
If you need basic to med complexity webpages at affordable rates, email me :)
Standard HTML, SHTML, MySQL + PHP or ASP, Javascript.
If you really want, FrontPage & DreamWeaver too.
But keep in mind you pay extra bandwidth for their bloated code
 
prove me wrong, dont yust say it :)
http://images.anandtech.com/reviews/cpu/intel/pentium4/prescott/cpuz3
700.gif

fixed 14x multiplier. Its normal processor, not an engeenering sample
with open multiplier.
Click to expand...

That slide was just demonstrating the overclock potential - he didn't
include any benchmarks at 3.7GHz.
He tested 2.8GHz, 3.0GHz, and 3.2GHz versions of both Northwood and
Prescott:

http://www.anandtech.com/cpu/showdoc.html?i=1956&p=15

There is nothing that says he did this by overclocking the FSB, so I
assume he had genuine chips at each speed grade.
 
Johnno said:
That slide was just demonstrating the overclock potential - he
didn't include any benchmarks at 3.7GHz.
He tested 2.8GHz, 3.0GHz, and 3.2GHz versions of both Northwood
and Prescott:
Click to expand...

yes, I;m wrong
Its good to be human.


Pozdrawiam.
 
The articles posted show potentially some unknown mirroring of certain
functions, and the chip real-estate would tend to agree along those lines.

Sure, Intel could be "hiding" some features, or could it be...
o Mirroring to increase yield?
o Yields/Marketing eventually dictate a new feature is brought online?

I'd agree the Prescott figures are hardly dramatic.

However, size & yield may allow Intel to make aggressive price cuts:
o Intel can compete on performance - or it can beat on price
---- which mix is dominant depends on the innovation/mkt cycle
o Economic-Buyers & Consumers know a new socket is due
---- to stop people waiting, Intel can price-cut with good performance
---- impatient people, nice upgrade perf boost, greed works everytime
o Once bought, the motherboard+CPU are obsolete re last of Skt-478
---- Intel gets more upgrades per person per time

With the P4 v M rocketing wattage difference, it does look like Prescott
could be going out on a wattage-limb only for sharp power drops later.
The M has a far more heat-efficient architecture than the Prescott.

Present heatsink size is an interesting point here:
o N-3.2Ghz heatsinks are no spring chicken re size / weight / cfm
o N-3.2Ghz heatsinks on a Prescott-3.4Ghz run 15-20oC hotter
---- in open air, 18oC, some figures suggest 65oC vs 46oC
o Considering a Prescott-3.4Ghz is 103W at typical (not theoretical)
---- that would put a P-3.8Ghz around 114W & theoretical ~130W

For a 130W CPU, it's not the cooler-cfm that's the problem
o You need about 25-27cfm
The problem is the cooler's recirculation of that air
o Most impingement coolers in a case recirculate 40-70%
o Taking a figure of 50%, your cooler needs 50-54cfm to compensate
Clearly the problem is getting as low a recirculation of air as possible
o Which requires blow-thro coolers & instant removal of that heat

Basically, BTX is not anything too soon - but in itself the basic problem
of its design rears its head - CPU + Graphics + RAM all inline re heat.
Plus the reinforced motherboard/case to take gigantic coolers too.

It's not just an Intel problem - AMDs chips aren't that much lower.
Interesting :-)
 
Back
Top