Dan, things may be rotten to the core (no pun intended) at 1 AMD Place if this guy is right:
Bad yields?
I have a different theory, look at it in this manner and you'll see that it would explain much....perhaps AMD's Palomino/Morgan are yielding badly and AMD was forced to increase the core voltage to get decent numbers of Morgan based Duron's. That might also explain the later then expected introduction of the Palomino core in the mainstream Athlon. And it would also explain why the Palomino based AthlonMP is at only 1.2GHz while the Thunderbird based Athlon is at 1.4GHz and overclocking beyond even that.
If this theory is correct it doesn't bode well for AMD's ability to stay competitive against a Pentium 4 processor that will be undergoing it's own transformation to improve performance and has also shown itself to be able to ramp in clockspeed amazingly well.
Who is Morgan?
Date: 8/24-2001
Category: Editorial
Author: Chris
Article Overview
There have been many people wondering why the Palomino is so late to take over from the Thunderbird core, and many wonder why the Palomino based AthlonMP is stuck at 1.2GHz. This article theorizes why that is, and in the process discovers some interesting facts about AMD's Morgan core that shows reveals perhaps the Morgan and Palomino aren't all that AMD would like you to believe...
Intro
There was much discussion surrounding the initial release of AMD's Palomino core, the Palomino core was expected to boost performance over the Thunderbird and reduce power consumption. The Palomino core has since seen release in AMD's Athlon4 processor for the Mobile market, and the AthlonMP for the server desktop market. Surprisingly however AMD has not yet allowed the Palomino to take over the place of the Thunderbird as the core utilized in their mainstream high end desktop Athlon processor.
I'll briefly cover what the Palomino is and how it's different from it's predeccessor the Thunderbird.
The Palomino boosts die size from 120 mm^2 in the Thunderbird to 128 mm^2 in the Palomino. It also boosts transistor count from 37 million to 37.5 million. There are only four real "enhancements" that the Palomino core offers over its predecessor outside of the reduced power reduction. Three of these features are performance enhancing and the other one is a mobile related feature called PowerNow!. The first advantage the Palomino offers is an increase in the number of L1 translation look aside buffer (TLB) entries. The second being a hardware data prefetch mechanism that works alongside its cache to boost performance. The third feature being that the Palomino adds 52 new instructions to add full support for Intel's SSE technology. The last feature of the Athlon 4 is an on-die thermal diode which measures the internal core temperature, much like what Intel's Pentium III has. The PowerNow! as mentioned above is also, features though it's only utilized in AMD's Mobile processors. AMD's PowerNow! implementation allows for a total of 32 speed/voltage steps between 500MHz at 1.2V and the maximum clockspeed and voltage of the processor. This can significantly reduce power consumption and significantly increase battery life. Why run at 1GHz when you can run at 500MHz with no performance loss? When you need the extra performance it will automatically compensate and increase clockspeed.
Morgan & Palomino
Why the explanation of the Palomino in an article about the Morgan core?
Well simple, the Morgan core is essentially the Palomino core. Just as the Palomino is expected to take over from the thunderbird core in the Athlon. The Morgan core will take over from the Spitfire core in the Duron. The Spitfire was a Thunderbird with 3/4 of it's cache cut off, and that's is exactly what the Morgan is to the Palomino.
AMD's Morgan core has been released in the 1GHz Duron. The Morgan core promised improved performance and reduced power consumption. In the Palomino at least both of these turned out to be true with the Palomino offering a small but definite performance advantage over the Thunderbird, and also reducing power consumption.
The Morgan core we expected the same from, and indeed it does boost performance... but AMD has also quietly revealed the Morgan core does not reduce power consumption in any way over the Spitfire. In fact not only does it not reduce power consumption... it increases it!
Home Next Page
Quick jump -------------------- Page 1 Page 2
Copyright © 2000-2001 TweakMax. All rights reserved. Privacy, Disclaimer and Copyright statements.
