What compiler does Apple use? As they are starting from scratch, they should be able to optimise for this specific chip without taking backward compatibility into account...

Steve Jobs said that he was talking about the processors being faster...and he specifically said not everything is going to be faster like the hard drives and memory etc etc. Just the processors which is why he showed the SPECmarks or whatever this phantom benchmark that, to my knowledge, isn't a free download from anywhere. Or was I the only one that heard him prefacing the results?

Oh well, let the Mac bashing continue, blood is in the water.

Lots of people here have run Linux or a Unix variant on very similar hardware. Surely they knew already the kind of performance they would get out of it, since OS X is basically unix under the covers. I don't think this should really be a surprise to many.

it should be noted that these are all single-threaded benchmarks so the second core doesn't help that much.
it should be interesting how these machines compare doing more things at once or running multi-threaded tasks.

Steve Jobs during the keynote at MacExpo when presenting Photoshop running on Rosetta:

While the performance of Photoshop is not gonna be strong enough in Rosetta for a professional that spends hours per day in photoshop, it's gonna be great for most of us who use photoshop occasionally.

Speed is a marketing issue. Real world performance not surprisingly lower.

Of course performance is behind. Mac's PowerPC hrdware was optimized for their code and their hardware. Apple must have time to make these tweaks on the new PC architecture. The G5's were also 64-bit running (at first) around 95% software emulation. Later versions of the OS brought this up to 100%. You see, its not all instant. They are developing on a new platform and need time to perfect their code.

And even though Mac carried on a subversion PC program for a while, they stopped a while ago. As the OS changed, the code changed, and they had to start all over.

Somehow, I can't help but feel this article is encouraging Microsoft-fanboy flaming.

Apparently the iMovie compression/export times were "dramatically slower" on the intel machine. They didn't list the results, stating that it was likely a bug; probably just the lack of Altivec support though. I think the value of Altivec on the PowerPC will only become more apparent over time.

Yep. I have an older 1GHz G4 PowerBook and just received the iMac 2.0 GHz dual-core. I compiled Qt 4.1 for both, and *roughly* (I wasn't paying exact attention to the clock) the iMac compiled the entire library (identical configure options) about 7 times faster than the single G4. About what I'd expect. For my shorter jobs it's also about 6-7x faster. The compiler (gcc) utilizes both cores nicely, as I can see with the system load monitor.

So the MacIntels using the Core Duo, seem to be faster than the G5, but not by much (at least in real life). From the keynote we learnt that MacBook Pros are also faster than PowerBooks G4. I wonder, in real life, how faster will the successors of the iBooks G4 be, since they will most likely use the Core Solo? I am really curious, since the iBooks G4 uses chips that are not significantly different from the Powerbooks G4. Will the Core Solo be up to the job?

And you know what? All of the above statements had significant elements of truth to them. Apple is doing nothing more than showing its products, accurately insofar as it goes, in the best possible light. Is this the least bit stunning?

Well, no, it's not but then I don't think the story was posted to try make us amazed and surprised.

This sort of thing tends to get blown off by Apple fans as "what did you expect", but Apple have a history of using basically meaningless measures of performance in their marketing literature and this should concern us. Sure, we follow the tech news and see these kinds of stories and maybe we knew better in the first place.

But statistics and relative measures of performance are going to be how many people who aren't into tech, lawyers, teachers, mothers, and so on, decide what products to buy. A computer is a serious investment at the best of times, and this trend of having hardware manufacturers (not just Apple) constantly walking the line between lying and merely being "creative" is harmful to the market as a whole. After all, Adam Smith pointed out several hundred years ago that the free market assumes a perfectly informed buyer, and this kind of crap from Jobs goes a long way to making people who matter not perfectly informed.

What's interesting about the article is that a lot of the operations should be taking advantage of Altivec on the PPC and whatever the heck intel uses (MMMXXX?) on the Core Duo. The fact that the intel chip was even slightly faster than the almighty altivec on tasks that should be optimized for it bodes exceptionally well.

How much does the extra core help here? Someone needs to fire up CHUD, turn off one of the procs and re-run the benchmarks.

Wow... I got modded troll with responses like this? I thought it would be common sense.

A) I was talking about the G5 in my comparison, the g4 laptops are irrelevant.
b) Dual core != 2x performance, not even close.

