mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Hardware

Reply
 
Thread Tools
Old 2004-02-03, 16:38   #1
Angular
 
Aug 2002

2×33 Posts
Default P4 Prescott - 31 Stage Pipeline ? Bad news for Prime95?

I have been reading the new Prescott reviews with a bit of shock. I expected the prescott doubles both the L1 and L2 cache. The biggest surprise to me is that the integer pipeline increased from 21 stages to 31 stages. This compares to the 10/15 stage (integer/FP) pipeline of the Athlon XP. I am not an assembly programmer but this sounds like it could be trouble for Prime95's highly optimized asm code. Given the market dominance of Intel it may be more critical to fix this problem (if it exists) than to fix the Operteron.

I recall George saying that SSE3 (PNI - Prescott New Instructions) did not look intresting. However, in anandtech's review he says that ADDSUBPS, ADDSUBPD, MOVSLDUP, MOVSHDUP and MOVDDUP are helpful in FFT.
http://www.anandtech.com/cpu/showdoc.html?i=1956&p=9

Does anyone have a Prescott CPU they can test using the existing code?

Assuming their is a Prescott preforms seriously slower with the current Prime95 code, does anyone support the creation of a fund to purchase a Prescott CPU for the development team? I pledge my support.

http://www.aceshardware.com/read.jsp?id=60000315

Intel Prescott: We expected a Lexus and got a Hyundai:
http://www.theinquirer.net/?article=13947

What's wrong with the Intel Prescott?
http://www.theinquirer.net/?article=13927

Last fiddled with by Angular on 2004-02-03 at 16:43
Angular is offline   Reply With Quote
Old 2004-02-03, 19:56   #2
adpowers
 
adpowers's Avatar
 
Sep 2002

1208 Posts
Default

My friend says he will be getting a Prescott for free. He doesn't know how long it will take, but he estimated a month. We'll see if really gets it :).

He is starting to become a stats whore like myself, so I'll ask if he'll want to help beta test the Prescott improvements.

Andrew
adpowers is offline   Reply With Quote
Old 2004-02-03, 20:13   #3
E_tron
 
E_tron's Avatar
 
Sep 2002
Austin, TX

10618 Posts
Default

It looks like intel has hinted the pending death of socket 478. It looks like intel is spending a fortune making 3.2 and 3.4ghz northwood processors. I feel that socket 478 will become a bargain system, so intel wishes to replace northwood with the less expensive prescott @ .09u.

Tejas will come to replace the northwood, canterwood, and prescott in about 7 months, so i feel that a expensive optimization project will be wasted.
E_tron is offline   Reply With Quote
Old 2004-02-03, 22:27   #4
gowen72
 
Aug 2002

72 Posts
Default

The increased cache latency time also looks like it will cause a considerable slowdown.

The anandtech article is interesting because it suggests that with increasing clockspeed Prescott starts to outperform a similarily clocked Northwood. Very similar to what happened when P4's were initially introduced, when compared to a P3

So at the currently available clock speeds, staying with Northwood seems to be the way to go. Now by the end of the year when the Prescott reaches 4 GHz, that might all change.
gowen72 is offline   Reply With Quote
Old 2004-02-03, 22:27   #5
ColdFury
 
ColdFury's Avatar
 
Aug 2002

1010000002 Posts
Default

Long pipelines are typically a problem with branch intensive code. I don't think a FFT qualifies, so it shouldn't hurt Prime95 that much.
ColdFury is offline   Reply With Quote
Old 2004-02-04, 05:30   #6
ColdFury
 
ColdFury's Avatar
 
Aug 2002

1010000002 Posts
Default

On the other hand, the increased cache latency might hurt Prime95.
ColdFury is offline   Reply With Quote
Old 2004-02-04, 18:04   #7
Angular
 
Aug 2002

2·33 Posts
Default

Quote:
Originally Posted by E_tron
It looks like intel has hinted the pending death of socket 478. It looks like intel is spending a fortune making 3.2 and 3.4ghz northwood processors. I feel that socket 478 will become a bargain system, so intel wishes to replace northwood with the less expensive prescott @ .09u.

