I've been thinking about something since reading about all the PC people complaining that Apple skewed the tests because they didn't use the compiler that produced the *higest* SPEC scores, not to mention that said compiler isn't even the most commonly used.
To me when you're testing the HARDWARE and *only the HARDWARE, you must try and keep everything the same as much as possible. That means using the same compiler for both machines. Sorta like testing automobile performance... You'd test the car using the SAME DRIVER and the SAME MODEL AND MANUFACTURER of the TIRES on the SAME TRACK. That means you must run these SPEC tests and see which machines produce the higher scores. VeriTest obviously did this, but the kicker is that even went further to see if the slower (Intel) scores could be boosted using the special features that might be available. Again, VeriTest did this.. They tested with SSE extensions turned *on*. Since SPEC doesn't test AltiVec on the PowerPC's I say that this test was MORE than fair -- even to the extent that I'd call it biased in favor of Intel! All that aside (and I'd really like some feedback on my speculation), is it at all possible that Apple issued these results with this one specific (particular?) benchmark knowing full well that it would cause such an uproar?
I'm beginning to wonder if they singled out SPEC in order to draw attention to it -- showing that it really is pretty useless in terms of what's important *computationaly* these days in terms of computing -- personal or otherwise. I think they did it to show that SPEC can no longer be *gospel* regarding system performance. There is too many tricks and tweaks that companies can use to show higher than usual SPEC marks. Apple may have done this to show that SPEC's results are misleading and somewhat flawed and that it's results have become *limited* in terms of what they tell you about a particular system. To be honest, I think it makes the PC side look bad, not Apple. Anyway, just something to think about.
Yeah. I'd like to wipe that guy's smug smile off his web-page...
Speaking of the Opteron...hardly a personal computer. Costs an absolute fortune.
We'll see if the Opteron is 50% faster when Altivec, Panther, the G5 are all optimised and IBM's 0.09 970 process their AMD ass... Plus, Opteron's still run on a crap OS. Xp is crap. I'm using it. I KNOW what I'm talking about.
You can get the G5 dual gigger for a reasonable (compared to last years Tower top end...) £2,300. That's damn good for what you're getting.
That's what I'm getting!
As for the Spec', Ed M, I think you're right. They do seem biased in Intel's favour if anything. As above...wait till Apple get the whole caboodle sorted by the Fall. Panther. Dual G5's optimised to run the OS. And real world apps to bench...not some x86 croney Spec test bed. Pixar said it was the highest performing machine. If they say it. I'll take their word.
I think the G5 will meet all comers...Opteron included...and even if an Opteron/Prescott came out slightly ahead...then I'm okay with that knowing IBM are moving 'swiftly' to 0.09 with the 970. How's AMD going to do with their 0.09? Wish them luck...
We'll have Panther. They won't.
Lemon Bon Bon
PS. Besides, for once, I don't care what Wintel has. The G5 is more than competitive. It's much cheaper than Intel's Xeon box. Kick ass G5 on Kick ass Panther...that's all I've been asking for...during these last four years. We NOW have hardware that can go toe-to-toe. That will do for me. And we've still got the Altivec factor. Can't wait to see it on that bus on real world apps... Nasty...
At first blush, the Opteron seems to be in the G5's ballpark performance-wise, although it seems to land in a higher price bracket.
It'll be hard to say more than that until a G5 actually ships, though. When people - including thorough, independent benchmarkers like Mike Breeden - can get their grubby mitts on the new PowerMac, we can start talking in more specific terms.
In the meantime, I wish AMD all the best against Intel.
To me when you're testing the HARDWARE and *only the HARDWARE, you must try and keep everything the same as much as possible. That means using the same compiler for both machines.
Am I the only one who doesn't buy this argument? The PPC and x86 code generation components of GCC were written by entirely different groups of people. They are not the same in any way, shape or form with the exception of the name. The parser and semantic analysis may be the same, but that seems to me to be largely irrelevant. It's not that I agree with the winlots on this one (I couldn't care less about the SPEC scores), but I still don't think the "same compiler" argument is valid (unless there's something about the compiler that I'm missing that makes it similar across platforms).
All in all I think Apple would have been better off sticking with real-world apps.
Also keep in mind, guys, that not only are the Opteron and Xeon systems MORE expensive (and underperforming in at least the Xeon's case), but also the fact that the Opteron will likely be running 32bit apps for some time to come. Not like the shrink-wrap consumer/professional applications developers are coming over in droves for the Opteron. Again, I'm not aware of any developers, Microsoft included, saying that they would port their wares to AMD's x86-64 for the desktop. If it's gonna happen, it will happen exactly like Microsoft said it would... No 64bit Windows on the personal consumer desktop systems until the end of the decade (or was that Intel? Or was it both?) Anyway.. the point is, I'm not sure there is a quick and efficient way of getting those x86 apps over in time. Think of all the bugs they will have to deal with all while trying to maintain backward compatibility or compatibility in general between al the many, many versions of Windows that are out there. Talk about the Windows OS *forking*... Sheesh!
Consumer confusion at an all-time high! Migration nightmares for the Wintelon crowd indeed. That's probably why they said by the end of the decade. There would be little incentive for Wintelon developers to code for these systems at this time, since the Opterons and Xeons are not aimed at the every-day-joe-blow-consumer space and the 32bit versions are claiming to run just fine through a 64bit OS, and since they're 32bit, it's unlikely that they will be able to address more than the 4-GB limit. Yep, Apple has the upper-hand right now and their timing couldn't have been better, given the current market conditions.