Not all companies go for showing the "absolute best case" benchmark. AMD is generally rather conservative with their performance ratings, and it's pretty rare that third-party benchmarks show an Athlon with a given performance rating having significantly different average performance than the "baseline comparison" CPU (One of the older P4s) running at the same clock rate as that performance rating.

A great example of just how conservative AMD is - The Venice core Athlon 64 3200+ has a 2.0 GHz core clock and 512k of L2 cache, using a 90nm process. Its closest dual-core variant (the Manchester core X2 3800+) has the same core clock, L2 cache per core, and manufacturing process. (They also have the same FSB speed, 1 GHz HyperTransport) Yes, that's right, the dual-core variant is only rated 18% higher than its closest single-core counterpart. (This is because currently, on average, a second core usually doesn't net you much benefit because so many CPU-intensive tasks do all the work in a single thread.)

Apple, on the other hand, is notorious for being overly optimistic in their speed comparisons - They always pick the benchmark which will make the competition look as bad as possible, to the point of even failing to use important performance features of the competition's CPU. (For example, back in the P2/P3 era, Apple constantly marketed their systems as being faster than a P2 or P3 with twice the clock speed - While the PPC did in general perform somewhat better per clock cycle than Intel's CPUs, the difference was not anywhere close to what Apple claimed it to be. The benchmark in question used Altivec on the PPC but failed to optimize for Intel whatsoever - No MMX or SSE was used, despite being available.)

To compare it to my previous example, Apple would have called the Athlon 64 X2 3800+ a 6400+ because it had two cores equivalent to the 3200+.

When it comes to inflated/BS benchmarks, Apple is one of the kings.

Ok, I don't think this is right. I watched part of the keynote the next day and when Jobs was talking about performance he said both chips were using the "best possible compiler" and these were identified as Intel's compiler for x86 and the IBM compiler for the PPC.

Although using Xcode, yes, they use GCC. I think at this point they were trying to get the best number possible. :)

Of course, when the G5 came out they used GCC when comparing it against the Pentium 4 as this was "fair". More likely, it was due to the fact the Pentium 4's architecture didn't show good performance when running under GCC due to it's long pipeline and SIMD dependenicies. They didn't show how an Athlon64 did either, which I suspect would have been close.

Not that I'm saying the G5 is a bad chip, but it is interesting how they used GCC then and now use Intel's compiler for benchmarks!

Did you read the article? The only non-Rosetta result that was slower was iPhoto (export to files). It came in at a "dramatic" 0.91x as fast as the G5. Well, I wouldn't call that dramatically slower.

Why was it slower? It's probably spending the vast majority of its time writing data to files. And guess what's the bottleneck there? The hard disk. The disk in the new Intel iMac is most likely slower than the disk in the older G5 (non-iSight) iMac. this post at the Ars forum [arstechnica.com] explains why. Apparently older iMacs had Maxtor disks, newer ones have Western Digital. And according to that post, the Maxtors are faster. Case closed.

As for the other tests, it would be interesting to see the results with varying (but equal) RAM configurations -- say, 512M, 1G, 2G. Does the Intel machine get faster relative to the G5 when both have more memory? Or does the memory help the G5 more? Does extra memory help Rosetta? What about running Rosetta apps multiple times?

It's a shame that none of the current reviews have done such a thorough enough test yet. It should be fairly easy to do, and vastly more informative!

And the tradeoffs between RISC/CISC can be complex and balance each other out in many ways. For example, on a CISC machine you might have an instruction to move a block of memory where on a RISC machine you might have to loop on load/store instructions. The CISC machine may break the move down internally to pretty much the same loop, but the instruction itself takes less memory. This leads to more instructions in the same cache space and possibly a shorter instruction pipeline needed for CISC. To balance that off, you might need more instruction cache memory for the equivalent amount of work on the RISC system.

Add "unrolling the loop" in the compiler for the RISC machine and you need even more physical instructions in the cache for the same function.

I'm not arguing that one architecture is "better" than the other here, just noting that the tradeoffs are complex and each architecture has quite an array of reasons for and/or against it. In the real world, the tradeoffs and complexities of either design philosophy have tended to pretty much equal each other out, and the performance of systems based on either architecture tend to be roughly equivalent on the same relative amount of chip real estate in the same physical circuit architecture.