Many people seem to have no idea about the differences between the Northwood and the Prescott. This guide will hopefully give a rough idea about the performance difference (or lack thereof) between the Northwood and the Prescott P4.

Info from this thread is a result of compiling and analyzing data from various webpages and some intel tech documents.
Some of the references:
http://www.aceshardware.com/
http://www.digit-life.com/articles2/rmma/rmma-p4.html
http://www.anandtech.com/cpu/showdoc.html?i=1956

Below is a list of basic differences between the Northwood and Prescott. More details to come soon.


Attached thumbnail(s)

The big question:
Why doesn't the Prescott with its twice larger L1 and L2 caches and architectural enhancements outperform the Northwood with ease? (refer to table above for numbers)

Prescott Advantages over the Northwood:
1. Twice larger L1 cache.
2. Twice larger L2 cache.
3. Integer Shift/rotate functions can run much faster due to improved ALU
4. Integer Multiply functions are much faster due to new dedicated integer multiplier (Northwood integer multiplies are done by the FPU unit)
5. New improved branch predictor to reduce pipeline stalls

Prescott Cache Disadvantages compared to Northwood:
1. Double the L1 cache latency
2. ~50% higher L2 cache latency
3. Half the L2 cache bandwith
4. 50% longer pipeline


ANALYSIS:
1) Although the cache sizes of the Prescott are twice higher, the cache latencies are also greatly increased, which somewhat offsets the increase in cache size. Caches generally work best when they are small and fast, not big and slow.
Also, according to the testing done by the guys at Digit-Life, the Prescott's L2 bandwith is only half of the Northwood. Perhaps intel reduced the Prescott's L2 bus width to 128bit (Northwood is 256bit).

2) The larger cache can barely offset the performance of the 50% longer pipeline (31 vs 21 stages). Longer pipelines means that a pipeline stall (due to branch misprediction) will be much more detrimental to the performance.

CONCLUSIONS:
Due to 1) and 2) above, the Prescott's performance is slightly lower than the Northwood in most situations, but they are roughly on par with each other.

An interesting observation made by the guys at Aceshardware is the power consumed per transistor of the Prescott. It is actually ~46% lower than the Northwood! But the immense number of transistors (more than double the Northwood) add up to make the total heat output higher.


So if you are looking for a P4 at the moment is get a Nortwood and stay away from the Prescott until Intel fixes the heat issue with the Prescotts, because the Prescotts feature a much higher heat output and lower performance than the Northwood.

This high power usage and heat output is due to the transistors leaking a lot of current ( 40% even in the OFF state!) with the new 90nm manufacturing process. The problem is happening with everyone's 90nm process (yes, AMD and IBM included) at speeds roughly above 2.5GHz.

OK, I'm sorry for mentioning AMD here. It's not my intention to create another "Intel vs AMD" debate here.

Don't you think that Intel's designing strategy has gone wrong ? A newer chips run even slower than the older one. And they have problem to increase the core frequency due to the excessive heat.

HAhahah....
suddenly change to AMD pulak..

I think P4E will sure catch up and perform faster than P4C. Later, will there be presscott with 1000mhz/1200mhz FSB ? I think this will really boost presscott's performance...

is pentium4 2.4A a prescott???

Nice 1 there ikanayam . So we hope there's no more stupid thread again , and shouldn't u pin it up ?

Did a quick search and found out that there is 2.4A Prescott. That's indeed confusing because there is a 2.4A Northwood as well.

Speaking of this new Precott, I see that it supports SSE3.

What extra does this SSE3 has to offer ? And what kind of application will take more advantage of it ?

Thanks for all the info.

Some new apps will take advantage of it especially those that got to do with encoding/decoding.

so 2.4c is pres or north?i m really noob about all of this

the Prescott is very Hot, while using the original Fan from intel
the Northwood is better, the 2.4c is better for overclocking......

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Amd Athlon XP 2500+ (Overclock to Athlon XP 3200+ = v1.75)
Thermaltake Silent Boost (Hydro wave bearing fan)
Abit AN7 v1.0 BIOS date 03/04/2004
Cosair 256mb PC3200 X2
GeXCube 9600XT Hardcor Gamer 2.8ns 128mb [570/700]
Creative Sound Blaster Audigy DE
Maxtor 120G SATA HDD
Samsung 523252 & Samsung 16x DVD
LianLi PC-07 full with room cover and Temperature detecter
Creative Gigawork S750
Samsung 753DF monitor