Hey, what's with the new developer tools that have been released that supposedly make it a breeze to port existing apps over to the new Panther/G5 systems? Can someone explain this a little better. Amorph? Programmer?
By the way... Where is StagflationSteve...? That character must be beside himself since the G5's were announced.. Let's face it, how is he going to explain that Apple's new G5 systems are much cheaper and better performing that the best the Wintelon crowd has to offer?
Am I the only one who doesn't buy this argument? The PPC and x86 code generation components of GCC were written by entirely different groups of people. They are not the same in any way, shape or form with the exception of the name. The parser and semantic analysis may be the same, but that seems to me to be largely irrelevant. It's not that I agree with the winlots on this one (I couldn't care less about the SPEC scores), but I still don't think the "same compiler" argument is valid (unless there's something about the compiler that I'm missing that makes it similar across platforms).
All in all I think Apple would have been better off sticking with real-world apps.
I agree that the conceit that any of this measures pure hardware is silly; pure hardware can't do anything but idle and heat your apartment. I certainly don't buy it. The conceit that using GCC across platforms somehow levels the playing field is equally bizarre, although at least it's no more absurd than claiming that any compiler or compilers could accomplish that goal. But if this is true, then comparing the "same" application across platforms is an unreliable measure of "raw" system performance as well, because applications are built with compilers.
It seems to me that if any use is to be made of SPEC, it should be admitted that what is being measured is a target platform, and then a target platform - that is, a combination of compiler (+ settings), OS and hardware that developers would actually target their applications for - is what should be tested. By the conventional wisdom of benchmarking, it would be almost useless to compare, say, GCC 3.3 on a 2x2GHz PowerMac G5 running OS X 10.3 to MS Visual C++ on a 2x3GHz Dual Xeon running Windows XP Professional, because so many things are different - but no matter what you do, too many things are going to be different. So you can either abandon the errand, or accept that what SPEC essentially is is a suite of application benchmarks (gzip is part of it, and that's certainly an application) and so it can't be used to test anything more finely grained than an application platform. Given that, the only platforms of any interest are the ones that people are actually going to use, so the people interested in comparos should get a couple of retail boxes, install the most common developer toolsets leaving the defaults in place, build SPEC, run it, and report the results. The results from this test will most closely mirror the actual capabilities available to the applications that people will use, and the methodology will make it difficult to tweak for benchmark results - if Intel wants better results on SPEC, they'd have to change their systems in a way that would be noticeable to anyone picking up an Intel-powered PC in retail, and in that case the higher SPEC score would be relevant in real terms.
I understand that this is a radical and outrageous thesis to most systems benchmarkers, but it's the only way I can imagine a cross-platform comparison that's in any way meaningful.
Also keep in mind, guys, that not only are the Opteron and Xeon systems MORE expensive (and underperforming in at least the Xeon's case), but also the fact that the Opteron will likely be running 32bit apps for some time to come. Not like the shrink-wrap consumer/professional applications developers are coming over in droves for the Opteron. Again, I'm not aware of any developers, Microsoft included, saying that they would port their wares to AMD's x86-64 for the desktop. If it's gonna happen, it will happen exactly like Microsoft said it would... No 64bit Windows on the personal consumer desktop systems until the end of the decade (or was that Intel? Or was it both?) Anyway.. the point is, I'm not sure there is a quick and efficient way of getting those x86 apps over in time. Think of all the bugs they will have to deal with all while trying to maintain backward compatibility or compatibility in general between al the many, many versions of Windows that are out there. Talk about the Windows OS *forking*... Sheesh!
I believe MS has announced an intention to release a version of Windows for AMD's x86-64, but I don't know more than that. If MS only releases a server edition of WIndows for x86-64, that would help keep Opteron based machines out of the PowerMac's market; but don't forget the UNIX-likes.
At any rate, the Opteron has none of the Itanium's difficulty running 32-bit applications. It doesn't do the job as elegantly as the 970 does, but it's not inelegant either. It looks to be an impressive performer, and for the sake of AMD's continued presence in the high-end CPU market (and for the sake of raising Intel's blood pressure) I hope it's everything AMD promises. The less room the Itanium has between the Opteron and the 970, the happier I'll be.
Quote:
Hey, what's with the new developer tools that have been released that supposedly make it a breeze to port existing apps over to the new Panther/G5 systems? Can someone explain this a little better. Amorph? Programmer?
As neither of us was sent to WWDC, I don't think we could muster any more information between us than is available to anyone else here.
The 970 supports the exact same instruction set that the 74xx does (the privileged and processor state instructions are different, but that's only relevant to a few parts of the kernel and the CHUD tools), so I can't imagine that "porting" means anything more than a recompile in all but the most pathological cases. Changes to the system frameworks shouldn't require recompilation in the overwhelming majority of cases - that's the big advantage of a framework. As usual, some Carbon apps will have to have code added to explicitly add support for some of the new features, but unless Apple takes this opportunity to deprecate the old event model and similar legacy cruft (please, Apple, please!), even the Bad Carbon Ports(TM) will run just fine as is.
