Submitted by “Apple have shown some absolutely stellar benchmarks with the Core solo and Core Duo processors outperforming their G4 based predecessors by quite a considerable margin in most areas, heavily vectorised applications are seemingly the only exception. The story is not quite so clear with the G5 based iMac but there is a difference even there. These benchmarks can be considered no doubt as justification for the decision to switch to x86 processors. However, in typical Apple style there is some slight of hand, in particular as regards timing of the switch. Making the transition 6 months early is no mistake, it was absolutely critical to get those benchmarks.”
While the author does make some good points, there are more “technical” reasons for the transition than the ones he outlines. Look no further than the now dual booting intel macs! Also look at the improved release times and faster ability to catch up with demand. I’m also pleasantly surprised with the ability to interchage parts now (like upgrading the mac mini to a dual core 2.16). Soon you’ll also be able to boot windows inside OS X using tools like VM Ware and Virtual PC at NATIVE speed. Darwine already works (although it’s in its infancy) and you can run software today on your mac without even booting up the windows OS. Given enough time the mac will now become a jack of all trades PC that can run unix mac and windows software without even rebooting, and that to me is a technical feat worthy of switching that no benchmark can convince me is a bad thing.
Don’t forget that one shouldn’t be restricted to buying special Mac graphics cards anymore.
The other point is that IBM was unable to provide a mobile G5 processor to Apple for the Powerbook on schedule. Continual delays caused Apple to look elsewhere. Intel was able to provide vendors such as Apple, HP and Dell the Yonah 32-bit dual core mobile processor built on 65nm design. Intel will also release Merom a 64-bit dual core mobile processor later this year also built on 65nm design.
Edited 2006-03-21 16:41
The author is missing one key point. The PPC architecture has been plagued by empty promises for years. Weren’t we supposed to see 3ghz chips by now? Oh, and what about the Motorola 7448? They are *just* now starting to materialise. The cell is a special purpose CPU which is not fit for general computing. I for one will not shed a tear over the passing of PPC. The PPC will continue to live on in the embedded space, but it just isn’t a compelling choice for a desktop.
Weren’t we supposed to see 3ghz chips by now?
You probably meant “a few years ago”…
The PPC will continue to live on in the embedded space, but it just isn’t a compelling choice for a desktop.
The problem with PPC is that no one bothered to produce a CPU that is optimized for desktop/mobile use. The 8641D, with its SOC design, would have been perfect for mobile use (memory controller, ethernet, PCI-E etc. are all integrated, which will produce substantial power savings) and solve the memory bandwidth problem once and for all, had they delivered the product in a more timely fashion…
Edited 2006-03-21 08:15
It seems like it should read “sleight of hand” instead of “slight of hand” as was originally posted.
Don’t forget that one shouldn’t be restricted to buying special Mac graphics cards anymore.
Apple could have fixed that easily long ago, all you need is a very limited x86 emulator for initialising the card. Other PPC companies do this.
It seems like it should read “sleight of hand” instead of “slight of hand” as was originally posted.
Fixed, Thanks.
The author is missing…
Yes there are other advantages but that’s not what the article was about…
The local supermarket sells a 3GHz dual core PC for €999 this week, not next year… wondering when Apple’s new Powermac is showing up 🙂
…
But I shouldn’t be to worried about the performance my 1GB Athlon 64 3500+ runs OSX very nicely (much better then the ppc iBook I tried earlier, even the eye candy looks cool instead of feeling slow)
The local supermarket sells a 3GHz dual core PC for €999 this week, not next year…
Remember the MegaHertz myth!
😉
This article is all complete fantasy. Some choice bits:
They also use the newer G5 bridge chip which fixes the latency problems the original G5 bridge had.
Uh, no. The 970MP in the dual-core PowerMacs use the same bridge chip, and its latency is still poor.
Well, technically, the memory controller latency went down, but the move to DDR2 wiped out any improvement. The resultant latency is still about 30% higher than Intel’s latency, and 150% higher than AMD’s.
