Aside from the droves of Apple and AMD fanboys whom have been commiting suicide in their millions which I’m finding most amusing, here are just a few of my thoughts on this “little bombshell”. The original News.com.com.com broke on the 3rd of June:
Apple has used IBM’s PowerPC processors since 1994, but will begin a phased transition to Intel’s chips, sources familiar with the situation said. Apple plans to move lower-end computers such as the Mac Mini to Intel chips in mid-2006 and higher-end models such as the Power Mac in mid-2007, sources said.
… which has since been confirmed at Apple’s WWDC 2005, on the 6th of June:
At its Worldwide Developer Conference today, Apple® announced plans to deliver models of its Macintosh® computers using Intel® microprocessors by this time next year, and to transition all of its Macs to using Intel microprocessors by the end of 2007. Apple previewed a version of its critically acclaimed operating system, Mac OS® X Tiger, running on an Intel-based Mac® to the over 3,800 developers attending CEO Steve Jobs’ keynote address. Apple also announced the availability of a Developer Transition Kit, consisting of an Intel-based Mac development system along with preview versions of Apple’s software, which will allow developers to prepare versions of their applications which will run on both PowerPC and Intel-based Macs.
As I myself scribbled on phpBB’s General Discussion forums regarding the initial news from CNET News.com:
Now the timeframe would mean Apple will not be using the power hungry P4 derived Netburst line, but rather a “beyond Penitum-M” line (at least Yonah) which will have a very high IPC and be at least dual-cored.
To elaborate a bit more, current G4’s used in iBooks, PowerBooks and MacMini are 32bit CPUs which consume about 30W when clocked at 1GHz. The desktop line are powered with G5’s are now 64bit and consume up to 80W when clocked at 2.5GHz. The former don’t exactly have what one would call stellar performance and the latter are far too power hungry to use in anything other than desktop applications, though do have performance parity with mid-end dual-processor Intel Xeon or AMD Operton workstations.
Now Intel’s CPUs are currently split into the mobile 32bit “Pentium-M” line and the desktop 64bit “Netburst”, P4/Xeon line. The Pentium-M “Dothan” CPUs have short pipelines of around 15 stages and use just 27W peak when clocked at 2.13GHz, with low-voltage variants using 10W and 5W when clocked at 1.5GHz and 1.2GHz respectively. The “Netburst” based CPUs, the P4 “Prescott” and Xeon “Nocona”, however do manage higher clockspeeds due to their much longer pipeline which consists of 31 stages. However, to get to a clockspeed of 3.8GHz it needs to guzzle a huge 115W. Intel also now the dual-cored “Smithfield” CPU to add this, which is a “Netburst” one and when clocked at 3.2GHz consumes a frightening 130W.
Whilst Steve Jobs did show a Pentium4 running OSX at WWDC 2005, the fact the transition is “bottom-up” is why I’m 99.999% certain it will be Pentium-M derivatives that will see the light of day as Apple moved from PowerPC to Intel. The reasons are the known performance of overclocked “Dothan” Pentiun-Ms and the forthcoming next generation “Yonah” Pentium-M which will see Intel migrate from the curent 90nm process to the finer 65nm one.
First though, are some laws of Physics which even Apple have to respect. Say a CPU was clocked at 1GHz and consumed 10W, to double the clockspeed, it’s power consumption does not double but in fact rises to the square of the increase. So at 2GHz it will consume 40W and at 3GHz consume a stonking 90W. This is the minimum change as it doesn’t factor into account power loss through transistor leakage at higher clockspeeds (transistors are leakier when made to switch faster) and generally higher voltage required for such feats which in turn increases power consumption (transistors are leakier when more voltage is put through them).
Of course, the other way to increase CPU performance is to have more “brains”, which is what multi-cored products do. Power consumption in the worse case will only be however many cores you use. So if a single-cored 1GHz uses 10W, worse case is that a dual-cored 1GHz will use 20W and a quad-cored 1GHz will use 40W. In the real world though, such numbers will be lower as various components of the CPU such as L2 will be shared. Take the “Prescott” based Pentium4 clocked at 3.2GHz. That uses 84W so one would expect a dual-cored version of that to use 168W. However, it’s dual-cored “Smithfield” variant, branded as the “Pentium D” actually uses, as mentioned earlier, 130W total or 65W per core.