2.4 - Northwood 400Mhz FSB
2.4B - Northwood 533Mhz FSB
2.4C - Nortwood 800Mhz FSB HyperThread
2.4A - Prescott

yayaya the 2.4A Prescott is extream overclocking

review at HardOcp......
http://www.hardocp.com/article.html?art=NjAz

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Amd Athlon XP 2500+ (Overclock to Athlon XP 3200+ = v1.75)
Thermaltake Silent Boost (Hydro wave bearing fan)
Abit AN7 v1.0 BIOS date 03/04/2004
Cosair 256mb PC3200 X2
GeXCube 9600XT Hardcor Gamer 2.8ns 128mb [570/700]
Creative Sound Blaster Audigy DE
Maxtor 120G SATA HDD
Samsung 523252 & Samsung 16x DVD
LianLi PC-07 full with room cover and Temperature detecter
Creative Gigawork S750
Samsung 753DF monitor

this is the box lable, serial number....

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Amd Athlon XP 2500+ (Overclock to Athlon XP 3200+ = v1.75)
Thermaltake Silent Boost (Hydro wave bearing fan)
Abit AN7 v1.0 BIOS date 03/04/2004
Cosair 256mb PC3200 X2
GeXCube 9600XT Hardcor Gamer 2.8ns 128mb [570/700]
Creative Sound Blaster Audigy DE
Maxtor 120G SATA HDD
Samsung 523252 & Samsung 16x DVD
LianLi PC-07 full with room cover and Temperature detecter
Creative Gigawork S750
Samsung 753DF monitor


Attached thumbnail(s)

Eh eh, Prescott is 533Mhz FSB and not 800Mhz FSB ?? Is it HT capable ??

it probably is
the A there stands for 533mhz FSB and not prescott only

i dun think so....the website tat michaelpng gave said tat the 2.4A dun have HT support.....

but then again..........B shows 533mhz FSB..........blur!!!!
*im lost

2.4A is Prescott only with no HT and a 533MHz FSB.
2.4B is Northwood with no HT and 533MHz FSB
2.4C is Northwood with HT and 800MHz FSB

I believe someone pointed this out somewhere in an earlier post in this thread.

The 2.4 A and B versions dun have HT
C, E and EE got HT

Thanks./.....

Sorry to butt in, but i think Prescott have more potential than Northwood, but not the current revision. Future revision b4 the new socket will see much more improvements, hopefully around the region of 4ghz or 5 ghz for overclockers.

Currently this is what i am able to hit, with AI7. Not much difference if i run with IC7 MAx3 tho, even with PAT disabled, AI7 kicks IC7 MAx 3 asses.

This post has been edited by Solty8: Apr 9 2004, 02:04 AM


Attached thumbnail(s)

Nobody doubts that Prescott will OC higher than Northwood in future revisions. The thing is, in its current state, you risk frying your mainboard when OCing a prescott, and i don't think that is a risk many people want to take. Currently, they OC on average only slightly better than the Northwood, and i think it is not worth the heat, the risk or even the performance.

Prescott will need the new socket in order to reduce the heat output. The greater number of pins will ensure a better power delivery methods and therefore reduce the heat output somewhat. Future revisions to the process technology will hopefully help also.

P4E needs very long pipeline stages in order to ramp up to high clock speed in the future. It also uses larger cache to compensate the long pipeline penalty. Well, it claims that the branch prediction has improved significantly. It's normal for longer pipeline CPU to perform weaker due to the pipeline latency in every stages, if 1 stage causes 1ns latency, 31 stages causes 31ns latency compare to 20ns latency on northwood. Moreover, longer pipeline means a task can only be accomplished after 31 stages of pipeline, compare to 20 stages of northwood. Therefore, if no enhancement on cache, branch prediction or prefetching, 31 stages of pipeline will definitely slower than 20 stages pipeline CPU at the same clock speed. Main advantage of having long pipeline is to ramp up clock speed , because each task are divided into more stages, where each stage equal to 1 clock cycle.

This is wrong on so many levels that I just can't keep quiet about it... sorry charge-n-go, nothing personal.

oh the decoder adds in another 8 stages to the pipeline

maybe you need to get a better HSF

It's suppose to work well with stock settings. Anyway, plannng to buy a huge table fan and blow straight into it.

The power of 100Watt ++ CPU

jarofclay, how u measure your PWM temperature?

Using CD4's thermal probes and also Abit's mGuru thingy. Normally, I would minus 10C from Abit's readings and it would serve up the correct figure, matching with my CD4 readings.

can somebody confirm the L2 cache bus width? 128bit? 256bit?

some one good enough to post a cpu-z image regarding the L2 cache for prescott?

what is command extension such as MMX, SSE, SSE2,SSE3, 3DNow Pro & AMD64?

