Slashdot Mirror
Intel Mac Performance Behind Hype
Barry Norton writes "Steve Jobs, at the MacWorld tradeshow, boasted: 'the new iMac [with] Intel processor is two to three times faster than the iMac G5.' MacWorld (the publication) has been putting the iMacs through their paces. The results are a good deal less impressive than Steve's boast, showing an average performance increase of 10 to 25 per cent while performing a series of everyday tasks with software specially designed for the new systems." Ars Technica had another perspective on the new systems earlier this week.
Apparently nobody watched the Keynote, in which Steve himself said that other components (hard disk, memory, etc) were not faster, so the overall experience would not be as fast as the 2-3x numbers he posted. Based on the specInt numbers he shows, sure, it's a 2-3 times improvement, but even he caveated it!
If this were digg I'd call for a "No digg!" right about now.
--- witty signature
At the moment I think they're using gcc.
"Leo Fender was in a 'state of grace' when he designed the Stratocaster." -- Paul Reed Smith
But what was more significant was his frank acknowledgement that Photoshop operating via Rosetta wasn't going to be usable by professionals. The people jumping on the accusation of hype bandwagon need to take those comments into consideration. It's not often that on a new product rollout something is said that directly translates into "Hey, don't go out and buy this right now."
GCC 4 to be exact. They helped make it.
/dev/random
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.
If only 't were so simple.
Unix-like operating systems (Linux, *BSD, Solaris, AIX, HP/UX etc.) present a common standard interface to the world, however the implementation details behind that interface differs radically amongst those platforms, and even between kernel versions with Linux.
As such, while these OSes may be able to run the same software, they do so with very different performance characteristics.
As a starting point, you should consider the differences between System VR4, Linux 2.4, Linux 2.6 and FreeBSD. There are many good books on the subject.
Stick Men
That's pretty much the way CISC has been implemented since the IBM 360. John Cocke started the RISC thing with the IBM 801 because he believed the microcode interpreter loop could be replaced wwith better compilation.
Steve may have prefaced his remarks, but the 2x speed claims are mentioned several times on the Apple website. http://www.apple.com/imac/ In most places they do include a footnote disclaimer or say "up to" 2x, but the boldface text on the intro page clearly says "Rev up your digital life at speeds twice as fast as the previous iMac." There's also a blurb about a "whole new architecture".
I'd love to see some tests with Pro Apps like Apeture and Final Cut Pro. The other telling one would be Maya for rendering. Most people don't need their word processor to run faster but higher end graphics software needs speed. The Apple tests seemed to lean on the side of graphics intensive software so I'm curious about those numbers. I did play with Apeture on one. It was a single chip dual core. Opening files and some functions hesitated but we're talking RAW files on a single chip machine. I was pretty impressed and I'm not a Mac person. I'm sure if most of that was Apeture and not the machine but it's pretty amazing either way. There definately seems to be an overall speed increase no matter who tests them. These are transitional machines and they are selling basically for what current Macs of a similar speed do. I have to believe once they settle in and the chips are better supported they will be much faster. One of the biggest benefits no one hardly talks about is hardware multitasking. I think if you started a shot rendering say in Maya then started working on a model in Modo you'd find little or no slow down if Maya was set to single node. Normally the apps would be stepping on each other. I haven't had a chance to try running multiple apps since I haven't had a chance to build out a dual chip PC system but there's a definate benefit over software multitasking. I'd give the new Mac a year to settle in before debating speed too seriously. Remember the debacle with the P4s when they came out? They cost a fortune and inspite of denials at the time turned out to be much slower because the apps weren't taking advantage of the P4 architecture. Apple switched to a whole new chipset. Having them come out faster is impressive on it's own. Even the apps that are called native I'm sure need refinements. Most of these aren't going to be optimized for dual chips. Non pro apps normally either don't take advantage or don't take full advantage. With dual core the new standard that will change.
That never stopped Apple from publicly bashing the architecture whenver they could.
Actually, the Intel turtle and the smoked bunny ads ended their run years ago. Ever since Jobs came back and re-hired Chat-Day for their adds, it's all been saccharine pop music and pretty colors. Apple hasn't bashed an Intel chip via their marketing since back when the G4 was actually considered a fast chip.
Information wants to be anthropomorphized.
SPEC = Standard Performance Evaluation Corporation
Formerly System Performance Evaluation Cooperative
http://www.pcmag.com/encyclopedia_term/0,2542,t=S
And no, it isn't a free download from anywhere
http://www.spec.org/order.html
CPU2000 V1.3
[Fuck Beta]
o0t!
Actually the CISC frontend is a very GOOD thing as it allows better utilization of cache resources and lowers overall latency due to memory fetching. A pure RISC architecture would fail to perform today because of the disparity between CPU performance and main memory performance. Even socalled RISC chips like the PPC bear little resemblance to earlier RISC chips like MIPS.
There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
Xcode includes 4.0.1 of GCC but apple was using GCC 3.x to compile the kernel in 10.4. Kernel modules are C++, so it wasn't possible to use GCC 4.0 yet. (since GCC 4 tried to be more compliant.. even KDE 3.x didn't compile on it) Apple said they used intel compilers for the testing though I believe on the intel macs and ibm's compiler for the ppc build. I wish they would have used GCC since its more fair in a way. If anything its optimized for the x86 platform more, but its more apples to apples. :)
Only intel zealots would think that an intel chip would be 3 times faster anyway. POWER isn't that bad or Microsoft wouldn't have put them in xbox 360s. Another factor is that the software "optimized" for x86 hasn't been out long. Sure apple's been keeping the old nextstep port alive all these years (it ran on intel and 68k), but making it run and tuning it for the latest pentium chip are two different things.