So with that out the way, the reason why overclocked “Dothan” Pentium-Ms are little marvels. When these little beasties are overclocked from their stock 2GHz to 2.7GHz, generally without the need to increase the core voltage, their real world performance can often better the very fastest desktop offerings of either Intel or AMD. Even when it is slower, it’s only a relatively small delta of about 10% to 20% and attributed to that platforms lack of SSE3 or raw memory bandwidth when compared with desktop beasties. The kicker is this tidbit though, even at 2.7GHz, a “Dothan” Pentium-M would consume only about 45W which is about 2 to 3 times less than power hungry desktop beasties yet overall it performs whatever task thrown at it faster.
Then as far as “Yonah” Pentium-M is concerned, it will fix the lack of memory bandwidth by increasing the FSB from 533MHz to 667MHz, plus the addition of SSE3 and other architectural improvements, all-in-all bringing about 10% improvements just from these. “Yonah” will also be dual-cored which depending on the task it’s doing will see an improvement of about 30% to 60% had it been single-cored. All this whilst using just 31W though the final clockspeed is unknown, best guess is that it’ll be comparable to “Dothan”, so about 2.0GHz to 2.5GHz. Collectively, “Yonah” has enough grunt to even kick an overclocked “Dothan” squarely in the nads yet reduce power consumption by 30% to 40%, dropping even more to aroung 15W to 20W with just one cored dynamically switched off, so it’s no wonder why the guys at The Inquirer have the following to say:
To impress your techie friends, point out to them that Yonah will have a shared L2 cache, and when it is running on battery power, one core will shut off, leaving the other core with the full 2MB cache. A 667FSB and the dual channel DDR2 should make this a potent chip.
Add in the massively improved SSE2 and new SSE3 units, and you have a multimedia monster. Other improvements, especially to IDIV will only add to this, you can probably encode a DVD on a flight, and still have battery life left over to watch the movie because this all will only consume 31W. Yonah, you go Grrl! Sorry about the lack of decorum, but it will be that good.
Sure the x86 ISA may be ugly compared to the “elegant” PowerPC one, though if I came to you with an iBook, PowerBook or MacMini where the performance would trounce any desktop offering based on PowerPC G5, Intel Pentium4/Xeon or AMD Athlon64 yet use less power than the G4 line allowing mobile devices to do useful work for as long as 8 hours, you’re saying you wouldn’t take it? Admittedly for the 8 hour mark, low-voltage “Yonah” will probably be required, therefore not clocking quite as high as it’s full-blown counterpart, it’s performance should still equal current desktop offerings.
“Yonah” will also include “Vanderpool” and “LaGrande” technologies from Intel. The former is hardware virtualisation and allows one CPU to run more than one operating system concurrently. This could give Apple a nice little niché, especially with enterprize as such machines could not only run OSX, but also Windows and various flavours of UNIX at the same time, aside from the fact they would also run natively as the same x86 architecture is used. This could be neatly protected with hardware DRM by the latter, locking Apple’s OSX out from commodity x86 boxes, which would leave them OSX free dispite the wonders of what hardware virtualisation brings.
So the PowerPC G4 to Intel “Yonah” transition is in the 32bit realm, meaning there is no “downgrade” and will give Apple’s platform a good kick up the arse as far as speed is concerned. For iMacs, as they use 64bit G5s, looking at Intel’s roadmap means Apple will probably use Intel “Sossaman”, which is the desktop variant of “Yonah”. The most obvious differences from “Yonah” is that it will support dual-processor configurations (giving you 4-cores total) and that it’ll be a 64bit processor as “Sossaman” supports EM64T, which collectively with AMD64 gets called x64.
