Arn posts a roundup of the Macworld 2006 rumors. I have always loved the work he has done there, even though MacRumors is an aggregator of rumors, not a news site itself, he seems to have a “taste†for what is good Mac news and what isn’t living up to his tag-line: “news and rumors you care about.†Truely.
One thing interesting in the report is the implication of 13″ widescreen Intel iBooks (“MacBooksâ€) and Intel-based Mac Minis in the spring. Caitlin notes that if true, it would leave a single non-Intel hole in the line up: pro desktops and servers (PowerMac G5 and XServes).
That is quite an intersting pickle. Intel reserves multi-processing for their “Xeon†brand, and an Intel Mac would have to go multi-processor to be comparable to the “quad†G5. These means we aren’t likely to see a replacement for these computers until the second half of next year.
I’m sure Apple is rationalizing this by saying that professionals make conservative purchasing decisions, the G5 is comparable to the fastest x86 computers, and many pro apps won’t “go universal†until later. All true, but given all the marketing muscle they’re putting behind how fast and efficient Intel is, don’t really know how they’ll pull this out without pissing off some of their fanbase.
Remember, in today’s world, hype cuts harder than the truth.
Intel iMac or IBM G5 Powermac
The lowest end PowerMac G5 is a 2×2 Ghz G5. There was some discussion on Flickr if this PowerMac G5 is slower than an iMac, including some bad advice from an “Apple rep†who seriously needs to be fired causing a little buyers remorse on one user.
These are dual processor systems, not dual-core which still affords some advantages when they are split (not so many advantages in the tight space and heat/power restrictions of a Powerbook or iMac). I think he is safe with the PowerMac purchase since evidence above points to it being the last one migrated to Intel.
It is simply not true that it will be slower than an iMac:
- As mentioned above, not all apps are “Universal” and thus need to run emulated on the Intel-based system. Jobs keeps harping Photoshop (notice he didn’t have one of his fabled Photoshop “shootoutsâ€?). The entire Apple Pro Line (Final Cut Studio, Aperture, Logic) etc. won’t be ready until March at the earliest (just in time for NAB). I guess MacOffice runs acceptably slow *shrug* I think we are well past the days when a spreadsheet and word processor were taxing our hardware.
- The memory and system bus is still slow relative to the Power Macintosh (though much faster than the G4). A lot of times your CPU is waiting on this or the disk.
- Apps that make heavy use of Altivec will need to be ported back to SSE. Yes, SSE2 is great compared to MMX, but not as good as VMX/Altivec/Velocity. You’ll probably notice this when you are using Compressor to do a batch encode—a G5 simply flies at these type of tasks.
- Graphics card performance. We are comparing a desktop computer to a notebook one, no matter how fast a notebook GPU they use. I notice the iLife and iWork apps are now making more heavy use of the CoreImage library which tries to use the GPU even more than Quartz Extreme does.
- Drive speed. This is especially true if you have a SATA card and run your drives in SATA RAID-0. Upgrade options here are very limited in the iMac world.
It is important to note that Apple’s own benchmarks are contrived: Those are two universal apps that they are comparing (check it out). I find this amusing that nobody has noticed this in their rush to bow to all things x86. Remember the days when we were quick to fault Apple for their contrived Photoshop benchmarks? How are these any different?
One thing the Intel iMac is going to do really well at is integer math. Notice how the SPEC_int delta relative to the iMac G5 (single CPU) is 3x while the SPEC_fp is only 2x? That’s one thing x86 has been weak at. Where should one see this? Well, if you have ever had to use Fink to compile kcachegrind you know how slow the PowerPC is (or rather, how really fast x86 is). Also you would see this in Safari web page rendering times—something Jobs showed in his keynote and similar “daily use†apps once they are Universal binaries.
(The flickr discussion mentioned the Intel iMacs have DVI out and spanned video. Neat! I missed that one.)
Craig Wood has posted some preliminary speed tests of the MacBook Pro.
Notice he compares it to the PowerMac 2x2Ghz under discussion above. In all but one test, (HD playback), the PowerMac beats the Mac Books Pro. This lends credence to my theory that the HDV rendering in FCP may be “contrived.”