Page 2/2 Power consumtion
Datasheets for the Morgan/Duron can be found here. You can compare them against this Spitfire/Duron datasheet.
Processor Clock Speed Vcc Power Power Maximum
Spitfire/Duron 900 1.6 35.4W 39.5W
Spitfire/Duron 950 1.6 37.2W 41.5W
Morgan/Duron 900 1.75 38.3W 42.7W
Morgan/Duron 950 1.75 39.8W 44.4W
Morgan/Duron 1000 1.75 41.3W 46.1W
Thunderbird/Athlon 900 1.75 44W 50W
Thunderbird/Athlon 950 1.75 47W 52W
Thunderbird/Athlon 1000 1.75 49W 54W
Thunderbird/Athlon 1200 1.75 59W 66W
Palomino/Athlon MP 1000 1.75 41.3W 46.1W
Palomino/Athlon MP 1200 1.75 49.1W 54.7W
The Palomino/Thunderbird follows the expected pattern with the Palomino consuming roughly as much power as a Thunderbird at 200MHz lower clock. We would expect the Morgan/Spitfire to follow a similar pattern... instead we discover the rather shocking fact that the Morgan actually consumes MORE power then the Spitfire due to the increased core voltage.
The Morgan has the exact same power characteristics as a Palomino. This is definitely NOT right. The Morgan should consume less power then the Palomino due to the smaller L2 cache size. Previously Thunderbird and Spitfires cores were built on entirely different dies. This being different from Intel, in that the Celeron and Pentium 3 processors are built on the same die and are in actuality the same processor... except the Celeron is composed of Pentium 3 cores in which half of the L2 cache is disabled. AMD's method increases their costs slightly, but it boost performance as the Spitfire gained all the advantages the Thunderbird offered while the Celeron due to it actually having half it's cache disabled also effectively lost half the set associatively in it's L2 cache.
This situation was expected to continue with the Morgan and Palomino... indeed we had no reason to think otherwise until now. Until you look at the power consumption figures, why would the Morgan have equal power consumption to the Palomino?
Increased Voltage
The increased core voltage from the 1.6V in the Spitfire to 1.75 volts in the Morgan could account for some of it. But the Palomino also runs at 1.75V, and given it has 3 times the L2 cache it should therefore consume more power at the same core voltage. But it doesn't, it has the exact same typical and maximum thermal characteristics as the Morgan at equal voltage and clockspeed.
Going by the power characteristics I theorize that AMD has adopted Intel's method and is now making the Morgan core from failed Palomino's and disabling part of the cache. This could weaken the Morgan core more then AMD would like it to be known as by disabling part of the cache they will also be influencing other aspects of the cache design and cutting them off also.
Another question to ask yourself, why did AMD increase the core voltage from 1.6V on the previous Spitfire based Duron to 1.75V for the Morgan based Duron? The feature set shouldn't require an increased core voltage... indeed the Palomino/Morgan was designed to reduce power consumption not increase it! The most likely answer therefore would be they raised it to ensure that all Morgan cores yield at clockspeeds of at least 1GHz so ther are no wasted processors that aren't good enough to be sold.
Bad yields?
I have a different theory, look at it in this manner and you'll see that it would explain much....perhaps AMD's Palomino/Morgan are yielding badly and AMD was forced to increase the core voltage to get decent numbers of Morgan based Duron's. That might also explain the later then expected introduction of the Palomino core in the mainstream Athlon. And it would also explain why the Palomino based AthlonMP is at only 1.2GHz while the Thunderbird based Athlon is at 1.4GHz and overclocking beyond even that.
If this theory is correct it doesn't bode well for AMD's ability to stay competitive against a Pentium 4 processor that will be undergoing it's own transformation to improve performance and has also shown itself to be able to ramp in clockspeed amazingly well.
/Chris@tweakmax.com
Discuss in the Forum | See related Articles
Back Home
Quick jump -------------------- Page 1 Page 2
Don't know who this guy is, or who taught him how to write, but, what he says can explain a helluva lot.
tweakmax.com |
