mersenneforum.org  

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

Reply
 
Thread Tools
Old 2004-03-07, 01:50   #1
db597
 
db597's Avatar
 
Jan 2003

7·29 Posts
Default Upgrading to prescott

Does anyone know if the Shuttle SB61G2 can support a prescott cpu? It's pin compatible etc, but the power supply for these small PCs is rather feeble, and the prescott is supposed to be really power hungry. I'm hoping to upgrade, but am worried that it won't work (or the box will die prematurely).
db597 is offline   Reply With Quote
Old 2004-03-07, 18:04   #2
E_tron
 
E_tron's Avatar
 
Sep 2002
Austin, TX

56110 Posts
Default

may i suggest not moving to a prescott cpu?

In my mind, the regular pentium 4(northwood and canterwood) is much better than the prescott. Regular P4'sconsumes less power, has less stages in the pipeline(better performance), is more compatible with current motherboards(because of power, cpu steppings, ect...), is readily available(unlike the prescott which i can't find in retail yet), run cooler(thereby more overclockable or less of a convection oven), and are a proven platform(unlike the prescott).

In my mind, intel is trying to phase out the regular Pentium 4 cpu's in favor of the prescott, because it is far cheaper to create prescott cores than to create northwood and canterwood cores. Plus, intel can make many more prescotts per wafer, because of the die shrink.

anyhow, to answer your question :). It looks like that setup cannot support the prescott. If it could, then Shuttle would say it in their specs(but they don’t). http://us.shuttle.com/specs2.asp?pro_id=396 . It looks like the power supply is 220Watts. A prescott will take at least 100W of that.
E_tron is offline   Reply With Quote
Old 2004-03-07, 20:24   #3
db597
 
db597's Avatar
 
Jan 2003

7×29 Posts
Default Cheers for the reply

Hi E_tron,

Thanks for the advise. I'm actually interested in the Prescott due to it's 1MB cache. This is useful for one of the other CPU intensive applications (CFD) that I use - though my machine does run Prime95 for the other 12 hours a day.

It's interesting that the Shuttle supports the P4EE, but not the Prescott. This is rather strange to me - the P4EE uses even more power than the Prescott.

If you're looking for Prescotts, I found stocks in the UK for the 2.8GHz model (www.microdirect.co.uk and www.cpucity.co.uk). It's around US$210 or so.

With regards,

Duane Bong
db597 is offline   Reply With Quote
Old 2004-03-12, 00:30   #4
QuintLeo
 
QuintLeo's Avatar
 
Oct 2002
Lost in the hills of Iowa

26·7 Posts
Default

The primary reason for the Prescott design is that Intel wants to continue it's "Mhz Myth" leadership - and the Northwood P4 was starting to badly run out of steam on upgradeability.

AMD was also starting to get semi-close on pure Mhz to the Northwood with their faster Thoroughbreds, and the early Opteron/Athlon-64 series was threating to ramp past the P4 on Mhz by the end of the year - remember, it takes at LEAST a couple months just to test a new CPU design, with lead time on design, tapeout, fab, etc.

They appear to have quickly realised that the extra pipeline stages was hurting performance a LOT, though, thus the increased cache to at least partially compensate - and the resulting increase in power consumption and heat.

I suspect we're in for a rerun of the early Athlon vs. late P-III scenario, where AMD actually takes the LEAD on raw performance for a while, not just performance per unit cost - but this time AMD has actually forced Intel to FOLLOW THEIR LEAD with the AMD64 extensions. Looks like lots of fun in x86 CPUland for the next couple years.

8-) 8-) 8-) 8-) 8-)

Last fiddled with by QuintLeo on 2004-03-12 at 00:31
QuintLeo is offline   Reply With Quote
Old 2004-03-12, 12:20   #5
TauCeti
 
TauCeti's Avatar
 
Mar 2003
Braunschweig, Germany

2×113 Posts
Default

I suspect, that Intel may be in deep trouble with the Prescott thermal design. They have just released the 2.4-non HT Prescott to get even _some_ of them into the channel.

I do not think it is unlikely that the 3.0+ GHz prescotts have problems not to get damaged permanently if you run them for some month/years under load in a typical ATX system.

If the prescotts get too hot, the TCC kicks in and modulates the clock. That is nothing new compared to the Northwoods. But with the prescotts thermal characteristics and the high idle-power-draw the TCC will kick in often in a _lot_ of normal systems. And i cannot help but shudder if i think about extended periods of TCC-clock modulation (on/off) with close to 100 Amperes(!) being switched on that tiny 90nm core design.

