Intel just don't fail to disappppoint me

[Leper Messiah]

Originally posted by: DAPUNISHER

Originally posted by: dmens
That's because some of the engineers who quit feel the need to toss blame around and run their mouths. Seriously, allegations regarding the "marketing driven P4" tend to be wildly exaggerated or outright false. Last time I checked, the arch and design leads of the old DPG group didn't report to the marketing VP, and the people who were actually involved with the critical design decisions of the P4 projects are going to be talking details any time soon.

FYI, if intel DPG were really run by the marketers, the whole damn group would have collapsed before willamette even taped out. yeesh.

I wish Wingz and pm would post more too, I frequently learn something from reading what you guys contribute, Thank you.

For real.

 

[Lonyo]

Originally posted by: Markfw900

The most basic objective for a company to market a product is to identify the basic requirements of customers. What do server computers need? Even a 15 years old can answer. Performance, Power consumption, Heat dissipation

A server needs to be first and foremost powered my a reliable CPU.
Reliability is the name of the game when talking about servers.

Reliabilty number 1 ? well, yes, but we have a serious power problem in one of our data ceters, and running all of our Wintel boxes at 1/2 their current consumption would fix that. Are you trying to say Opterons aren;t reliable ? Don't even go there.

Yup, power in huge server rooms gets more and more of an issue.
It's not always just CPU power consumption though. The higher the CPU power consumption, the higher electricity costs, but also you put out more heat, more heat means you need moor cooling, which requires even more electricity.
Hence the move to power per watt on Intel's part, where AMD are already ahead of the game, and the move to using (IIRC) Pentium-M's in blade servers.
AMD and Intel are equally reliable pretty much, unless you're living in the past.

 

[IntelUser2000]

The facts:
1. Process technology isn't the reason that Prescott runs so hot, its the bad design of the CPU, don't believe me?? Look at Dothan.
2. Dempsey, the successor to Paxville, along with Bensley platform will increase performance by 80% as tested in initial benchmarks: http://www.tecchannel.de/server/hardware/432957/
3. The design group for Presler is said to be mostly moved to next generation desktop chips, like Conroe, so the engineers that are supposed to be doing power tweaks and performance tweaks are simply not there. Hence, almost no improvement with power consumption for Presler.

 

[dexvx]

So AnnihilatorX what kind of hardware have you been a part of designing?

...

But seriously, NetBurst was looking good when Northwood/Gallatin was the processor. They dominated AthlonXP's at the time. It all came downhill due to Prescott.

 

[dmens]

What's this about no improvement in power consumption with Cedarmill/Presler? It might not be the cubic power reduction with process size like they talk about in school, but that's all complete gibberish anways lol.

 

[IntelUser2000]

What's this about no improvement in power consumption with Cedarmill/Presler? It might not be the cubic power reduction with process size like they talk about in school, but that's all complete gibberish anways lol.

Cubic reduction in power consumption??? Where did you get that from??? There has never been cubic reduction.

Cubic reduction only comes when
-voltage reduces
-frequency reduces

And if you do read the initial Presler/Presler EE results, there are almost no benefits to power consumption for those 65nm processors, and I am explaining its not Intel's process tech fault.

 

[dmens]

That's because they don't underclock and undersupply presler as much as smithfield. For a better comparison, look at cedarmill vs prescott. There was power work done on cedarmill... not bad for a quick shrink.

 

[IntelUser2000]

Q]That's because they don't underclock and undersupply presler as much as smithfield. For a better comparison, look at cedarmill vs prescott. There was power work done on cedarmill... not bad for a quick shrink. [/quote]

Not cubic reduction as you say.

Why does Smithfield come into this equation?? Looking at voltages, Smithfield isn't undervolted from Prescott and although Smithfield does clock lower than Prescott, they are at the same process.

If you look back at Northwood days, 2GHz P4 Willamette were 75W while 2GHz Northwood was 55% and there was lower power versions of Northwood with like 45W and Intel were able to make Pentium 4-M's which consumed only 30W.


Cedarmill, with same L2 cache size, only reduces power consumption by 20% or so, while Northwood had double the cache size of Willamette.


Though, its true that Northwood had more attention from engineers unlike Cedarmill.

 

[dmens]

LOL I didn't say... when I was in school a lot of people said that. Northwood was not a shrink, it was a pretty significant redesign with a large design team dedicated to it, unlike cedarmill. Wait smithfield isn't on a lower voltage? They said they'd do that when I was working on it, haha.

 

[IntelUser2000]

Wait smithfield isn't on a lower voltage? They said they'd do that when I was working on it, haha.

Prescott 2M 3.2GHz: http://processorfinder.intel.com/script...3&PkgType=ALL&SysBusSpd=ALL&CorSpd=ALL

Smithfield 3.2GHz: http://processorfinder.intel.com/script...9&PkgType=ALL&SysBusSpd=ALL&CorSpd=ALL

They are being vague on actual voltages as they show only one range but I can even say Smithfield's voltages are EVEN higher than Prescott 2M, to the contrary.

 

[coldpower27]

Originally posted by: IntelUser2000
Q]That's because they don't underclock and undersupply presler as much as smithfield. For a better comparison, look at cedarmill vs prescott. There was power work done on cedarmill... not bad for a quick shrink.

Not cubic reduction as you say.

Why does Smithfield come into this equation?? Looking at voltages, Smithfield isn't undervolted from Prescott and although Smithfield does clock lower than Prescott, they are at the same process.

If you look back at Northwood days, 2GHz P4 Willamette were 75W while 2GHz Northwood was 55% and there was lower power versions of Northwood with like 45W and Intel were able to make Pentium 4-M's which consumed only 30W.


Cedarmill, with same L2 cache size, only reduces power consumption by 20% or so, while Northwood had double the cache size of Willamette.


Though, its true that Northwood had more attention from engineers unlike Cedarmill.[/quote]

You also got to keep in mind that Northwood was a second generation 0.13 micron product, while Cedar Mill is a first generation 65nm product, it's purpose is to test out 65nm technology. This may also explain why the transistor density of Yonah isn't too high for a 65nm product, when compared to the 90nm Dothan Processor. They aren't the optimal 65nm designs, those come later with the Merom/Conroe/Woodcrest line.

More emphasis is probably being placed on Intel NGMA products rather then Pentium M or NetBurst.