If OS X isn't the UNIX by which all others are judged, then it likely will be very soon. It's already has the best UI of any UNIX in history. Bar none. And it covers all users from the extreme geeks right on through to the first-time computer users. Opteron will likely be facing a drought of native Windows apps; especially if Microsoft doesn't release a "consumer" version of Windows-64 for it. We'll see though.
And regarding Amorph's comments on SPEC... This only goes to show that SPEC may very well have outlived it's usefulness of being the main benchmark people often look to when attempting to surmise which platform is fastest. I just think there are way too many variables that can be manipulated to skew the test one way or the other, depending how you look at it. SPEC is supposed to test the *entire* system. That must include every aspect of the CPU. If it doesn't test SIMD (AltiVec) or perhaps some other new CPU feature that might exist, then it IS NOT evaluating the *entire* system, therefore it's results provide less than optimal information.
Amorph writes:
Quote:
At any rate, the Opteron has none of the Itanium's difficulty running 32-bit applications. It doesn't do the job as elegantly as the 970 does, but it's not inelegant either.
OK, so then what's the catch, that users will "have to know" or "be aware of" what types of apps they are running. Will the 32-bit apps run smoothly alongside the 64-bit native apps simultaneously? If so, how much of a system slowdown is expected? Then there is the OS question again. The applications/developer support questions, the driver and hardware support questions, the end-user-confusion questions, etc., etc., etc., eeeeeesh... Not something I'd be looking forward to. And in the end, Intel and Windows folk are likely to be stuck with x86 architecture for some time to come if AMD's chips take off.
It does look like machines with Intel chips are way overpriced and underperforming compared to Apple's G5s, but what about AMD's Opteron?
Are there Opteron systems out there that are competitive with the G5?
I went to www.boxxtech.com (a web site that crashes my current build of Mozilla, by the way) and priced a dual Opteron 244 system that matched the base dual 2.0 G5 as best as I could. The price came out to about US $4200 -- the same price range as the dual Xeon the G5 was put up against in the keynote.
It would be interesting to see this system in a bake-off with the G5 with real apps.
Does anyone know of any such tests run between Xeon and Opteron?
[Edit: I'd thought that "Opteron 244" meant 2.44 GHz. 244 is just a model number. The 244 runs at 1.8 GHz.]
I didnt really finish reading this thread cause I got to pissed. Of course the PC hardware junkies are going to go right for the spec tests trying to say apple cheated. Look Apple used an independent firm to do the testing, I`m quite sure they wouldnt have if they where going to cheat. Second if apple cheating then why did the mac perform so much better in all the bake-offs.. Look at mathmatica they couldnt even show the potential for that because the PC couldnt handle it. Its all in the apps baby. Unless you run spec for a living.
Thats my I bought the marketing point of view. I`m to tired to give out my own arguements right now. I`ll do so later.
It's a mixed bag of results with no clear overall winner. In these tests, the Opteron system falls behind the Xeon in the kinds of things Apple wants to emphasize, like audio, video, and 3-D processing.
So, SPEC numbers aside, I get the impression that Apple's bake-off might have been just as, if not more impressive against an Opteron-based system.
Level 3 cache on the G4 gets more bang for buck than on the G5, mainly because the G4 is so bandwidth starved (especially in Altivec). Although it might make some difference to the G5, it wouldn't be worth it (furthermore, the L3 cache would have to be faster and even more expensive to have a worthwhile effect on a faster, hungrier processor).
Do any Intel/AMD mainboards actually have L3 cache? The Power 4 does, but it's not a personal computer CPU. The last x86 CPU I can remember with L3 cache was the AMD K6/3.
Quote:
Saaay. Where's that smart-ass Digital Video site that was largin' it with their 'Dell twice as fast as powerMac G4' smugness? They didn't mind dishing it...can they take it?
To coincide with the Panther release...at an even higher price bracket? Uber-Uber-Unix Workstation?
It suddenly occured to me...that a rev B Powermac G5 will get you an extra (assumption by LBB here...) 500 970 style mhz. Dual that. Yer talking 1 gig of 970 oomph over the current top of the line. On fpu performance, that would be like having a 3 gig G4 just with such a mere bump! Or a 1.7 G4 on integer? Or a 2 gig G4 overall. They don't make them yet! And if the bus stretches with each mhz bump...you get more bandwidth from a 970 04 San Fran' speed bump than you have in the current top of the line G4 Tower. Hmmm. Whatever, that is a handy extra 'oomph' to have. It should run the Opteron alot closer if the Opteron IS ahead.
If Apple don't ship the 970 towers until a little later due to shipping schedule crushing demand...then I may wait until Rev B turns up.
I've watched the keynote rebroadcast 3 times now...and I'm still in shock. That machine is a work of art. Panther is so much more the complete article...and I can't wait to reclaim my soul as a Mac user...
All this Spec' bull is a little churlish in my eyes. Alot of analysts have swallowed Intel's 'Mhz' bull for years. Seemingly unwilling to challenge those speed claims. The most blatant 'cheat' being the Pentium 3 1 gig to Pentium 4 at 1.4 gig transition. I often wonder...if the Pentium 4 was actually a Pentium 3...what would that performance really be? 2.5? 2.3? I suppose that's 'ifs' and 'maybes.'
