Why is Intel's 90nm may not be a factor in Prescott power consumption

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Okay there was a forums topic about comparing Intel's and IBM's 90nm process. It suggests that 90nm process may not be the contributing factor(or not as much) as most people think.

Link: http://www.aceshardware.com/forum?read=115069088#TEXT

After reading look at my article, which is next paragraph. According to rumor(discussed even in Anandtech), Prescott's transistor count somehow doesn't up. Most people predicted its disabled transistors in Prescott. It's possible. Hyperthreading was apparently disabled in Willamette. Which was actually a good thing because performance loss would have been big with first-gen HT(for example the Xeons). Now its like, we can start a war about whether HT loses performance or not. I don't think EMT64(aka CT), or LaGrande will amount to much since 64-bit needs like 2% die and LaGrande, according to www.chip-architect.com pictures, it is a small portion of the die. So is write-combining buffers, branch prediction enhancements, 11 more pipeline stages and Trace cache improvements. Some transistors must be disabled, right? It's too many extra transistors that we don't know its functions. But first please look at the Aceshardware link.


You can't assume that power consumption is lower in the processors that have caches and logic disabled. Look at the Pentium III and the Pentium III core Celerons.

Link Celeron PIII: http://processorfinder.intel.com/sc...ocFam=49&PkgType=ALL&SysBusSpd=ALL&CorSpd=ALL

Link Pentium III: http://processorfinder.intel.com/sc...ocFam=25&PkgType=ALL&SysBusSpd=ALL&CorSpd=ALL

Both at 1GHz, both at 0.18 micron, both at 100MHz bus, both cDO stepping, both at 29.0 watts. Difference? Pentium III has higher L2 cache. That tells disabling the caches(and probably logic too) has no effect on wattage(Even though the cache consumes very little power, it consumes some power, so according the theory of most people, power consumption should be greater on the Pentium IIIs since it has more cache enabled). So yeah, according to the Aceshardware article(Yes I know its a rough estimate), Prescott is having more advantage by going to 0.09 micron than PowerPC does(or Intel has more advantage than IBM).
Want another link? Here look at P4 and P4 Celerons.

P4 Celeron link: http://processorfinder.intel.com/sc...ocFam=49&PkgType=ALL&SysBusSpd=ALL&CorSpd=ALL

Pentium 4: http://processorfinder.intel.com/sc...cFam=483&PkgType=ALL&SysBusSpd=ALL&CorSpd=ALL

Again, both at 1.8GHz, both at 400MHz bus, both E0 stepping, AND both at 66.1W. Difference(Again)? L2 cache size. So, disabling transistors don't do anything to power consumption decrease(I don't like that fact but its true). Interesting Eh? Take time to read it and I would like to see some informative opinions about this topic.
 

MichaelZ

Senior member
Oct 12, 2003
871
0
0
I'll tell u what the problem is right here and now. Socket 478. 478 Prescott is a reborn 423 Willamette going through public testing.
 

THUGSROOK

Elite Member
Feb 3, 2001
11,847
0
0
the northwoods are fast enough.
intel has a problem, and that problem is that we dont need faster cpus.
 

BigBadBiologist

Platinum Member
Nov 30, 2002
2,156
0
76
Originally posted by: THUGSROOK
the northwoods are fast enough.
intel has a problem, and that problem is that we dont need faster cpus.

This quote comes to you from one of the best overclockers on the board :)
 

clarkey01

Diamond Member
Feb 4, 2004
3,419
1
0
Intel?s problem is that they think or USED to think more Ghz sells more chips, which is relatively true, since you still get dum ass?s buying 2.6 Celerons when they could probably get a 2500+ Barton for roughly the same price, or cheaper.

Back to the point, they feel the need to scale much higher in frequency even if it means lower the value of the Mhz & increasing heat, e.g ?Nehalem? is planned or was planned to reach 10.20Ghz in 2005, AMD would probably be only at 3.5ghz if that. A customer sees a 3.5ghz chip & a 10.20ghz chip, forget the AMD?S number system, or at least that what Intel hopes they do.

I think AMD/IBM have the right approach, higher instructions per clock for a start, its not like they both have the heat problems of Prescott.
 

Acanthus

Lifer
Aug 28, 2001
19,915
2
76
ostif.org
Originally posted by: clarkey01
Intel?s problem is that they think or USED to think more Ghz sells more chips, which is relatively true, since you still get dum ass?s buying 2.6 Celerons when they could probably get a 2500+ Barton for roughly the same price, or cheaper.

Back to the point, they feel the need to scale much higher in frequency even if it means lower the value of the Mhz & increasing heat, e.g ?Nehalem? is planned or was planned to reach 10.20Ghz in 2005, AMD would probably be only at 3.5ghz if that. A customer sees a 3.5ghz chip & a 10.20ghz chip, forget the AMD?S number system, or at least that what Intel hopes they do.

I think AMD/IBM have the right approach, higher instructions per clock for a start, its not like they both have the heat problems of Prescott.