Maybe (just speculation) the possibility of permanent damage is the fear Intel has and the reason for the delay of the higher clocked Prescotts?

Intel has a list of those thermally advantaged cases. Looks like boxed-Prescotts need cases that deliver a maximum internal temperature of 38C. And those cases mentioned in the link are supposed to manage that with an ambient temperature of 35C. Insane. No typical user will buy those kind of systems for normal home use in the years to come.
TauCeti is offline   Reply With Quote
Old 2004-03-15, 01:40   #6
db597
 
db597's Avatar
 
Jan 2003

7×29 Posts
Default Not so bad

Despite all the negative publicity, the Prescott still has much to offer:

1. x86-64 - this might actually be better than the AMD implementation. If it can process the same number of instructions per clock cycle, it will be like SSE2 on intel versus SSE2 on an opteron. And to think that Intel will be beating AMD at it's own 64bit game!

2. SSE3 and 1MB cache - for most applications, we've seen no improvement in performance between Northwood and Prescott. BUT, there are some cases where it would actually help.

One example is CFD and FEA code. These are very intensive in terms of CPU, and I normally run simulations for weeks at a time. Most of the time is spent in LOOPS calculating stresses, pressures etc for each cell in the computational domain. With this in mind, let me now point out that some loops that just won't fit into a 512KB cache. But with 1MB, you suddenly get a dramatic speed up, because everything can be done within the cache.

Old games and current applications naturally won't be affected (the loops are not as uniform in size, and don't get repeated as often - in CFD/FEA, you often run the same loop 100,000 times). Also, these existing software were compiled by their developers some time ago using compilers that expect a 512KB cache and 20 stage pipeline.

But the fact that the Prescott matches the Northwood despite the lack of optimisation for it's cache size, SSE3 and pipeline stages, is a good showing. When optimised software starts to come out, there's bound to be some speed up.

As for the heat, this might be similar to the TBred A and B scenario.
db597 is offline   Reply With Quote
Old 2004-03-17, 13:16   #7
QuintLeo
 
QuintLeo's Avatar
 
Oct 2002
Lost in the hills of Iowa

1C016 Posts
Default

> 1. x86-64 - this might actually be better than the AMD implementation. If it can process
> the same number of instructions per clock cycle, it will be like SSE2 on intel versus
> SSE2 on an opteron. And to think that Intel will be beating AMD at it's own 64bit game!

VERY doubtfull. Nothing in P4 has ever managed to match the Athlon in per-clock on the same instructions. Intel's x86-64 is going to suffer from the SAME extremely-long pipeline effects that all the other instructions on the P4 suffer from. For the record, SSE2 on the Opteron seems to run FASTER PER CLOCK than on the P4 - the P4 is clocked enough faster to make up the difference, but that gap is narrowing.

> 2. SSE3 and 1MB cache - for most applications, we've seen no improvement in
> performance between Northwood and Prescott. BUT, there are some cases where it
> would actually help.

Prescott seems to NEED that bigger cache to overcome it's other inefficiencies relative to the Northwood core - and most software seems to run slower on Prescott ANYWAY, per all of the benchmarks I've seen done on both.

This is looking like another P3 to P4 transision, just not AS wide a gap per-clock - but the Prescott doesn't clock enough faster *yet* to make up for the loss in instructions-per-clock. The Prescott thermal/power consumption problems aren't helping, though I suspect Intel will eventually get those under control.

> As for the heat, this might be similar to the TBred A and B scenario.

Exactly.

Looks like the next year or two in the x86 CPU world is going to be VERY interesting. 8-)
QuintLeo is offline   Reply With Quote
Old 2004-03-18, 04:08   #8
ColdFury
 
ColdFury's Avatar
 
Aug 2002

26×5 Posts
Default

Who cares how more "efficient" a processor is compared to another? I care about end performance and don't worry about how it achieves it.
ColdFury is offline   Reply With Quote
Old 2004-03-18, 11:06   #9
Dresdenboy
 
