12-19-2005, 10:14 PM

I know all the hype about the new ibooks that "may" come out in January. However, if these new computers are forced to used the emulator Rosetta to encode all the OSX stuff to X86 stuff, wont that slow the speed down a lot. So in other words you won't really get the true performance of the intel chip.

Also, as time goes on and more applications are made for OSX with Intel technology, will the Rosetta emulator still be needed? Or will the need for it go away and the true power of the Yonah processor can be unleased?

Thanks in advance.

12-19-2005, 10:22 PM

The rosetta transcodes the code a near realtime speed, so you won't notice a slowdown.

As for how long the rosetta is needed, it depends on how long apps are developed for the G# processors.

12-19-2005, 11:20 PM

From what I hear the Rosetta emulation runs almost invisibly... and I don't think that it's used for all apps. I think developers are simply porting some to the x86 platform right?

12-20-2005, 02:12 AM

OS X is already Intel native.

It's the applications that aren't all Intel ready, in fact most are still PPC only binaries, however Rosetta will emulate for those.

However I do not believe for one minute yet that it will be fast. I have written a fair bit on this in on other posts and I'm sure a search will yield those posts. However; knowing what I do on emulation technology, Rosetta probably isn't as fast as a lot of the hype is saying and for some time yet I believe that a native PPC machine will wallop an Intel emulating the PPC completely.

One of the big challenges is the endian swapping.

This will begin to change as more apps begin to become Intel native or PPC ones are replaced with new Intel or universal binary alternatives.

Vicky

12-20-2005, 03:07 PM

Didn't Steve demonstrate the mactel machine at one of the WWDC's? It showed the speed in which they ran at and how it worked with a maths program???

12-20-2005, 05:01 PM

Nothing saying those weren't Intel native...

And Mathematica was shown as was a case in point for the Intel transition and how relatively straightforward it is if you are using XCode already.

Again I await to be convinced when I see it in the flesh, if Rosetta can be so fast, how come the other way around - ie: Virtual PC at present is slow?

As I say I'll believe it when I see it, I respect Steve Jobs greatly, but it doesn't hurt sometimes to see things in the flesh than believe the marketing speak and hype.

Vicky

12-20-2005, 07:03 PM

I remember seeing some numbers that suggested Rosetta ran a PowerPC app at about 80% the speed of an Intel app.

I expect Rosetta will stay around for some time...modern Macs still ship with Classic, 5 years after OS X became the standard on new Macs. Even a Motorola 680x0 emulator is still included, eleven years after PowerPC became the standard. (Though both will be dropped with the move to Intel.)

12-20-2005, 07:37 PM

Originally Posted by lil

Again I await to be convinced when I see it in the flesh, if Rosetta can be so fast, how come the other way around - ie: Virtual PC at present is slow?

Although I agree with your overall point... I'll believe 80% when I see it too..... but in the case of VPC you are emulating an entire Proc/RAM hungry OS inside of another Proc/RAM hungry OS... and then running your apps... So I don't know that this comparison is really applicable.

12-21-2005, 05:28 AM

I suppose that much is true.

I'm just not sure where the quantum leap in speed will come from in terms of emulation. JIT emulation technologies are very good and probably is what Rosetta uses instead of the old 68k emulators transcription method.

The other things is, 80% is quoted but 80% of what. 80% of the speed of the Intel chip in it, ie: 2GHz -> 1.8GHz PPC; but then, MHz myths and all that, or 80% the speed of a PPC 750?

I think for the time being the best way to run PowerPC applications (ones that haven't made the transition to universal binary or you can't afford to upgrade to the new version which is a universal binary) will be on a PPC native machine.

Especially in the case of the Power Mac G5; we have read for a couple of years now how that machine batters Dual Xeon set-ups, and hence probably why it will be on of the last to undergo the ppc->x86 transition.

Vicky

12-21-2005, 03:34 PM

Originally Posted by lil

Especially in the case of the Power Mac G5; we have read for a couple of years now how that machine batters Dual Xeon set-ups, and hence probably why it will be on of the last to undergo the ppc->x86 transition.

As I said on another thread, the only PC that can currently match a quad G5 is the HP quad Opteron which is $3000 more expensive.

Nothing from intel currently even comes close.

Amen-Moses

12-22-2005, 04:37 PM

Think of this though.

With rosetta running, and intel processors, there will be a gap between that and PPC. I don't know how to phrase it. If Rosetta is fast enough, and so is the Processor, then wouldn't it overpass anyway. The Intel would have to be just that much faster, or better phrasing, the PPC would have to be that much slower.

I do agree, though, since the g5quad is so fast, the intel wouldn't catch up (while running rosetta). Hopefully, when the powermac goes intel, rosetta won't be needed.

But remember, we're talking about single core g4's here.. not quadg5s. I think that a dualcore yonah w/ Rosetta would not only close the gap, but make it noticeably faster

Guess we'll see, really soon (hopefully)

Pierre

12-24-2005, 10:26 PM

The real issue I have here is the problem of universal binaries. Programs that are UBs won't have ANY processor optimization, theoretically "flatlined" in that area, whereas many of the programs that we now enjoy are sped up because of optimization that comes with being native to only one processor. Unless of course, the "dual coding" allows optimization for both processors? I don't know enough about it, but it seems like they'd really be crippled in that area.

12-24-2005, 11:03 PM

well, if you have a 2GHz intel running at 70% speed, that would be 1.7GHz, and you had that in an iBook, that would be a ~350MHz speed boost. and then again for native apps (safari, iChat, mail, iPhoto, iMovie, DVD player etc) it would run at 100% speed because you wouldnt need rosetta... so I think because the intels are avialable at a much higher clock speed (I've heard the new yohnas are starting at dual cre 1.66GHz) that it will make up for the speed lost in emulation of certian apps

12-25-2005, 07:12 AM

I am guessing that apps would have to be optimized to take advantage of the dual-coreness of the Yonah's too..

12-25-2005, 12:29 PM

Originally Posted by Meyvn

The real issue I have here is the problem of universal binaries. Programs that are UBs won't have ANY processor optimization, theoretically "flatlined" in that area, whereas many of the programs that we now enjoy are sped up because of optimization that comes with being native to only one processor. Unless of course, the "dual coding" allows optimization for both processors?

A universal binary can be as optimized as an ordinary executable. Most apps will be optimized sufficiently by the compiler.

Sometimes compiler optimization is not enough; Photoshop for example is heavily optimized by hand both for AltiVec on the Mac (and SSE on Windows.)This is still possible with a UB; the existing AltiVec code can be used, and probably some of the SSE code can be brought over from the Windows version. This does require some serious effort by the developers; probably only compute-intensive software will get this sort of attention.