They both have SOI now, Intel doesnt. When (not if) intel implements SOI on their CPUs they will dramatically decrease gate leakage, which will decrease the amount of heat they put out dramatically, as well as allowing higher clockspeeds.
 

Alkaline5

Senior member
Jun 21, 2001
801
0
0
Originally posted by: clarkey01
Intel?s problem is that they think or USED to think more Ghz sells more chips, which is relatively true, since you still get dum ass?s buying 2.6 Celerons when they could probably get a 2500+ Barton for roughly the same price, or cheaper.

Back to the point, they feel the need to scale much higher in frequency even if it means lower the value of the Mhz & increasing heat, e.g ?Nehalem? is planned or was planned to reach 10.20Ghz in 2005, AMD would probably be only at 3.5ghz if that. A customer sees a 3.5ghz chip & a 10.20ghz chip, forget the AMD?S number system, or at least that what Intel hopes they do...

I'm not knocking you--you're right--but you're post got me thinking. The recent announcements of 1) Intel's Model Number system and 2) the planned re-desktop-ification of the Pentium-M in 2007 indicate that the GHz marketing push has backfired somewhat, as Intel now seems to be moving toward performance instead. These announcements along with the Prescott release and the "sudden" adoption of AMD64 make me wonder wtf is going on at Intel these days?
 

Stoneburner

Diamond Member
May 29, 2003
3,491
0
76
THe inquirer has an article on that question

http://www.theinquirer.net/?article=15029

WHat confuses me is what they are doing with teh notebook pentium M successors, DOthan. OKay you made a more efficient chip, excellent performance and battery life.... so you just decide to raise the ghz and fsb and consequently the heat and wattage as a follow up? DId they at least consider working harder on greater efficiency AT Least for notebooks?
 

zephyrprime

Diamond Member
Feb 18, 2001
7,512
2
81
But do the missing transistors in the Prescott so numerous that they can account entirely for the increased power consumption?
 

Accord99

Platinum Member
Jul 2, 2001
2,259
172
106
Originally posted by: zephyrprime
But do the missing transistors in the Prescott so numerous that they can account entirely for the increased power consumption?

The number of unaccounted transistors is about 30-40 Million in Prescott; the entire Northwood core is only around 27M transistors.
 

Budman

Lifer
Oct 9, 1999
10,980
0
0
Originally posted by: BigBadBiologist
Originally posted by: THUGSROOK
the northwoods are fast enough.
intel has a problem, and that problem is that we dont need faster cpus.

This quote comes to you from one of the best overclockers on the board :)

better wipe that Brown stuff off your nose. ;)
 

Sahakiel

Golden Member
Oct 19, 2001
1,746
0
71
Originally posted by: Stoneburner
THe inquirer has an article on that question

http://www.theinquirer.net/?article=15029

WHat confuses me is what they are doing with teh notebook pentium M successors, DOthan. OKay you made a more efficient chip, excellent performance and battery life.... so you just decide to raise the ghz and fsb and consequently the heat and wattage as a follow up? DId they at least consider working harder on greater efficiency AT Least for notebooks?

Tweaking a chip takes a few months and do relatively little.
Redesigning a core takes years but the results can be dramatic.
Increasing clock speed is still viable. We haven't reached optimum pipeline length yet. Obviously, diminishing returns kicks in as we get closer.
Increasing IPC has the unsavory effect of decreasing clock speed. K6 sucked. K7 thoroughly beat it into the ground. It's all a question of balance.
 

clarkey01

Diamond Member
Feb 4, 2004
3,419
1
0
Where does Intel go from here ? Does anybody know what time frame their numbering system will come into play in desktops? Are Tejas & Nehalem still going to be labled by Ghz or by numbering?

For now it looks like presscott will reach 4ghz by the end of the year, there is an article on Intel over coming heat problems.
http://www.theinquirer.net/?article=15062
 

TerryMathews

Lifer
Oct 9, 1999
11,473
2
0
The disabled transistors theory makes a lot of sense. Normally with electronics, you cut the data line to disable something and leave the Vcc intact, as doing the opposite can cause problems with the data bus.

It will require more work, but Intel will realize a significant power savings if they come up with a way to cut both the data line and power line from unused sections of the chip. Also note that when I say line, I use it in a figurative sense as semiconductions don't use wires in the same sense that we use them in the real world. 'Line' would refer to what is most likely a part of the plane the transistor is built upon.
 

apoppin

Lifer
Mar 9, 2000
34,890
1
0
alienbabeltech.com
Originally posted by: i82lazyboy
I'll tell u what the problem is right here and now. Socket 478. 478 Prescott is a reborn 423 Willamette going through public testing.

Within a year (actually 6 months, IMO), we will have all forgotten the heat "issues" with prescott as it matures; Intel fans will be happy (again) and AMD fans won't know exactly "why" but they will still despise Intel.
rolleye.gif



And are we really 'ON' for 10Ghz next year? :Q
I woulda guessed late '06 at the (very) earliest.