Even IF the 'G5' is a 'catch up' machine. It's one hell of a catch up. A grand cheaper than its competition. Certainly the Boxx boxes. It looks ahead to me in real world apps. That's where it counts. And instinct says it won't be long before we have some kind of speed bump. I doubt we'll have to wait 9 months. 12. Or 18! And I'm sure the bump won't just be .250! Each bump you get with a 970 gets you double yer G4 progress! So 50% may not sound that much...to some...but to me...1 gig of 970 mhz improvement actually gives you 1.7 integer or 2 gig overall or 2.7 fpu G4 of G4 performance in 1 year! Put like that, it IS IMPRESSIVE!
Taken over a year between the last available G4 tower and the 3 gig 970 available this time next year...it's just a whole 'nuther ball game. The G4 will seem like the dark ages.
II just found this Opteron vs. Xeon article: Duel of the Titans: Opteron vs. Xeon
It's a mixed bag of results with no clear overall winner. In these tests, the Opteron system falls behind the Xeon in the kinds of things Apple wants to emphasize, like audio, video, and 3-D processing.
So, SPEC numbers aside, I get the impression that Apple's bake-off might have been just as, if not more impressive against an Opteron-based system.
Yeah. When I looked at the app benches at Tom's hardware I came away distinctly unimpressed by the Opteron in single cpu form and in dual? It seemed to be hanging out with the Xeon. So what's the story there..?
I suppose we'll have to wait for real world performance.
But for those creative pros who had to leave the Mac for performance reasons but who loved the OS...Panther and G5 will probably bring them home! And they can buy all their 'X' versions of their software with the £1000 they saved over buying the Xeon workstations...
If OS X isn't the UNIX by which all others are judged, then it likely will be very soon. It's already has the best UI of any UNIX in history. Bar none. And it covers all users from the extreme geeks right on through to the first-time computer users. Opteron will likely be facing a drought of native Windows apps; especially if Microsoft doesn't release a "consumer" version of Windows-64 for it. We'll see though.
Apple has some work to do before OS X is the UNIX by which all others are judged, and they might never do some of that work: It's not really necessary for OS X to have either the bulletproof uptime or the robust enterprise-grade toolset available on, say, AIX or OSF/1 (or whatever they're calling it now - Compaq UNIX?). Of course, it won't have their price tags, either.
Quote:
[re: the Opteron]
OK, so then what's the catch, that users will "have to know" or "be aware of" what types of apps they are running. Will the 32-bit apps run smoothly alongside the 64-bit native apps simultaneously? If so, how much of a system slowdown is expected? Then there is the OS question again. The applications/developer support questions, the driver and hardware support questions, the end-user-confusion questions, etc., etc., etc., eeeeeesh... Not something I'd be looking forward to. And in the end, Intel and Windows folk are likely to be stuck with x86 architecture for some time to come if AMD's chips take off.
I'm not recalling the exact nature of AMD's solution - I believe the processor has to switch modes, which it can do on the fly (say, between context switches). It's definitely refined enough that end users won't have to care whether they're running 32 or 64 bit applications, or both at once. Note that applications compiled for the Opteron's 64-bit mode gain access to twice as many registers, so there's an incentive to compile to 64 bit for a bit of extra performance even if the application doesn't need the 64-bit support itself. (The 970 reveals the same number of registers to both 32- and 64-bit applications, and it sports the generous number of registers common to PowerPCs.)
THG noted in the shootout article linked above that since the Opteron has a memory controller on board, it can only be connected to DDR333 RAM. If you wanted to hook an Opteron up to, say, dual channel 128-bit DDR400 RAM, you'd have to either wait for AMD to release a revised Opteron that supported that configuration or purchase a chip that bypasses the onboard memory controller in favor of its own (THG discusses this). On the other hand, each Opteron gets its own bank of RAM, up to 8GB - but that advantage wouldn't really manifest until you got to 4+ CPU systems.
Based on that article, I'm going to speculate that the G5 will still be at the front of the pack when it rolls out - if not in SPEC, than in real-world application benchmarking. If there is something faster on the landscape, it won't cost $3K.
It's a mixed bag of results with no clear overall winner. In these tests, the Opteron system falls behind the Xeon in the kinds of things Apple wants to emphasize, like audio, video, and 3-D processing.
So, SPEC numbers aside, I get the impression that Apple's bake-off might have been just as, if not more impressive against an Opteron-based system.
This article showed also some interesting points : in some task a single opteron 1,8 is slower than the athlon 2400. The opteron is basically an athlon with a different memory controller and a bigger L2 cache (and of course 64 bit integer and SSE2 instructions set). I fear that the athlon 64with only 256 KB of L2 cache will sucks.
Based on that article, I'm going to speculate that the G5 will still be at the front of the pack when it rolls out - if not in SPEC, than in real-world application benchmarking. If there is something faster on the landscape, it won't cost $3K.
Agreed.
Let me also add that this is only the *beginning* for this new line or processor from IBM. I'm sure we'll see additional compiler optimizations and applications developers willing to take advantage of the CPU's features.
I also remember Steve Jobs mentioning that the 970's were built for massive parallelism in mind -- full SMP support. I then remember him mentioning that the powermacs have 2 processors and that it's only the beginning ... or something to that effect. Perhaps someone here can recall what Jobs said about SMP. I may have gotten it wrong.
