23,
I hate the current naming as well, it is grotesquely unclear what the Hell these processors are anymore. But, if you create a totally different processor line (you couldn't call it Pentium 4, or for that matter Pentium 5, which is probably the name of the Merom/Conroe) I think it would go from bad to worse. I doubt computer manufacturers would be too happy about it as well. There may be a solution for it, but it escapes me. Without more information, I don't know what makes the most sense for Intel right now. From a purely technological perspective, the Yonah seems a clear solution, but it would create confusion.
Anandtech's article typically misses the mark. I can't help but get a little irked when people mention that additional pipeline stages mean higher clock speeds and lower IPC. In fact, the additional stages for the Athlon 64 were not to increase clock speed, but were to improve IPC. In a general sense, yes, but when they use this generality to make a specific claim against a specific processor, it is poor reporting. Without knowing, one should reserve judgement.
Also, rather stupidly he seems to discount any IPC improvements in the Merom. He minimizes them at best, and tends to believe the benefit will mostly be from higher clock speeds. Why? Because the Merom is four issue and the Yonah three issue? Hmmmm, that shouldn't help IPC, huh? Oh wait, it might just barely offset the longer pipeline, which of course is always a big loser in IPC.
So, I think his understanding of this technology is remedial at best, and his conclusions not firmly planted. Of course, it could be that the technology goes the way he says, but I think any firm conclusions will have to wait and his view seems a little divergent from what we currently know about the processor. I believe it will show improved IPC. The reason no one added a fourth pipeline is not because of transistors so much as the fact as they could not figure out a way to keep it busy. If Intel has, it could translate into substantial performance improvements. Also, having a fourth integer pipeline (and with it the safe assumption it will be used) implies strongly that the second and third pipeline will show greater utilization since Intel has figured out a way to better schedule instructions. Currently, scheduling is not strong enough to show benefit from a fourth ALU, so I would think it had to be improved and that improvement should fall to the other pipelines too.
Another reason I think the fourth ALU is very important is related to Intel's focus on lower power/performance. Obviously, this should negatively impact this unless it sees some pretty decent use. If it isn't working much at all, they wouldn't keep it and suffer the power use for something so nearly useless.
One ugly thought is that the fourth pipe could be useful mainly in Hyperthreading. That would be a huge disappointment. With multiple cores becoming more and more prevalent, I would think Hyperthreading would become less a focus for Intel, but it is hard to be sure.
Only time will tell.
[Posted by: TA152H | Date: 12/23/05 10:53:58 AM]