Re: Powerbooks?
Originally posted by Frobozz
After all, PC makers put super hot chips in their laptops all the time. It's not insane in the PC world for 40+ watt chips in a laptop. The below is a 2001 article from Tom's Hardware on the heat issue:
http://www6.tomshardware.com/cpu/20010917/index.html
I saw no reference to mobile processors in the link you provided. Regardless, I wouldn't use the P4-M as a role-model for what a mobile chip should be. According to Intel, it typically dissipates 30 watts when running at full clockspeed...and at this power drain the laptops only last about 1 hr 15 min! You can get 2 hours out of the P4-M by cranking the clock speed way down (1.2 Ghz...slower real world performance than the PB G4). I would guess that a 1.2 Ghz 970 (19 watts) would not be able to get more than 2 hours of battery life unless IBM has added some power saving features which allow the core voltage (and hence clock speed) to fall even further below 1.1v (we have no indication that they have done this, though we don't know that they haven't). So yes, it is certainly possible to put a 970 in a PB, but do you really *want* the Powerbook to have the battery life of a P-4M laptop?
As for the "rumors" of lower power consumption for the 970 than IBM has stated, I haven't seen any, and if I did, I would put no faith in them. Even the "easy" rumors (e.g. has it entered production yet) are typically wrong, so it really would take blind faith to believe that they will report esoteric technical details accurately.
For a look at heat output on pre-production 970's (NOT the same specs/watts reported in recent rumors) and the G4e (oct 2002), take a look at this ArsTechnica article:
http://arstechnica.com/cpu/02q2/ppc970/ppc970-1.html
We've already disproven these figures using Motorola's and IBM's own documents. Plus it would be mathematically impossible for the PB G4 to get more than 1.5 hours of battery life out of a 55-61 watt-hour battery if the processor alone (G4 @ 1 Ghz) were consuming 30 watts. Simply put, the Ars Technica figures are wrong. I'm sure it was accidental, but they are wrong nonetheless.
Where I _do_ agree with the skepticism is WHEN. I don't think we'll see the 970 PB until Jan. '04, and for good reason... the PowerMacs need to take the throne for speed. Simple.
Yes, I agree with this. Eventually, the 970 will almost surely appear in a PB. But it is very unlikely to happen at the same time as it appears in the towers. At the very least, I think it will take a later version of the .13 micron 970, if not waiting for the .09 micron 970. It would be nice to be wrong about this, but I just don't see it happening...
I think it's safe to say that until June, we're all doing a great deal of speculation on (somewhat) outdated documents.
And this will somehow change after June? We will stop speculating on this site after that? ;-)
Most docs in publication right now do not match that of current rumors.
Yes, and most rumors (probably 90%) are false, whereas most published documents are not. That is the difference.
and user wattage reported in current docs, that I choose to believe them.
That is you choice, but in my experience you are just setting yourself up for disappointment!
But even at 1.8 Ghz and only released initially in the towers, I think there is plenty to be excited about. I actually think that the 970 will be more competitive than the SPEC marks indicate. I would guess that what the IBM rep said at MPF and they will be competitive with the Intel at "twice the clock speed" (i.e. a 1.8 Ghz 970 will compare favorably with a 3.6 Ghz P4 rather than a 2.8-3.0 Ghz P4, which is what the SPEC numbers would imply). The reason is that the 970 literally scores twice as high as the G4 at an equivalent Mhz: the 1 Ghz G4 scores 300 and 185 on SPECint and SPECfp respectively, and a 1 Ghz 970 scores over 550 and 600 on SPECint and SPECfp respectively. So according to SPEC, the 970 will easily be twice as fast as the G4 on average. Therefore, if the 970 is *not* twice as fast as the P4 at a given clock speed, then the implication is that the G4 is actually slower PER CLOCK CYCLE than the Pentium 4 - i.e., we have the Mhz myth in reverse!!! But nobody seriously questions that the G4 is more efficient than the P4 per clock cycle, it's just that it's not twice as efficient (so a 2.8 Ghz P4 is faster than a 1.4 Ghz G4, but it's not twice as fast).
So if we accept the notion that the G4 is at least as efficient as the P4 PER CLOCK CYCLE (and a myriad of real world benchmarks confirm that, on average, the G4 is actually faster than the P4 per clock cycle...it just doesn't clock nearly as high), then the one of two things must be true:
1) For whatever reason, SPEC does not provide an entirely accurate comparison BETWEEN the Pentium 4 and the PPC architecture. It's not off by an order magnitude, but for whatever reason it does seem to overstate the performance of the P4 by a non-trivial margin.
2) For whatever reason, SPEC does not provide an entirely accurate comparison between processors WITHIN the SAME processor architecture, i.e. it is inaccurate in comparing the PPC 970 and the MPC 7455.
Clearly, 1) seems far more reasonable than 2). I don't think SPEC is a terrible benchmark, but obviously either 1) or 2) must be true. Both cannot be false, because then we have a contradiction. And given my choice, I would strongly suspect that 1) is far more likely to be true than 2)...it would be bizarre to think that SPEC somehow is a more accurate benchmark when comparing an x86 processor to a PPC processor than comparing one PPC processor to another. So I suspect that the 970 will be pretty competitive even when Intel switches to the 90 nm process for the P4.
Of course this is just talking scalar integer/fp. In terms of vector processing and MP aware apps or generally multitasking, the 970 will smoke the P4 (assuming Apple releases at least one MP 970 config).