Note. By “contrived” I do not mean false, but rather contrived to highlight a particular superiority/optimization instead of it being “typical”.
More benchmarks comparing the iMac Core Duo, this time from Ars Technica. Notice that, as predicted, it performs well on tests that exercise Universal Binaries, but it is slow on non-Universal apps (Photoshop). It isn’t sluggish when the apps aren’t used heavily.
Someone compared the Intel transition to the OS9->OSX transition and went on the quote his Steveness about how the Intel Macs are not for pros because of the poor Photoshop performance.
My response:
It isn’t exactly like other transitions.
OS9->OSX still ran natively, the issue was that a lot of libraries were not thread safe and thus software (pre-Carbonized) had to run in a virtual machine.
If anything this transition is closer to 68k->PPC, where you have emulated, native and "fat" binaries (except now they are Rosetta translated, ??, and "universal" binaries).
However there are some really important differences to consider.
1) Translation in Rosetta is greatly improved from emulation in the PPC transition. Basically Rosetta is a "just-in time" compiler. This means that it isn’t emulated so much as "compiled and cached" into x86. This is much faster and also alleviates some issues like endian.
2) Much software out there is Universal-Ready out of the box. This is because the primary development platform is unchanged and free (XCode). In those days the development platform moved from say ThinkC or MPW to CodeWarrior and none of those were free.
3) Coders today do not optimize as much in machine specific code as they used to.
4) Apple itself was running Mac OS X on x86 long before it was running in PPC, it was called NeXTSTeP, back then. They always maintained that, most of their effort has been in slowly migrating Carbon apps like iTunes.
Because of these, only large apps like Photoshop or Office which have a lot of legacy code or are written in older development environments will slow the transition.
As for Jobs saying that Pro’s shouldn’t adopt it, what choice does he have? Intel PowerMac’s won’t be out until Q3. If I am a pro who uses Final Cut Pro or Aperture as my main program, what should stop me from migrating to Intel when the Universal Binaries of those apps are available in March? As far as I can see, the only thing stopping me is the lack of an Intel PowerMac comparable to a quad-core G5.
This is exactly why I hate slashdot. Today, a week later, they finally clue into the contrived benchmarks Apple is using and they jump all over it by slinging the bullshit in the other direction.
Just because it only performs 10-20% faster on a universal binary single-threaded app like iMovie doesn’t mean the performance gain isn’t good. For all I know, the task may be disk or bus limited, but more likely Apple let iMovie remain single-threaded to encourage people to buy their pro products. Also the tasks benchmarked by MacWorld are likely heavily optimized to highlight the strengths of SIMD core in the G4/G5 (Altivec/VMX/Velocity).
After all, a user cares that it runs decently fast, but it isn’t like they aren’t going to be checking their e-mail or IMing while their iMovie is rendering. What is the CPU utilization of the two computers when you did the test, MacWorld?
Want to prove this thing is 2-3x faster? Open up a new install of fink and compile kCacheGrind from source.
This sort of thing reminds me of the day the G5 came out and everyone was smearing Apple for turning off hyperthreading in their Intel SPEC_fp, not realizing that hyperthreading on lowers x86’s performance on that benchmark.
Idiots.
I don’t know what is worse, Slashdot, or “reporters†like this who have a clear agenda.
Jobs has “contrived†tests which I mentioned long before these “benchmarks†came out proving me to be right. But he has a reason: to sell Macs, nor was he deliberately misleading: “Now everything’s not going to run 2-3x faster, you know the disks aren’t 2-3 times faster, etc. But in the most important benchmarks, 2-3 times faster.â€
What the fuck is this guy’s excuse? The loser couldn’t even perform his own benchmarks and deliberately misinterprets the data. I wonder if I dig up his old articles if I find him bagging Apple for doing Photoshop application benchmarks instead of SPEC_int/SPEC_fp in the G4 years? Or bagging Apple for “turning off hyperthreading†in the SPEC_int benchmark when the G5 was released?