The 970GX will be faster per clock then the 970FX so it might actually lead in the single threaded benchmarks.
The 970GX is just the single-core 970MP. It isn’t any faster in single-threaded benchmarks. The only difference is the larger cache (1MB versus 512KB), but even that is a mixed advantage because the latency increased as well (13 clocks versus 11 clocks).
The 8641D has dual cores and dual on-chip memory controllers, they should pretty much blow the existing G4s out of the water.
The 8641D is pretty much the same as the G4, with the only difference being the integrated memory controller. The IMC is going to help, but there is no getting around the fact that the G4 architecture is seriously out of date.
There’s also a completely new Power processor due from PA-Semi which looks like it’ll not only match the Intel laptop parts on performance but use less power doing so.
Did he check the estimated SPEC results? It’s a 2007 part, and its expected to have a SPECint of 1100. Based on the initial benchmarks of the Merom architecture, that’s equivalent to a 1.2 GHz Merom chip. It’s SPECfp looks very good at 2200, but in real world code, the crappy integer performance is going to be a yoke. The 970 is already unbalanced at ~1500 SPECint at ~2000 SPECfp, the PA Semi chip is going to be even worse.
the POWER6 derived P6L
There has been no indication that a P6L will even exist. Without Apple, I have my doubts about such a chip ever being built. Even if it was built, POWER6 isn’t exactly Apple’s ideal chip. It could very well be an awesome chip, but look at the design point: 4+ GHz, tens of megabytes of cache, quad-channel DDR2 memory controller. The indication so far have been that IBM is tuning quite finely around that design point. The core isn’t as aggressively out-of-order as the POWER4/5, as it counts on the large cache and SMT to make up for potential core stalls. You can’t afford such a cache on a workstation, and SMT does nothing for single-threaded performance.
Apple could if they so wished pick and chose processors from both.
Not going to happen. Apple’s burned its bridges with Freescale and IBM, and it can’t afford to do that with Intel.
So let’s sum up Apple’s choices in the PowerPC realm: a low performance, low-power chip intended for embedded applications, a high FP performance low-power chip intended for signal processing, and a high-performance, high-cost, high-power chip intended for servers. All very nice chips, and all very suited to their respective tasks, but none competitors to Conroe in the desktop/laptop realm.
I’ve said this before: Apple needs a desktop chip. Not an embedded chip tricked out for desktop use, or a server chip neutered for desktop use, but an actual desktop chip optimized for desktop tasks. Nobody in PowerPC-land is willing to make one for Apple, and that’s why Apple switched.
The 8641D is pretty much the same as the G4, with the only difference being the integrated memory controller. The IMC is going to help, but there is no getting around the fact that the G4 architecture is seriously out of date.
Not quite… 8641D uses SOC design (I guess it took Freescale a long time for Freescale to realize that the solution used in their low end chips can also be applied at the high end), which not only has integrated memory controllers, but also integrated ethernet, PCI-E etc. This will not only improve performance, it will also lower the power consumption dramatically.
8641D also uses the new e600 core (different from the core used in 7447)
Integrating ethernet and PCI-E will improve performance significantly for the type of tasks the 8641D is aimed at, but networking and I/O performance isn’t nearly important enough on the desktop for that to be noticeable.
As for the E600 core, it is almost identical to the G4e core, aside from the bus interface. The buffers are all the same size, the OOO window is the same size, the instruction schedulers are the same size, the branch prediction tables are the same size, the execution units are the same configuration, the register files are the same size, etc. The biggest difference in the execution core I can see is that the FP scheduler has been enlarged to 2 entries, allowing (limited!) OOO execution in the FPU.
In most of these measures (except BHT size), the original Pentium Pro is a more sophisticated design than the E600. While the 8641D might make an awesome network router processor, it’s not going to cut it as a Merom competitor…
I think we may see IBM going into obscurity. Like a cash-strapped Roman Empire with a lot of territory. They were unable to build a good desktop processor. It’s sad, because I’ve been a fan of the PPC processor.