Tejas will come to replace the northwood, canterwood, and prescott in about 7 months, so i feel that a expensive optimization project will be wasted.
Why is intel speding a fortune making 2.3/3.4 GHz Northwood processors? What are the Tejas changes beyond the socket 478 -> socket T?
Angular is offline   Reply With Quote
Old 2004-02-04, 19:21   #8
gowen72
 
Aug 2002

72 Posts
Default

It is more expensive to make Northwood's simply because the chips are bigger, so you don't get as many chips per wafer. I can't remember the figures but I think it was something like 30% more Prescott's per wafer.

It makes sense for Intel to keep making Northwoord's at the moment until they can ramp up the production and yields of the Prescott chips. But it is pretty obvious that Intel want to stop producing Northwood's otherwise they would have priced Prescott's at a premium.

The other benefit of Prescott chips is that since they have a longer pipe, the clock speed can be increase pretty quickly. Which is when they'll start out-performing Northwoods.

Isn't Tejas, rumoured to have x86-64 (ala Athlon64) or something?
gowen72 is offline   Reply With Quote
Old 2004-02-05, 07:54   #9
QuintLeo
 
QuintLeo's Avatar
 
Oct 2002
Lost in the hills of Iowa

26·7 Posts
Default

Quote:
Originally Posted by gowen72
The increased cache latency time also looks like it will cause a considerable slowdown.

The anandtech article is interesting because it suggests that with increasing clockspeed Prescott starts to outperform a similarily clocked Northwood. Very similar to what happened when P4's were initially introduced, when compared to a P3
The P4 on *most* work has NEVER matched a P3 per clock - even the upgraded cache in the Northwood P4s didn't help enough to overcome that entirely. I doubt that the huge cache on the "P4 EE" managed it either.

The *only* reason the P4 eventually passed the P3 on performance is that the P3 topped out at 1.4 Ghz - and it took Intel a lotta work to get it that far.

I seriously doubt that the Prescott will ever match the Northwood on per-Mhz efficiency - it appears that Intel is going the "more pipeline stages to increase the CPU Mhz faster" route yet again, since that lets them keep their "Mhz matters" BS ad campaign going.
QuintLeo is offline   Reply With Quote
Old 2004-02-24, 11:02   #10
thestu
 

32·5·59 Posts
Default

well i just recived a prescott the day and i already notice the improves ments on the cache. I have 1 machine that has a northwood 3.2ghz and my new prescott is a 3.0ghz both machines are the same spec apart from the processor obviously and i see my prescott out running the northwood easily
  Reply With Quote
Old 2004-09-06, 07:04   #11
Dresdenboy
 
Dresdenboy's Avatar
 
Apr 2003
Berlin, Germany

192 Posts
Default

Michael S, who helped together with magicfan to measure the Banias FPU throughput/latency values, now published results for Nocona/Prescott:
http://realworldtech.com/forums/inde...7801&roomID=11

The long pipeline causes really high latencies for SSE2 instructions (compared to Northwood).
Dresdenboy is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
GMP-ECM & Prime95 Stage 1 Files Gordon GMP-ECM 3 2016-01-08 12:44
How to use prime95 for stage 1 & GMP-ECM for stage 2 Prime95 Lone Mersenne Hunters 111 2015-06-13 03:18
The news giveth, the news taketh away... NBtarheel_33 Hardware 17 2009-05-04 15:52
Stage 1 with mprime/prime95, stage 2 with GMP-ECM D. B. Staple Factoring 2 2007-12-14 00:21
Prescott impact to prime95. nucleon Hardware 35 2003-07-22 06:02

All times are UTC. The time now is 12:50.

Mon Sep 28 12:50:13 UTC 2020 up 18 days, 10:01, 1 user, load averages: 1.53, 1.75, 1.86

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.