eSnow said:
Well, yes and no.
The PentiumM lacks the SMT-cabilities of the P4 as well as the cache coherency protocol needed for dual- or quad-CPU boards. Furthermore, it has a much shallower pipeline than the P4 which is good on the one hand since it makes it more performant per Mhz, but on the other hand limits its clock frequency. While the P-M can match a P4 in some tasks, this is not true for all - there are some parallels to the comparison between 7447A and 970.
For desktop use, SMT is currently poorly implemented, if at all. The reason that most home users by anything like an Athlon FX or a P4 EE is to game on, and the performance just isn't showing up the way that the chipmakers have hyped it. This is probably due to coding not handling the extensions and taking advantage of the chip, but that's still not going to do much good when a Dothan revision a year from now starts to beat the P4 at most home-user tasks.
Professionals with big cases and no need to avoid noise can still use Xeon and Athlon FXs, but I don't think that there's going to be a huge market for them once anyone with half a brain starts realizing what they can do for home design with a decent Pentium-M design.
Interestingly, Intel's roadmap shows some indication that this is in the works, and that it will happen by 2006.
ffakr said:
Dothan will be faster, and it will have a fairly massive 2MB L2 cache.. but there are limits where L2 becomes a rapidly diminishing return in terms of performance boost. It's likely that, to some extent, the large L2 is probably tacked on as filler for the relatively small PM core.. so it's fabbed at a manageable size. No one wants to try and hook up 500 or 700 leads to a 20mm^2 chip. The 970fx is a much more complex (wide) processor than the PM and it only comes in at 65mm^2 with 512K L2. It's pretty small.
Well... It's the old P6-generation core with some new stuff bolted on. What did you expect?
😉
Honestly, if you've got the processor budget to add cache on the chip without really screwing things on cost, I don't see a reason
not to do it. Less fetching means better performance, in general. Yes, it hits a wall after a while, and this is one reason the P4EE just isn't as much of a killer as Intel would like.
Also, keep in mind that though Intel has been saying for years that they are serious about controlling waste heat, they've continued to develop cpus that follow the design mantra of the P4 to an ever greater extent. The Prescott is really like a P4 plus. It's less efficient with longer pipes. Once again, Intel decided to trade more top end speed for less efficiency. Tejas is next in line and it's expected to clock even FASTER.
They can't just drop their old design philosophy, not after it's been entrenched the way it has by their massively successful focus on the consumer tendency to think that numbers are good, and big numbers are better. It's going to take time, money, and effort to correct that perception before they can attempt the move to the Pentium-M descendants on the desktop around, as noted above, 2006.
The variant in question is a derivative of Jonah, the Dothan's successor, with 2MB L2, a 90nm process, and dual-cores. No figures released on speculated clockspeed. In the second half of 2006, Intel is slated to release Merom, and will be both mobile and desktop chip (in the form of Conroe), with multiple cores, 65nm process, 4MB of cache, and a target heat of 45w. Dothan is roughly on schedule, but Jonah and Merom are falling off schedule.
Power5 from IBM was salted for a June release. It's out now. The Power5+ is expected in 2005, and slated to clock up to 3.0ghz
in the server version. To extrapolate from the Power4 to 970 clock: Power4 1.3ghz to 970 2.0ghz (54% increase), so Power5 3.0ghz to 980 4.6ghz. That's assuming that they're not being concurrently designed, in which case we might even be able to expect an even greater leap in performance because the lower end chip isn't a bastardization so much as a carefully created alternative.
All processor information from
endian.
I'm just guessing that the Pentium 4 might step back from the podium as Intel's flagship general purpose 32 bit processor and become more of a niche player, with the desktop and notebook implementations of the cooler and less power hungry Dothan M becoming much more common. That's not even mentioning the competition that Intel's Pentium 4 and Pentium 4 Xeon is facing from AMD in the high-end desktop, workstation and entry-server market, courtesy of the Athlon 64 and Opteron.
Am I the only one who noticed that, this past quarter,
AMD outsold Intel?
Like I said before, with competition from AMD, IBM, a (hopefully) re-focused Motorola, plus improved Pentium M chips from Intel themselves, the CPU market is a much more diverse and interesting place than when it was just a singular focus on Pentium 4s hot enough to fry an egg 🙂
Amen.
nek said:
I think that there is a real possibility that a Power5 derivative was made concurrently, and therefore will be ready for introduction at WWDC. Moving the Powermac and Xserve to this new chip would allow the 970fx to be used in the iMac. The 970fx can likely reach 2.6 or 2.8GHz, so that would provide enough room to upgrade the the iMac and eMac a couple times over the next year or so.
I'm in agreement on everything but one factor... Let the iMac die! Do something new, Apple. Come on, Ives and Jobs. I know you have to have something up your collective sleeves that you could really wow the punters with, if only you'd do it.
Surprise me!
And if Freescale can deliver on its plans for the e600 and e700 in the near future, then Apple won't need to try to put the 970fx in the PowerBook.
Once again, amen.