As far as the Opterons go... I just don't see them hitting the CompUSAs and CircuitCity type outlets any time soon, but considering how much trouble AMD has been having with getting their tech to market, it does make you wonder just how much longer they can keep pushing the envelope.. the same goes for Intel. This CPU war has caused them to push harder and faster in a race to bring the fastest CPUs to market as quickly as possible no matter what the cost. How much further does either company have with respect to pushing the envelope? Remember, Intel was working on the Itanium for a VERY long time. I suspect that they saw the end of the x86 line coming sooner than they would have hoped -- that's probably why they started the project. Now AMD goes and brings out AMD-X86-64 -- x86-64 is something that Intel *didn't* want to have to deal with. I can't imagine how it will pan out for the Wintelon crowd. I don't think they've been thinking about those types of problems.
Comments
To me when you're testing the HARDWARE and *only the HARDWARE, you must try and keep everything the same as much as possible. That means using the same compiler for both machines. Sorta like testing automobile performance... You'd test the car using the SAME DRIVER and the SAME MODEL AND MANUFACTURER of the TIRES on the SAME TRACK. That means you must run these SPEC tests and see which machines produce the higher scores. VeriTest obviously did this, but the kicker is that even went further to see if the slower (Intel) scores could be boosted using the special features that might be available. Again, VeriTest did this.. They tested with SSE extensions turned *on*. Since SPEC doesn't test AltiVec on the PowerPC's I say that this test was MORE than fair -- even to the extent that I'd call it biased in favor of Intel! All that aside (and I'd really like some feedback on my speculation), is it at all possible that Apple issued these results with this one specific (particular?) benchmark knowing full well that it would cause such an uproar?
I'm beginning to wonder if they singled out SPEC in order to draw attention to it -- showing that it really is pretty useless in terms of what's important *computationaly* these days in terms of computing -- personal or otherwise. I think they did it to show that SPEC can no longer be *gospel* regarding system performance. There is too many tricks and tweaks that companies can use to show higher than usual SPEC marks. Apple may have done this to show that SPEC's results are misleading and somewhat flawed and that it's results have become *limited* in terms of what they tell you about a particular system. To be honest, I think it makes the PC side look bad, not Apple. Anyway, just something to think about.
--
Ed
Speaking of the Opteron...hardly a personal computer. Costs an absolute fortune.
We'll see if the Opteron is 50% faster when Altivec, Panther, the G5 are all optimised and IBM's 0.09 970 process their AMD ass... Plus, Opteron's still run on a crap OS. Xp is crap. I'm using it. I KNOW what I'm talking about.
You can get the G5 dual gigger for a reasonable (compared to last years Tower top end...) £2,300. That's damn good for what you're getting.
That's what I'm getting!
As for the Spec', Ed M, I think you're right. They do seem biased in Intel's favour if anything. As above...wait till Apple get the whole caboodle sorted by the Fall. Panther. Dual G5's optimised to run the OS. And real world apps to bench...not some x86 croney Spec test bed. Pixar said it was the highest performing machine. If they say it. I'll take their word.
I think the G5 will meet all comers...Opteron included...and even if an Opteron/Prescott came out slightly ahead...then I'm okay with that knowing IBM are moving 'swiftly' to 0.09 with the 970. How's AMD going to do with their 0.09? Wish them luck...
We'll have Panther. They won't.
Lemon Bon Bon
PS. Besides, for once, I don't care what Wintel has. The G5 is more than competitive. It's much cheaper than Intel's Xeon box. Kick ass G5 on Kick ass Panther...that's all I've been asking for...during these last four years. We NOW have hardware that can go toe-to-toe. That will do for me. And we've still got the Altivec factor. Can't wait to see it on that bus on real world apps... Nasty...
When was the last time Apple could say that?
I can't whinge about the price. That's why I'm getting one...
C'mon, Apple...ship it already!
Lemon Bon Bon
Are there Opteron systems out there that are competitive with the G5?
It'll be hard to say more than that until a G5 actually ships, though. When people - including thorough, independent benchmarkers like Mike Breeden - can get their grubby mitts on the new PowerMac, we can start talking in more specific terms.
In the meantime, I wish AMD all the best against Intel.
Originally posted by Ed M.
To me when you're testing the HARDWARE and *only the HARDWARE, you must try and keep everything the same as much as possible. That means using the same compiler for both machines.
Am I the only one who doesn't buy this argument? The PPC and x86 code generation components of GCC were written by entirely different groups of people. They are not the same in any way, shape or form with the exception of the name. The parser and semantic analysis may be the same, but that seems to me to be largely irrelevant. It's not that I agree with the winlots on this one (I couldn't care less about the SPEC scores), but I still don't think the "same compiler" argument is valid (unless there's something about the compiler that I'm missing that makes it similar across platforms).
All in all I think Apple would have been better off sticking with real-world apps.
Consumer confusion at an all-time high! Migration nightmares for the Wintelon crowd indeed. That's probably why they said by the end of the decade. There would be little incentive for Wintelon developers to code for these systems at this time, since the Opterons and Xeons are not aimed at the every-day-joe-blow-consumer space and the 32bit versions are claiming to run just fine through a 64bit OS, and since they're 32bit, it's unlikely that they will be able to address more than the 4-GB limit. Yep, Apple has the upper-hand right now and their timing couldn't have been better, given the current market conditions.
Hey, what's with the new developer tools that have been released that supposedly make it a breeze to port existing apps over to the new Panther/G5 systems? Can someone explain this a little better. Amorph? Programmer?