Thats what I am talking about....sadly what a drawback it is for the Prescott....

u can say that again n again...

user...

erm.....
bus width is certainly 256-bits
i dun know how to post the picture
coz upload size is 100k oni
mininum cut here cut there is 1mb
and refer to nutz's aceshardware link
the bus width of prescott is 256-bits
if the bandwidth is half the size of northwood
then how is their performance on par? (i know northwood better then prescott, but not THAT much)

edit:and ikanayam gogo ar, better change the picture in the first page
coz the picture says bus width of prescott is 128-bit oni

This post has been edited by Lucifer: Apr 22 2004, 12:12 AM

I don't know what methods those 2 used, but digit-life's analysis was very complete and measured the useable bandwith of the Prescott's L2 to be half of the Northwood's. It's not exactly half, but it's more or less half. Looks like the bus speed may have been slowed down somehow, like being made to transfer data every other clock instead of every clock or something like that.

Review here: http://www.digit-life.com/articles2/rmma/rmma-p4.html

fishy:

What i believe the problem with the benchmark that gave only 1/2 the real L1-L2 bandwidth is because the new set associativity parameters. Its either the set associativity increased from 4 way to 8 way or the string length increase from 64-bytte to 128-byte in prescott. (I can't really remember). What i'm guessing here is that the data is not aligned properly with the new set assosiativity settings.

So don't worry about it... besides the P4 has a write through L1 cache (i.e. writes directly to the main memory, skipping the L2 cache all together).

This post has been edited by nUtZ`: Apr 22 2004, 02:11 AM

Thanks for explaining nUtZ`

Yeah thanks for the explaination NutZ

here's something I got from digging around Intel's Technology Journal

Willamette Block diagram


Prescott Block diagram


No Northwood loving, unfortunately. Both diagrams were cut-n-paste from pdf articles available at the link above (prescott on the front page, willie in the archives). Anyway that ought to put to rest the 128 bit wide bus width from L2. So the mystery remains where does the bottleneck lie?

nUtZ`: L2 cache line length is 128 bytes long on all the P4 implementations, apparently, and the L1 D-cache line is also 64 bytes long throughout. You raised a good question about data alignment. L2 is a unified instruction+data whilst L1 is just data, after all, hmm... According to the original Willamette paper, the 128 byte L2 cache line is further divided into two 64-byte sectors, hence L1<->L2 data transactions should be 64-byte aligned.

Not sure if the increasing the L1 set associativity from 4-way to 8-ways has anything to do with the lacklustre performance. I mean, after x mod 4 vs x mod 8 is just a 1-bit difference; one extra shift to the right. one more bit to mask out.

This post has been edited by silkworm: Apr 22 2004, 03:54 PM

for those who are commenting prescott L2 cache bandwith but do not own a prescott, i can confirm here:

Prescott L2 Cache is 256Bits
Prescott have 1MB L2 Cache, Twice as much as Northwood
Prescott L2 Cache have bigger latency, Twice as much as Northwood

Last come to last, precott is not really that bad compare to northwood.

i just got my prescott yesterday,

SuperPI 1M 39second, prescott 2.8E oc at 3.5GHz
SuperPI 1M 43second, nothwood 2.4C oc at 3.3Ghz

but Prescott is really hot, get yourself a water cooling kit before OC it...

They focus more on synthetoc and application benchmarks
Those are where the L2 cache power comes into place

I guess it has been mention few times in the forum that I will surely support future applications and windows. Hopefully it will make things faster with SSL3.

finally i got my fact right about these two..thanks ikan n ayam

got a customers 2.8e system with heat problems in yesterday
idle was at 46, after 3dmark2001SE about 60 degree C

room temperature is at 20 degree C, im just reading from the room thermometer.

after some mods and cable management, idle was at 35. didnt have time to do 3mark2001SE cuz customer needed his system back for work

Go to tomshardware website
Got comparison between Northwood and Prescott at 4Ghz

i dun really trust tomshardware anymore.....
i only believe in 50% of what they say

anyway the prescott seems to perform poorly in gaming..

Let's compare this 2 benchie

Man the Pressie really sux bad
The Northy is better

the prescott is actually very good, only at higher clockspeed... but at lower clockspeed it's hot and slow.
Anyway, tomshardware looks like an INTEL promoter to me...
Their benchmark shows that Prescott is faster than Northwood even at slower speed...