Come late 2006, Intel’s “Netburst” architecture will be dead in the desktop space and as of 2007 dead in the server space too. A new architecture, derived from Pentium-M “Banias” where “Banias” also forms the basis of “Dothan” and “Yonah”, will be created. It is rumoured to have a pipe-line of just 12 stages, be very power effcient yet improve performance by 20% to 30% per clock & EM64T will feature throughout this family. The codenamed for this family are as follows; the mobile dual-cored Pentium-M successor “Merom”, the desktop dual-cored Pentium-D successor “Conroe”, the dual-processor dual-cored Xeon DP successor “Woodcrest” and finally the multi-processor quad-cored Xeon MP successor “Whitefield”. In terms of alignment to Apple’s transition, this will probably mean that higher end PowerBooks will get “Merom”, Power Macs will use “Conroe” and XServe will use either “Woodcrest”, “Whitefield” or perhaps both, expanding Apple’s server line!
As for why Apple chose Intel over AMD? Less you’re an AMD fanboy, the answer it pretty fucking obvious: Platform. In other words, Intel know how to make chipsets and really damn good chipsets at that too. By the time Apple shifts over in 2006, Intel’s chipsets will not only have WiMAX in addition to WiFi, a speed-bump to dual-channel DDR2 800MHz SDRAM support, plus all it’s existing features such as PCI-Express, IEEE 1394, USB 2.0, SATA 300 and so on. For the desktop side of things, how can one not be enticed with dual ATI or nVidia next generation graphics running through 2 PCI-Express ×16 slots as Apples wouldn’t be limited to boring old AGP?
Of course, hardware is no good without software and as far as compiling binaries for OSX is concerned, it is possible to make one “fat” binary which runs on PowerPC architecture as well as x86/x64 ones:
-arch arch
Compile for the specified target architecture arch. The allowable values are i386 and ppc. Multiple options work, and direct the compiler to produce “fat” binaries including object code for each architecture specified with -arch. This option only works if assembler and libraries are available for each architecture specified. (APPLE ONLY)
However, this is of buggerall use if you have a pre-compiled binary where you can’t just recompile it (plus fixing whatever compile errors that pop up due to the move from big-endian PowerPC to little-endian x86/x64) as you don’t have access directly or by proxy to the source-code. Apple’s cunning trick is that it appears to be a licencee of Transitive’s “QuickTransit” product which would allow OSX on x86/x64 to run PowerPC binaries at 80% native speed with a memory penalty of 25%, as per this eWeek article:
On average, translating the various instructions will produce about 80 percent of the computational performance of a native compilation, said Frank Weidel, lead solutions engineer at Transitive. The QuickTransit kernel also requires a memory penalty of about 25 percent per application, Weidel said. The amount of memory an application uses for data is not affected. However, the multiple instances of the technology will run side by side; for example, the company has been unable to break the QuickTransit application running 200 instances of the technology alongside one another, he said.
Intel’s future Pentium-Ms which Apple shall be using will have enough of a performance delta over the PowerPC CPUs they are replacing to run native PowerPC code via emulation faster than what the native line of PowerPCs can. Reminds me a lot of how back in the old days of Windows NT4 that the fastest way to run x86 code was to do it via emulation using FX!32 which came with Windows NT4 and running it on a DEC Alpha.
If there is still doubt, a certain company by the name of SGI have been happy with it having migrated their hardware products from MIPS Rx000 to Intel Itanium. Whilst Apple haven’t yet announced they will be using “QuickTransit”, though the eWeek.com article strongly hints that this is the case, with other suspects being Microsoft, whom would find a perfect way to use this for XBOX360 and run olderXBOX games. Ironically the move here was from x86 (Pentium-III 700MHz) to PowerPC “Xenon”.
The technology will be adopted and deployed by an undisclosed computer OEM in the fourth quarter, company executives said, and five more manufacturers are expected to announce support for the technology next year.
So what will really happen? We’ll just have to wait and see, though the thought of a “Yonah” based PowerBook or MacMini does make me want to tweak my nipples, even more so if it’ll run OSX, Windows Longhorn and some Linux all at the same time. :D