Dresdenboy's Avatar
 
Apr 2003
Berlin, Germany

192 Posts
Default

Quote:
Originally Posted by db597
1. x86-64 - this might actually be better than the AMD implementation. If it can process the same number of instructions per clock cycle, it will be like SSE2 on intel versus SSE2 on an opteron. And to think that Intel will be beating AMD at it's own 64bit game!
It's not that easy. There are differences in the way, how these CPUs implement 64bit operations. It's not the time to say "it will". Just some thoughts:
  1. As you can see in the photo shown here, 64 bit operation is only possible at lower clock frequencies than 32 bit (as expected due to tree like adder structures etc.) at the same voltage level. For 1.3V this is 4GHz vs. 7GHz. We know, that Prescott is planned to run at 4GHz at the end of this year or later. That means, the double pumped ALU inside this chip (I think it's still present) would run at 8GHz giving nearly the performance we'd expect from a 4GHz Northwood.
    But while executing 64 bit operations it is likely that the throughput could be lower. And even if not - don't expect that the performance relations between these CPUs with 32bit integer code would change much in 64bit mode (where we still can use 32bit but with twice the amount of registers).
  2. For SSE2 in 64bit mode the improvements could very likely be more different.
    First: The throughput is still the same. So the main areas for improvements are:
    • loosening the dependencies thanks to availability of more registers and
    • larger workingsets fit into the registers (reduced register spilling).
    Already the Northwood core helped addressing both issues because hyperthreading works around dependencies (by executing independend instructions of the second thread) and the load and store units are very effective. So the changes with availability of 64bit mode will be small (you can't be faster than max throughput).
    The K8 doesn't have SMT (but lower instr. latencies) and it's behaviour with SSE/SSE2 loads is not that good. In this case the extra registers will have more effect. An example can be seen here and here. Ignore the 64bit int stuff and look at divx encoding - 17% faster without marture compilers (maybe just some hand optimized main loops). 17% doesn't say much but 3.75 vs. 3.2GHz does I suppose


Quote:
Originally Posted by ColdFury
Who cares how more "efficient" a processor is compared to another? I care about end performance and don't worry about how it achieves it.
I agree, end users don't need to worry about that. However, I care about it as a programmer.
Dresdenboy is offline   Reply With Quote
Old 2004-03-18, 16:27   #10
nucleon
 
nucleon's Avatar
 
Mar 2003
Melbourne

5·103 Posts
Smile



What a complete bore CPUs are this year? (Speaking from Intel point of view)

The planned speed improvements for this year is maxxing @ 4.0GHz. Big deal. 25% clock improvment (vs 3.2GHz). Like most people are discussing this cpu atm - this clock improvement is needed to overcome the inefficiencies (longer pipeline etc..) Given current code, lets say 15-20% improvement _maybe_ for 4GHz. (okay I'm getting that figure from where the sun isn't shining)

But on the plus side - this is the year for the support teams to pull into line. New busses (PCI express x1- PCI replacement, x16 - AGP replacement), new RAM (DDR2), new case standards (BTX), and expect more SATA HDDs.

So my opinion, buying a system now, to upgrade to prescott etc.. is wasting a fair bit on your PC 'infrastructure'. Of course if you need a machine now, you could always go for say a basic 2.4-2.8 HT CPU now, and use it for a prime node when you grab your top end Tejas behemoth next year. :) If you really itching to buy something maybe upgrade the externals - monitor, UPS etc..

But as for CPUs - 2005 is the year to watch. Tejas with clock speeds 4-9GHz. And mutiple cores per die in server side.

Later this year to keep pushing the performance curve, I'd expect some multi-cpu machines to appear. These will come into their own if memory subsystems can improve to the hype around them.

-- Craig
nucleon is offline   Reply With Quote
Old 2004-03-21, 17:46   #11
QuintLeo
 
QuintLeo's Avatar
 
Oct 2002
Lost in the hills of Iowa

26·7 Posts
Default

Right now, it looks like BTX is being soundly rejected by the hardware manufacturers, but the other new stuff like PCI Express is starting to show up.

I'd agree that buying a P4 system now expecting to upgrade it to Prescott is not a good idea, though - even IF the motherboard will support the power requirement (which is REAL iffy on most current P4 motherboards) of the current Prescott CPUs, it's likely to be missing other features by the time a Prescott CPU is finally released that has better performance than the current Northwood CPUs. Then there's the 64-bit question.

Welcome to the next version of the PIII to P4 transition....
QuintLeo is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Upgrading a cat. 0 box ckdo PrimeNet 3 2018-03-12 15:24
Upgrading a LGA 775 foxmccloud123 Hardware 14 2013-05-19 06:09
Upgrading a 1U server fivemack Hardware 4 2012-12-01 17:44
Prescott ET_ Hardware 6 2004-07-06 02:38
P4 Prescott Benchmarks! Dresdenboy Hardware 0 2003-09-28 19:37

All times are UTC. The time now is 08:40.

Fri Oct 23 08:40:35 UTC 2020 up 43 days, 5:51, 0 users, load averages: 0.89, 1.22, 1.30

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.