By the way... Where is StagflationSteve...? That character must be beside himself since the G5's were announced.. Let's face it, how is he going to explain that Apple's new G5 systems are much cheaper and better performing that the best the Wintelon crowd has to offer?
--
Ed
Originally posted by Fluffy
Am I the only one who doesn't buy this argument? The PPC and x86 code generation components of GCC were written by entirely different groups of people. They are not the same in any way, shape or form with the exception of the name. The parser and semantic analysis may be the same, but that seems to me to be largely irrelevant. It's not that I agree with the winlots on this one (I couldn't care less about the SPEC scores), but I still don't think the "same compiler" argument is valid (unless there's something about the compiler that I'm missing that makes it similar across platforms).
All in all I think Apple would have been better off sticking with real-world apps.
I agree that the conceit that any of this measures pure hardware is silly; pure hardware can't do anything but idle and heat your apartment. I certainly don't buy it. The conceit that using GCC across platforms somehow levels the playing field is equally bizarre, although at least it's no more absurd than claiming that any compiler or compilers could accomplish that goal. But if this is true, then comparing the "same" application across platforms is an unreliable measure of "raw" system performance as well, because applications are built with compilers.
It seems to me that if any use is to be made of SPEC, it should be admitted that what is being measured is a target platform, and then a target platform - that is, a combination of compiler (+ settings), OS and hardware that developers would actually target their applications for - is what should be tested. By the conventional wisdom of benchmarking, it would be almost useless to compare, say, GCC 3.3 on a 2x2GHz PowerMac G5 running OS X 10.3 to MS Visual C++ on a 2x3GHz Dual Xeon running Windows XP Professional, because so many things are different - but no matter what you do, too many things are going to be different. So you can either abandon the errand, or accept that what SPEC essentially is is a suite of application benchmarks (gzip is part of it, and that's certainly an application) and so it can't be used to test anything more finely grained than an application platform. Given that, the only platforms of any interest are the ones that people are actually going to use, so the people interested in comparos should get a couple of retail boxes, install the most common developer toolsets leaving the defaults in place, build SPEC, run it, and report the results. The results from this test will most closely mirror the actual capabilities available to the applications that people will use, and the methodology will make it difficult to tweak for benchmark results - if Intel wants better results on SPEC, they'd have to change their systems in a way that would be noticeable to anyone picking up an Intel-powered PC in retail, and in that case the higher SPEC score would be relevant in real terms.
I understand that this is a radical and outrageous thesis to most systems benchmarkers, but it's the only way I can imagine a cross-platform comparison that's in any way meaningful.
Originally posted by Ed M.
Also keep in mind, guys, that not only are the Opteron and Xeon systems MORE expensive (and underperforming in at least the Xeon's case), but also the fact that the Opteron will likely be running 32bit apps for some time to come. Not like the shrink-wrap consumer/professional applications developers are coming over in droves for the Opteron. Again, I'm not aware of any developers, Microsoft included, saying that they would port their wares to AMD's x86-64 for the desktop. If it's gonna happen, it will happen exactly like Microsoft said it would... No 64bit Windows on the personal consumer desktop systems until the end of the decade (or was that Intel? Or was it both?) Anyway.. the point is, I'm not sure there is a quick and efficient way of getting those x86 apps over in time. Think of all the bugs they will have to deal with all while trying to maintain backward compatibility or compatibility in general between al the many, many versions of Windows that are out there. Talk about the Windows OS *forking*... Sheesh!
I believe MS has announced an intention to release a version of Windows for AMD's x86-64, but I don't know more than that. If MS only releases a server edition of WIndows for x86-64, that would help keep Opteron based machines out of the PowerMac's market; but don't forget the UNIX-likes.
At any rate, the Opteron has none of the Itanium's difficulty running 32-bit applications. It doesn't do the job as elegantly as the 970 does, but it's not inelegant either. It looks to be an impressive performer, and for the sake of AMD's continued presence in the high-end CPU market (and for the sake of raising Intel's blood pressure) I hope it's everything AMD promises. The less room the Itanium has between the Opteron and the 970, the happier I'll be.
Hey, what's with the new developer tools that have been released that supposedly make it a breeze to port existing apps over to the new Panther/G5 systems? Can someone explain this a little better. Amorph? Programmer?
As neither of us was sent to WWDC, I don't think we could muster any more information between us than is available to anyone else here.
The 970 supports the exact same instruction set that the 74xx does (the privileged and processor state instructions are different, but that's only relevant to a few parts of the kernel and the CHUD tools), so I can't imagine that "porting" means anything more than a recompile in all but the most pathological cases. Changes to the system frameworks shouldn't require recompilation in the overwhelming majority of cases - that's the big advantage of a framework. As usual, some Carbon apps will have to have code added to explicitly add support for some of the new features, but unless Apple takes this opportunity to deprecate the old event model and similar legacy cruft (please, Apple, please!), even the Bad Carbon Ports(TM) will run just fine as is.
but don't forget the UNIX-likes
If OS X isn't the UNIX by which all others are judged, then it likely will be very soon. It's already has the best UI of any UNIX in history. Bar none. And it covers all users from the extreme geeks right on through to the first-time computer users. Opteron will likely be facing a drought of native Windows apps; especially if Microsoft doesn't release a "consumer" version of Windows-64 for it. We'll see though.
