?

Log in

No account? Create an account
 
 
 
 
 
 
Date: 2005-06-06 18:42 Security: public optimistic
Well, the cat's out of the bag - Jobs has confirmed in his WWDC keynote speech that the first Intel-based Mac will be shipping next June, with the lineup transitioned by June 2007. The demos on stage have been on OS X running on a 3.6GHz P4.

And 10.5 will be known as "Leopard".
 
 
 
 
 
 
Oh, I'd say IBM's chipmaking's not doing too badly - bear in mind they've landed all three of the next-gen consoles! Therein may well lie a good part of Apple's problem - IBM may have decided it's simpler to just develop new PPC variants and churn them out by the million, rather than spending time on high-performance desktop processors. Traditionally, the POWER family's not been at the leading edge of clock speed, being fabricated more for high reliability. Still, there must have been a pathetically weak commitment by IBM for Apple to've elected this route.

Not, I suppose, that it's all that surprising, inasmuch as it's been an open secret that OS X for Intel existed, but it was interesting to hear that the project's been maintained completely current, with all projects mandated to work on either platform. And Mach-O's long been fat binary capable, as the NextStep days showed, where binaries were routinely delivered for all four supported architectures (68k, x86, SPARC, PA-RISC. We actually had a couple boxen of the latter flavor in the office - two of HP's Gecko workstations. Very nice bits of kit indeed, but fearsomely expensive - something like $35k, ISTR), so we'll be seeing a return to those, in the form of "Universal Binaries". (Sounds very 007 :)

One aspect that caught my eye was the demonstration of "Rosetta", translating PPC to Intel on the fly, complete with demos of apps like Potatoshop running without modification on the new hardware. Perhaps this is the first public sighting of Transitive's technology?

As for OS X, there's some useful replies downthread. ^_^ Basically, Mach 3.0 microkernel with FreeBSD personality, with a blend of FreeBSD 4.4 and 5.x with a little NetBSD and some of their own work. Then the main APIs are Cocoa (from the NextStep side of the family) and Carbon (from old Mac OS), along with Java 1.4.2/1.5. There's more on Apple's developer site, needless to say.
Has there been any indication that Rosetta is Transitive-based? That would make things especially interesting...as between that and Universal Binaries, this need not be a one-way or one-time transition.
Still don’t know the answer, but reading Apple’s guide on Universal Binaries does bring up some educational tidbits: the emulation only goes as far as a G3. Applications that can optionally make use Altivec and the like on a G4 or G5 will still work, they’ll just fall back on the (emulated) G3 code. Applications requiring a G4 or G5 won’t work with Rosetta.
BTW, I notice the keynote's now available to view, with H.264 available for those able to take advantage of it. (Which now includes Windows users, with the QT7 preview now available)

Still churning through the Ars thread on the general topic, but it's mostly pure speculation, aside from a side note of some interest, explicitly stating that Intel Macs (still seems odd, writing those two words together) won't be using Open Firmware. I'd imagine this might signal a boost for Intel's EFI as a replacement for the supremely crusty BIOS of old.
Well, it is C|Net, but they're running a story confirming that it is indeed Transitive's work behind Rosetta.
Yay.

I'd seen some Xbench scores posted (PPC code being emulated on one of the Apple x86 development boxes), and the results were not disappointing. Somewhat retarded, certainly, but not horrible, and averaging out to about the same ballpark as my current PowerBook, which is fast enough to keep me employed.

The tidbit I found particularly interesting is that system calls and official API's will fall through to native code; these are not emulated. So those Xbench CPU scores, being fully self-contained, don't reflect what "real world" use will be like...it may actually be downright snappy. Encouraging stuff.

This is all pretty much moot for me, since about 95% of everything I do involves either the OS-bundled apps, or Adobe who've pledged OS X86 support. Bring it on.
It's not really very clear at the moment, but it sounds like the current Rosetta only translates PPC to Intel - certainly, it'd seem possible for Apple to offer the process in either direction, but they may choose to leave x86 compatibility to folks like the WINE project.

Appendix A in the Universal Binaries guide has some further information. Of particular note is that Rosetta only pretends to be offer a G3 - anything requiring Altivec appears not to be handled. But the basic architecture suggests that it does check what binaries are available, and what it's running on, automatically getting involved as necessary.

Just idle musing - what other viable architectures could Apple target? Would they be able to make use of, say, something from the UltraSparc or MIPS families? (After all, PA-RISC and Sparc were never exactly high on NeXT's priorities, but available nonetheless. And wasn't Canon interested at one point?)

Ah! Just seen you've spotted the same G3 limitation. Erf, time to snuffle around for SSE3 details and find out how many architected registers are available - presumably Altivec translation just wouldn't fit while providing any sane level of performance.
SPARC and MIPS are as dead as...well, PA-RISC. I was thinking perhaps if a few years down the road the whole x86 thing is tapped out just as a POWER8 or whatever were to come along and wow us all...Transitive would offer options there, since it sounds like they have pretty good emulation in every direction...or if they had occasion to use POWER and x86 on different product lines simultaneously later on, not just during the transition period. If it’s their own homebrew, don’t know if they’ll have that flexibility.

The only other possibly-reasonable architecture I could think of was XScale, if they were to do some sort of handheld or tablet device. And I don’t think that one’s listed among the current Transitive options (though presumably could be added if demand and performance allow).