E has heat issues at high speeds, already consume 100W++ at stock 3 Ghz unless you got super cooling ler

So its better to take the C version as it is cooler and one more thing, you planning to get the Nvidia 5600, why not go get a 9600PRO instead??

im using
2.8E (standard hsf)
is7-E Intel i865-ICH5 mobo
GEIL DDR-SDRAM PC3200 - 256 MBytes x 2
wd 80Gb

before this....my idle cpu temp ~61 sys temp ~41 pwm 50
overload cpu temp ~79 syt temp 50 pwm 55

after i flash my bios with latest abit flashmenu i got this

idle cpu 51 sys 45 pwm 45
overload 65 sys 45 pwm 50

is it something todo with winbond sensor or bios?

This post has been edited by slainn: May 27 2004, 11:48 AM

haha, the bios is cheating u loh~ ahah, just kidding.

brother, 3.2c is faster than 3.0e lah....

The sensor got some reading problems and u need to flash to the latest v20 BIOS to bring down the temp

The Prescott is better when it comes to overclocking, but for stock the Northwood kicks its butt

the prescott (2.8ghz) im having is now reaching 69degrees!!!

Hahhaa, now only you know ah Ket
Should choose Northwood not Prescott
Get a super cooler ler, necessity for Pressie procs

Go for northwood...!no temperature problems...

"Overclocking notes:

The PR does require good cooling for a good OC, even with the higher threshold.

The PR is capable of more per clock cycle than an NW. this means that a 200mhz OC on a PR will give a bigger performance boost than an eqivalent OC on an NW.

The PR and NW even out performance-wise, at around 3.4-3.5 ghz. The PR really starts to overtake the NW at around 3.6-3.7.

The PR has more potential as it is designed for way higher clocks than the NW which is reaching its limit."

taken from http://www.hardcoreware.net/

.. i like this quote too :

"To put it bluntly: Prescott becomes interesting after 3.6GHz; in other words, after it has completely left Northwood's clock speeds behind." from http://www.anandtech.com/cpu/showdoc.html?i=1956&p=25


...so u see, it ain't that bad after all. just slap on a big heat sink (or water cool) and fan, oc it til it's sizzling and some salt,spice... and u should be able to beat some NW a$$!

I'm gonna be a guinea pig coz i've got a 2.8e. Wish me luck!

This post has been edited by bum8um: Jun 10 2004, 11:43 AM

The only problem is, not many of us are willing to risk frying our mainboards or spend a few hundred ringgit more to go into watercooling when many of the 2.8c M0 northwoods are hitting 3.6+ GHz at little or no voltage increases, and without the power or cooling issues that the prescott presents.

hmm...after reading all the arguments, i am still at a cross-road

should i
1. get the 2.8C
2. get the 2.8E

their price is the same ( quoted from a pricelist), i am planing to change the stock HSF to something from THermaltake or coolermaster or something equivalent, and my system is going to be a micro-atx ( i.e. tiny tower).

any sugestions?

2.8C

better mobo er ...

try abit A17

gigabyte 1100k or soemthing like that

asus p4p800

question: is the IS10 from ABit a good Mobo for the northwood processor?

just install p4 3.0e abit vt7 (via pt880) cheap board, my advice is, don't try it. maybe this board is meant for lower end market n i should not use a 3.0e on it. now swich to gigabyte 848 board (still consider cheap board, but w/ gigabit lan), then okay, stable liao. but then, don't use prescott in a office environment, stock fan sound like there is a motobike in your office, with engine on.

ya...low end motherboard just for the low end Prescott P4 2.4A.
My DFI PT880 run this prescott very well, stable... no noisy sound too.....But still dont know what is my processor temperature running now , bcoz the MB dont have sensor for temperature(bcoz the cheap n low end Mbaord)
PC Running 4 night n 5 day without shutdown.........using Overnet to downloading movie n music, then playing Ragnarok........
!st night, restart for try the 3DMark2001se, can score 11071(Little bit low) with my G4 Ti4200 n Kingston 512MB PC3200.
Then try C&C General, ok!
2ND night,restart > try WE7Int n NFS:UG.... no problem.... finish download GXngYr SnZps 2. After movie, Ragnarok again.
3rd night, restart> try Silent Hill 3, WE7Int n Ragnarok again.
tonight will play WE7Int for warm up, later will go watching England VS Swiss match...


PS: X=i , Y=e , Z=a

may i know wat is cache??wat is that use for??

hi all,

just wanted to know is there any heating or other issues on the Extreme Edition (EE) of Pentium 4 CPU?? hope to hear some views on this.


cheers mates.

not sure.but the presscott im using is darn hot.can reach abt 69 degrees! only a P4 2.8