And regarding Amorph's comments on SPEC... This only goes to show that SPEC may very well have outlived it's usefulness of being the main benchmark people often look to when attempting to surmise which platform is fastest. I just think there are way too many variables that can be manipulated to skew the test one way or the other, depending how you look at it. SPEC is supposed to test the *entire* system. That must include every aspect of the CPU. If it doesn't test SIMD (AltiVec) or perhaps some other new CPU feature that might exist, then it IS NOT evaluating the *entire* system, therefore it's results provide less than optimal information.
Amorph writes:
At any rate, the Opteron has none of the Itanium's difficulty running 32-bit applications. It doesn't do the job as elegantly as the 970 does, but it's not inelegant either.
OK, so then what's the catch, that users will "have to know" or "be aware of" what types of apps they are running. Will the 32-bit apps run smoothly alongside the 64-bit native apps simultaneously? If so, how much of a system slowdown is expected? Then there is the OS question again. The applications/developer support questions, the driver and hardware support questions, the end-user-confusion questions, etc., etc., etc., eeeeeesh... Not something I'd be looking forward to. And in the end, Intel and Windows folk are likely to be stuck with x86 architecture for some time to come if AMD's chips take off.
--
Ed
Originally posted by Tuttle
It does look like machines with Intel chips are way overpriced and underperforming compared to Apple's G5s, but what about AMD's Opteron?
Are there Opteron systems out there that are competitive with the G5?
I went to www.boxxtech.com (a web site that crashes my current build of Mozilla, by the way) and priced a dual Opteron 244 system that matched the base dual 2.0 G5 as best as I could. The price came out to about US $4200 -- the same price range as the dual Xeon the G5 was put up against in the keynote.
It would be interesting to see this system in a bake-off with the G5 with real apps.
Does anyone know of any such tests run between Xeon and Opteron?
[Edit: I'd thought that "Opteron 244" meant 2.44 GHz. 244 is just a model number. The 244 runs at 1.8 GHz.]
Thats my I bought the marketing point of view. I`m to tired to give out my own arguements right now. I`ll do so later.
It's a mixed bag of results with no clear overall winner. In these tests, the Opteron system falls behind the Xeon in the kinds of things Apple wants to emphasize, like audio, video, and 3-D processing.
So, SPEC numbers aside, I get the impression that Apple's bake-off might have been just as, if not more impressive against an Opteron-based system.
Do any Intel/AMD mainboards actually have L3 cache? The Power 4 does, but it's not a personal computer CPU. The last x86 CPU I can remember with L3 cache was the AMD K6/3.
Saaay. Where's that smart-ass Digital Video site that was largin' it with their 'Dell twice as fast as powerMac G4' smugness? They didn't mind dishing it...can they take it?
Sensing some glee and smugness
Could the 2.5 yields be stock piling now?
To coincide with the Panther release...at an even higher price bracket? Uber-Uber-Unix Workstation?
It suddenly occured to me...that a rev B Powermac G5 will get you an extra (assumption by LBB here...) 500 970 style mhz. Dual that. Yer talking 1 gig of 970 oomph over the current top of the line. On fpu performance, that would be like having a 3 gig G4 just with such a mere bump! Or a 1.7 G4 on integer? Or a 2 gig G4 overall. They don't make them yet! And if the bus stretches with each mhz bump...you get more bandwidth from a 970 04 San Fran' speed bump than you have in the current top of the line G4 Tower. Hmmm. Whatever, that is a handy extra 'oomph' to have. It should run the Opteron alot closer if the Opteron IS ahead.
If Apple don't ship the 970 towers until a little later due to shipping schedule crushing demand...then I may wait until Rev B turns up.
I've watched the keynote rebroadcast 3 times now...and I'm still in shock. That machine is a work of art. Panther is so much more the complete article...and I can't wait to reclaim my soul as a Mac user...
All this Spec' bull is a little churlish in my eyes. Alot of analysts have swallowed Intel's 'Mhz' bull for years. Seemingly unwilling to challenge those speed claims. The most blatant 'cheat' being the Pentium 3 1 gig to Pentium 4 at 1.4 gig transition. I often wonder...if the Pentium 4 was actually a Pentium 3...what would that performance really be? 2.5? 2.3? I suppose that's 'ifs' and 'maybes.'
Even IF the 'G5' is a 'catch up' machine. It's one hell of a catch up. A grand cheaper than its competition. Certainly the Boxx boxes. It looks ahead to me in real world apps. That's where it counts. And instinct says it won't be long before we have some kind of speed bump. I doubt we'll have to wait 9 months. 12. Or 18! And I'm sure the bump won't just be .250! Each bump you get with a 970 gets you double yer G4 progress! So 50% may not sound that much...to some...but to me...1 gig of 970 mhz improvement actually gives you 1.7 integer or 2 gig overall or 2.7 fpu G4 of G4 performance in 1 year! Put like that, it IS IMPRESSIVE!
Taken over a year between the last available G4 tower and the 3 gig 970 available this time next year...it's just a whole 'nuther ball game. The G4 will seem like the dark ages.
Lemon Bon Bon
Originally posted by Lemon Bon Bon
That 2.5 gig 970 IBM PR page. That was on .13?
Could the 2.5 yields be stock piling now?
I think that the 2.5GHz 970 was a typo since it had a 900MHz bus.
II just found this Opteron vs. Xeon article: Duel of the Titans: Opteron vs. Xeon
It's a mixed bag of results with no clear overall winner. In these tests, the Opteron system falls behind the Xeon in the kinds of things Apple wants to emphasize, like audio, video, and 3-D processing.
