What's so special about duo-core?

[davidspalding]

Splitting off a thread into a new one...

Not to single out any user's comment, but:

... I purchased a T42 on 1/3 and had it shipped back without opening the packgage. I am waiting for the UXGA 1600x1200 spec so I can close the deal on a T60. I will not wait any longer. I asked the sales rep if he new when the duo core technology will be available and he pleaded the fifth. After waiting long enough I ordered the T42 to learn that I was misled.

I fear that my waiting is going to be much longer than I could ever imagine.

I'm not up to speed on this duo-core technology. What's so special about it that makes it a deal-maker feature? Does it save power, or double performance, or have a specific application like photo/3d graphics software, etc.? Feedback and discussion welcome.

[K. Eng]

This very good article should get you up to speed:

http://www.anandtech.com/mobile/showdoc.aspx?i=2663

In short, performance blows the doors off the old Pentium M, and battery life is generally better.

You should see moderate gains in single threaded applications, and huge gains in situations with multiple threads. I think this is the biggest leap in mobile processors since the original Pentium M.

[Troels]

Actually just saw a review of the Yonah, and was about to say that the Yonah wouldn't mean a good battery life at all, especially not equipped with some of ATIs latest offerings. Infact sucked for laptops unless you will make use of both cores.

http://www.tomshardware.com/2006/01/16/ ... _response/

especially http://www.tomshardware.com/2006/01/16/ ... age22.html

of course the dual core (or their crappy name Duo core - whatever) should perform better, it's just, why do Anandtech think that more laptops will get past 5 hours (in useless idle tests) when the dual core dissipates more heat (power) than even the Sonoma - which again dissipates more heat than the first gen Dothan ?

edit: it was in no way meant as a flame/insult, just a question.

[K. Eng]

Troels, the poor showing of the Napa platform machine in the THG test is probably due to the fact that the machine reviewed is a pre-production machine and not a shipping one.

I've looked at the Lenovo datasheets for the T60 and T60p. The T60 is rated at up to 6 hours battery life, which would tend to go against the conclusion of THG.

It is true that Yonah has a higher TDP (31 watts) than the Sonoma Dothan (27 watts), but this does not necessarily mean more heat dissipation. TDP is the maximum dissipation under realistic worst-case conditions, and is different than average power use. On average, I think Yonah probably dissipates less power than Dothan.

I've not confirmed this, but several people at Ars Technica have called Apple about the battery life on the MacBook Pro and are being quoted "up to 6 hours" as well.

[Troels]

K. Eng,
Yes, i think you are right. Goes to show that every part in a laptop must be carefully chosen.
The Yonah must dissipate less heat than the (2nd gen.) Dothan at idle i think, just sad that "Tom" jumps to conclusions that early then. Either that or the Yonah has SpeedStep down to 300 Mhz to achieve 6 hrs.

Looks promising

David Spalding,
Voted "other feature" because i think the lack of video ram was limiting especially the most normal T42/T43 with only 32 or 64 mb. Also speedier graphics is also welcome, the 9600 and X300 does look a little dated compared to what is available now.

[christopher_wolf]

Well, this is more on the theoretical side of it; but I was, at the time, thinking of a Beowulf Cluster and ran across a farily simple description of Amdahl's Law (Who just so happened to be working at IBM )

http://www.phy.duke.edu/~rgb/Beowulf/be ... ode21.html


To me, this would only make sense if, for every task to be done on the computer and noticed by the user as work, a Pentium M would have a certain threshold for how much can be done in a given time and still not be noticed by the user or termed as work; much like the idea behind Hyperthreading. If, indeed, a user had a low enough threshold for things like time-to-completion for a given task and the OS handled everything in a serial manner then it would appear as if Dual Core would have an advantage over Single Core. So, it ends up depending on the user. This also assumes that the user puts a very high value on things like speed of the program; some progressive tasks would then be the worst-case scenario.

Yet that means, say if the process had was forced through a serial procedure call; the fastest you could ever get it is the time through that serial portion. So on a 1.86GHz Yonah, you may get better performance, defined as high values of R and T, on 95% of all the tasks and still be the same or quite possibly slower than, say, a 2.0GHz Dothan on the remaining 5%. I think if you could consider web browsing/email/word processing/excel/powerpoint as high level tasks, then you would only get a tad more performance out of the current Yonahs. Unless you are doing very large amounts of work (going to the compute limits of excel), you may only realize a 25% gain in speed. Stuff like Word Processing will not benefit much, if at all, since that is limited by the user which can be assumed as near-serial (Try typing out several essays at the same time); further, the only advantage email and web browsing will gain is in terms of pure compute speed, then they get limited by the connection you are on and bandwidth becomes more important.

Going parallel would greatly help only a certain part of the user-base as much as would generally be thought. So the question just becomes how much the user doing non-intensive compute/business/research tasks would benefit from dual core. Assuming that OSes can multitask perfectly, and the code was near perfect when it came to multi-tasking (that is a far reach, I assure you), then you only gain by getting a larger increment,per chipset model release and generation, of application speed, again assuming memory doesn't limit it; those are quite a few assumptions, full OS support, full application support, etc.

So it really gives you the most tangible results if you use your system to the maximum and have tasks that could benefit from a dual core (Video, still limited by parts of the Video Card, Rendering, again limited by the Video Chipset) and already have the maximum values for every other resource. That is also why I would wait for the "Performance" models of the T60 and X60 come out; it makes little sense to me to pay that much for a dual core whilst the tasks I am thinking of already tax the system to the maximum when it comes to other compute resources.

[BillMorrow]

chris is right..
how much will all this technology benefit the regular, non-numbercrunching/scientific super-user..?
i think not enough to justify the overall cost..
for most..

but there are always those, like me, who like to mess with the high end just to see how it all hangs together and works..

[bert]

For a long time Intel was able to increase CPU performance by increasing the clock speed, but lately they have hit something of a cealing there. So now the approach is instead to increase performance by packing multiple CPUs into one component. It starts now with two. Soon we will see four, eight, sixteen, and so on until some kind of limit is reached again. So the Duo is the start of a new generation.

[davidspalding]

Add to number-crunching/scientific uses other niche specialties like Photoshop, 3D rendering apps, sound/video editing. Now, you might think that these are DESKTOP applications, but since my TP is my new primary system, I do use it for Photoshop and Premiere Elements on it. Sounds like, if Photoshop can use it (can it use the HT -- hyperthreading -- chips?), duo core indeed would be nice. Even if I just used it to do batch processing of 25+ files in Adobe Bridge while doing a crop and correction job in Photoshop (a feature in CS2, btw).

[K. Eng]

I can't remember where I read it, but I think Intel claimed that four general purpose cores were optimal, and after that, it would make more sense to add specific purpose cores.

I think the limit to the # of cores will be how much more transistors can shrink.

It starts now with two. Soon we will see four, eight, sixteen, and so on until some kind of limit is reached again. So the Duo is the start of a new generation.

[christopher_wolf]

Quite a few cores can be placed on a chip right now without much of a chage to the feature size (65nm in thte Yonah; 90nm in the Dothan). Certain things in the chip itself have to change, but you could divide it up into many smaller cores or a few, large cores; that depends mainly on the bus and other internal features of the chip.

[bert]

removed duplicated message