MidnightBSD: The BSD for Everyone
I think people may be missing the fact that the dual core thing might be throwing a wrench into the mix, which could account for all of the discrepencies. There aren't many native apps, and I think the iLife suite isn't multi-processor capable yet, so a benchmarks won't show the benefits of the second core.
From a email to the xcode-users list...
...also from a blog entry...
In our tests, a large C++ project finishes a full clean build slightly (a matter of seconds) sooner on a Quad Tower than it does on a Core Duo iMac. So the 2-core Intel is only slightly slower than the 4-core Quad for full builds.
Warning: every project is different, and the dynamics of disk and cache speed and latency, processor saturation, process threading, and system memory will affect your results significantly. But we are very pleased with the IDE and compiler performance on the Intel chip.
gcc is certainly faster. Subversion compiled in 5 minutes, 16 seconds on my dual 2.7 g5 with 1.5 gigs of ram. It compiles in 4:32 on the 1.83ghz intel mac with 1 gig of ram. Which makes me happy.
Actually, the ads to which you refer were made by TBWA Chiat/Day, when Steve Jobs was CEO of Apple. However, you're right that it was a long time ago (1998). See the Great Apple Ads page for details.
On Integer tests, the Intel was already the faster chip. It's floats where the G5 owns. Put two x86 chips on a die, and of course it's always going to kick G5 ass six ways from Sunday on a SPEC test that relies mainly on Integer calculations. Why is this surprising anybody?
Apple bashed x86 because it was CISC. But since it no longer is, Apple's complaints no longer apply, so it's not hypocritical of Apple to use x86.
This is not true, even if you belive this is why Apple bashed the x86 platform.
1) Apple based the x86 platform because it was not their CPU or architecture. (Adobe even had to stand up to Apple, and say, hey Photoshop and Illustrator will run faster on Windows x86 PCs than on Macs)
2) The x86 CPUs are not any more CISC than the PowerPC CPUs are non-CISC. The x86 CPU architecture is a bit more complex than to call it a CISC processor, go look up some of the technologies AMD and INTEL are actually using. Even the Intel 486 used RISC technologies to augment the included CISC portions of the CPU.
3) The 68xxx CPUs in Prior Macs were CISC as well, and Apple bashed the Intel x86 CPUs then as well.
It comes down to Apple's disdain for whatever they were not using at the time. Just like they bashed Color displays in the 80s to as far as two-button Mice as late as the early 2000s.
You act like the x86 changed in the last couple of years, and it really hasn't. And even in the last couple of years, Apple was STILL bashing the x86 CPU market.
The only change is Apple got a sweet deal from Intel, and no longer wanted to develop the PowerPC technologies.
(And the Apple excuses that the PowerPCs/IBM development was not moving along fast enough was pure crap. MS took the PowerPC architecture and in less than a couple of years, pulled out a tri-core 3ghz version of the technology for a Video Game Console. - Just imagine if Apple has put this type of work in the PowerPC line, we might have Dual Tri-Core G6s from Apple by now instead of them running to Intel, and in the long run pushing out products that are AGAIN slower than the average Windows PC sold.)
In theory, you're right. There's a lot of factors that go into how efficient a CPU and an instruction set archietecture are.
The ARM, for example, added a subset of their usual instructions (Thumb mode) that used 16-bit ops instead of the usual 32 bit. On a system with slow memory, this turned about to be about 30% faster, even though the number of operations increased.
However, I'm reasonably certain that if you were to design a CISC instruction set today, it would not resemble the x86 ISA. You could certainly come up with instructions that would be more compiler friendly, be easier to pipeline, etc.
that's the very DEFINITION of CISC.
A true RISC chip has no decoder, and no micro-ops.
The code you write is the code it runs - thus making the chip extremely simple and therefore able to run very fast. The tradeoff being that it takes more instructions to do some things, and a pure RISC design wouldn't have OO execution and other such nice performance improving stuff.
A chip is either RISC, or it isn't, it can't have a decoder on top and still be RISC.
These days there is no RISC / CISC divide - because chip design has evolved in such a way that such over simplified lables can't be applied anymore.
But to say that Intels chips are RISC-like because instructions get decoded to micro-ops is to fundamentally misunderstand the philosophy of RISC.
Nope. The Pentium Pro had the micro-op stuff, as do its descendants.
Barry Norton writes "Steve Jobs, at the MacWorld tradeshow, boasted: 'the new iMac [with] Intel processor is two to three times faster than the iMac G5.'
No, that's not what he said, stop twisting his words to set up a straw man you can then revel in knocking down. If you watch Jobs' full keynote presentation you'll see that he specifically compares only processor benchmarks, not system benchmarks. He even made the disclaimer: "Now everything's not gonna run 2 to 3 X faster, you know the disks aren't 2 to 3 X, etc., but on the most important benchmarks, [the Core Duo] is 2 to 3 times faster [than the G5]."
Flying is easy, just throw yourself at the ground and miss. -Douglas Adams
I think that Apple has usually done a pretty good job at being very specific and far about their benchmarks.
Like when Apple used old 486 code to test on a Pentium, but used new PPC code on the Mac side?
Like when Apple used an old MMX code to test on a SSE equipped x86 CPU, but used new Altivec code on the Mac side?
Like when Apple's spec results for a particular x86 CPU did not match the official results, Apple used a weaker x86 compiler?
To be fair, this was all in marketing info, not in an engineering analysis, so I guess such games are fair. Apple's "tricks" were disclosed in the fine print.