So, SPEC numbers aside, I get the impression that Apple's bake-off might have been just as, if not more impressive against an Opteron-based system.
Yeah. When I looked at the app benches at Tom's hardware I came away distinctly unimpressed by the Opteron in single cpu form and in dual? It seemed to be hanging out with the Xeon. So what's the story there..?
I suppose we'll have to wait for real world performance.
But for those creative pros who had to leave the Mac for performance reasons but who loved the OS...Panther and G5 will probably bring them home! And they can buy all their 'X' versions of their software with the £1000 they saved over buying the Xeon workstations...
Lemon Bon Bon
Originally posted by Ed M.
If OS X isn't the UNIX by which all others are judged, then it likely will be very soon. It's already has the best UI of any UNIX in history. Bar none. And it covers all users from the extreme geeks right on through to the first-time computer users. Opteron will likely be facing a drought of native Windows apps; especially if Microsoft doesn't release a "consumer" version of Windows-64 for it. We'll see though.
Apple has some work to do before OS X is the UNIX by which all others are judged, and they might never do some of that work: It's not really necessary for OS X to have either the bulletproof uptime or the robust enterprise-grade toolset available on, say, AIX or OSF/1 (or whatever they're calling it now - Compaq UNIX?). Of course, it won't have their price tags, either.
[re: the Opteron]
OK, so then what's the catch, that users will "have to know" or "be aware of" what types of apps they are running. Will the 32-bit apps run smoothly alongside the 64-bit native apps simultaneously? If so, how much of a system slowdown is expected? Then there is the OS question again. The applications/developer support questions, the driver and hardware support questions, the end-user-confusion questions, etc., etc., etc., eeeeeesh... Not something I'd be looking forward to. And in the end, Intel and Windows folk are likely to be stuck with x86 architecture for some time to come if AMD's chips take off.
I'm not recalling the exact nature of AMD's solution - I believe the processor has to switch modes, which it can do on the fly (say, between context switches). It's definitely refined enough that end users won't have to care whether they're running 32 or 64 bit applications, or both at once. Note that applications compiled for the Opteron's 64-bit mode gain access to twice as many registers, so there's an incentive to compile to 64 bit for a bit of extra performance even if the application doesn't need the 64-bit support itself. (The 970 reveals the same number of registers to both 32- and 64-bit applications, and it sports the generous number of registers common to PowerPCs.)
THG noted in the shootout article linked above that since the Opteron has a memory controller on board, it can only be connected to DDR333 RAM. If you wanted to hook an Opteron up to, say, dual channel 128-bit DDR400 RAM, you'd have to either wait for AMD to release a revised Opteron that supported that configuration or purchase a chip that bypasses the onboard memory controller in favor of its own (THG discusses this). On the other hand, each Opteron gets its own bank of RAM, up to 8GB - but that advantage wouldn't really manifest until you got to 4+ CPU systems.
Based on that article, I'm going to speculate that the G5 will still be at the front of the pack when it rolls out - if not in SPEC, than in real-world application benchmarking. If there is something faster on the landscape, it won't cost $3K.
Originally posted by shetline
I just found this Opteron vs. Xeon article: Duel of the Titans: Opteron vs. Xeon
It's a mixed bag of results with no clear overall winner. In these tests, the Opteron system falls behind the Xeon in the kinds of things Apple wants to emphasize, like audio, video, and 3-D processing.
So, SPEC numbers aside, I get the impression that Apple's bake-off might have been just as, if not more impressive against an Opteron-based system.
This article showed also some interesting points : in some task a single opteron 1,8 is slower than the athlon 2400. The opteron is basically an athlon with a different memory controller and a bigger L2 cache (and of course 64 bit integer and SSE2 instructions set). I fear that the athlon 64with only 256 KB of L2 cache will sucks.
Based on that article, I'm going to speculate that the G5 will still be at the front of the pack when it rolls out - if not in SPEC, than in real-world application benchmarking. If there is something faster on the landscape, it won't cost $3K.
Agreed.
Let me also add that this is only the *beginning* for this new line or processor from IBM. I'm sure we'll see additional compiler optimizations and applications developers willing to take advantage of the CPU's features.
I also remember Steve Jobs mentioning that the 970's were built for massive parallelism in mind -- full SMP support. I then remember him mentioning that the powermacs have 2 processors and that it's only the beginning ... or something to that effect. Perhaps someone here can recall what Jobs said about SMP. I may have gotten it wrong.
As far as the Opterons go... I just don't see them hitting the CompUSAs and CircuitCity type outlets any time soon, but considering how much trouble AMD has been having with getting their tech to market, it does make you wonder just how much longer they can keep pushing the envelope.. the same goes for Intel. This CPU war has caused them to push harder and faster in a race to bring the fastest CPUs to market as quickly as possible no matter what the cost. How much further does either company have with respect to pushing the envelope? Remember, Intel was working on the Itanium for a VERY long time. I suspect that they saw the end of the x86 line coming sooner than they would have hoped -- that's probably why they started the project. Now AMD goes and brings out AMD-X86-64 -- x86-64 is something that Intel *didn't* want to have to deal with. I can't imagine how it will pan out for the Wintelon crowd. I don't think they've been thinking about those types of problems.
--
Ed