| Thursday, August 15, 2002 |
11:18 - The G4's Last Hurrah
http://www.denbeste.nu/cd_log_entries/2002/08/Anunbelievablekludge.shtml
|
(top)  |
...I think he's right. All the signs certainly point to this being the case. The G4 chips in the new Powermacs are 7450s, just like in the previous machines, and it's entirely possible that they've simply been overclocked and rebranded as 1.25GHz chips just for the purpose of this speed-bump.
But I don't buy the "embarrassment of engineering" angle. Depending on what one is trying to see this as, it can be an act of desperation-- or a fairly ingenious strategic interim solution, a very competitive product in a demanding market, ideally positioned for the new POWER4-derived IBM chips (the G5?) which seem to be targeted for consumer availability sometime early next year. It'll be a while in coming, but that just means we'll have to hold our breath for a while. And the current machines are far from useless.
The architecture is, after all, notable. Dual everything. Bridgeless direct PCI. Lots of L3 cache. And there's nothing inherently wrong with overclocking a chip; just ask any gamer. If they can do it and keep the chips stable, more power to 'em.
But this isn't an act of desperation, thrown together in a hasty three weeks of scrambling. This is an architecture that's been in planning for what must have been months. My guess is that as far back as January, or possibly earlier, Apple realized that the G4 was doomed as long as Motorola was so preoccupied with its business woes and seemingly uninterested in making desktop semiconductors. So they started coaxing IBM to develop the new chip as a contingency plan-- to develop the architecture, figure out the backwards-compatibility issue, make it desktop-capable, make it 64-bit Book E-compliant, try to get the power consumption down to a reasonable level, implement something to take the reins of Altivec-- and then, when the time was right, to build their new semiconductor foundry and announce the existence of the new chip. Finalize the design, get it ready to move, and meanwhile Apple would build an architecture that squeezes the last drops of juice out of their existing stock of effectively end-of-lifed G4s-- overclocking them when they run out of top-end chips that can legitimately push the speed barrier-- and put it into a machine that's designed for the new IBM chip rather than for the G4. Den Beste says it himself:
In fact, there's good reason to believe that these new machines are still going to be bottlenecked, because the processors share a single bus to the controller. PC duallies have separate buses for each CPU. Irrespective of how much L3 cache is connected to the controller, or how fast the RAM behind it runs, the data is choked on that FSB which runs 133 MHz for the 1.0GHz duallies, and 167 MHz for the 1.25 GHz duallies. The new "faster" bus merely keeps pace with the degree of choking; it doesn't relieve it. To relieve it, the G4 bus interface would have had to be redesigned, but that would have required Moto to roll the chip design, and it's clear they are not going to be doing that. If Apple was expecting a new G4 with a new FSB architecture, they would never have created these monsters.
Yes, exactly. These machines are architected for the G5, not for the G4. They'll hum along nicely with the G4 for a while, but they're overengineered in their current configuration. Put in G5s, with new memory buses, and they'll come into their own. (Likewise with the Xserve, which many are now considering to be an interim design, anticipatory of the 64-bit G5 as well.) Or that's the hope, anyway.
Some speculation (between myself and Chris, mostly) claims that the 133- and 167-MHz front-side bus speeds in the new Power Macs seem lackluster in comparison to the 266 and 333 MHz of PC motherboards-- but that this is simply because by PC-style accounting, those latter numbers are really 133 and 166, but doubled because of the effective double speed of DDR RAM, which Apple has only now received. The P4-based systems can get away with claiming such numbers because their CPUs have direct access to the RAM at such speeds, whereas Apple's don't-- yet. The front-side bus speed itself isn't crippled compared to that of PC motherboards. It's crippled by the fact that the FSB is designed for a later chip which will take full advantage of this architecture.
Apple has pulled this slow-ramp CPU trick before, incidentally. They've done things like releasing 100MHz machines when the chips could really do 120MHz, or underclocking the RAM or the FSB, often for the purposes of selling pro-vs-consumer targeted products-- but more often than not, to allow themselves a smooth ramp target. If the previous generation of machines were 60MHz, and they now had chips that could do 120, they would first release them at 100-- still a significant speed improvement, and a sales driver, but it also gave them the ability to do another speed bump three months later-- another sales driver, and it kept the speed ramp smooth. Overclockers wishing to void their warranties could often soup up their machines, and when the clone market was in existence companies like Power Computing could blow away Apple's own machines with 150MHz boxes against Apple's machines running at 120MHz and so on, but Apple kept up the smooth and behind-the-curve pace-- so that much like corporate financial officers deferring profits in a successful quarter so the company doesn't totally blow away its numbers and then predispose everybody to the same kind of success the next quarter, setting themselves up for a fall if the results are merely "normal", Apple could always ensure a predictable, on-schedule speed increase. And with a couple of notable exceptions (like the infamous downclocking of the G4s right after they were announced), they've been able to hold to that schedule throughout most of their history.
So now we're beginning to cross a desert, one in which we're unlikely to see any truly new G4 models (there's a possibility of a 7470 appearing later this year, but it's not looking good-- and the 7500, with its RapidIO architecture and longer faster-clockable pipeline, which was the subject of much optimism earlier this year, seems to be a lost dream now). I suspect that there will be room in the schedule for at least one more speed-bump between now and whenever the IBM G5s appear; rumors tell of the current 7450 G4s running at 1.4 or even 1.6GHz, and even if that's a matter of overclocking, it may well be right in with Apple's plans to do just that. It's a gamble, but if timed correctly, we could reach the other side of the sands with water left in our canteens.
As to whether IBM's chips are in fact on the way, we have no facts but a lot of speculation-- though, granted, that speculation is extremely compelling. For instance: is it a coincidence that IBM's chip will not support any 68K emulation code, and that Apple has announced that early next year new Macs will not be able to boot into OS9 at all? I doubt it. (Then again, as Kurt Revis points out, 68K emulation is all software-- it can work just as well on a new CPU as on the old.) The evidence seems to be in pretty strong supply for the idea of Apple being wholly committed to the IBM chips, and engineering around the G4 now rather than specifically for it.
The new Power Macs are very competitive machines, and they're only going to get better. That is, if our speculations pay off. Steve Jobs is rolling the dice in a major way here, but the situation isn't as dire as it can be made out to be. I still would like to see one of these machines in action; it's bound to be a significant leap past the previous generation, and it isn't because of marketing smoke-and-mirrors. There's a plan in place here, and there's good engineering going on. Considering how many rabbits Jobs has pulled out of his butt in the last few years, and how few failures, I'm inclined to put some faith in the man's ability to marshal his resources and strategize with the best of 'em.
|
|
