Asymmetrical Information

Jane - I could be wrong, but I think the new IBM chip is a continuation of the family of "Power PC" chips designed by Apple, Motorola, and IBM. (The new chip is called a "G5". Apple uses G3s and G4s in their other machines.)

As for the switch to 64 bit processing, you are correct. It will take a while for the software to catch up with the hardware. In the meantime, Apple says the new chip will run 32 bit applications in native mode -- so users should still get a substantial speed bump.

So they're making 64 bit machines. So what?

Didn't Mr. Jobs tell all of us for years about the "megahertz myth"?

James

The G5 is the PPC970, which is a cut-down version of IBM's POWER chips (64-bit chips designed to run in big machines). It is not a new version of the G3/G4 etc, despite Apple's naming convention.

All other things being equal, a 64-bit chip will *not* run twice as fast as a 32-bit chip. The only thing the extra 32 bits allow is the use of more memory (32-bit chips are mostly restricted to 4GB of RAM, though there are software workarounds). Since there aren't many consumers who need more than 4GB of RAM, 64-bit chips won't change much of anything for at least a couple more years.

Also, Slashdot has a link to an analysis of the benchmarks that Apple used to say the new machines are faster than top-of-the-line P4s and Xeons. I was shocked, shocked to find gambling^H^H^H^H^H^H^H^H that the benchmark setups were, um, ah, adjusted so that the G5 ran faster and the intel boxes ran slower. According to quotes from Apple's own documents on the benchmarks, the intel machines had performance enhancing features of the CPU (SSE and hyperthreading) turned off. "lies, damn lies, and benchmarks" - Mark Twain, Mod N.

I wonder if the switch to the new Unix-like OS will doom the Mac. Once people realize they can run applications on plain vanilla boxes running Linux, will they still feel the need to stay in the proprietary hardware world Apple has created?

And I think Eric is wrong to suggest that only the
address space is increased. I would expect that a 64-bit chip would have a 64-bit address space and a 64-bit wide data bus. If you are doing full 64-bit computations in the CPU, it would only take 1 fetch to read in a register, not two as in a 32-bit data path processor. Maybe in some applications something like a doubling of speed could occur.

But even though we call them computers, direct numerical computation is seldom what we use them for. The most computationally intensive tasks are often graphics manipulation, which increasingly is done not in the CPU but in specialized processors on the video cards. Smart peripherals are not new to the Mac world: for quite some time there was more memory and processing power in the laser printers connected to Macs than in the Macs themselves.

But take anything I say about Macs with a grain of salt, as I am still smarting from the derision and neglect the computer press has always shown for my first computer love, an Amiga.

1. The Macintosh always ran on a 32-bit platform. The 68K family is 32 bits--the 68000 has a 16 bit external data bus, but that is an implementation detail, not a feature of the architecture, e.g. the 68008 was a 68K implementation with an 8-bit external data bus.

2. MacOS X is built on top of Darwin, but there's a heck of a code there on top for the user interface. If all you want is a command line
or some other X-based interface, then sure, you could run on plain Darwin, *BSD, or some Linux flavor quite fine.

3. You have to keep in mind the difference between the architecture and implementation. Simply because the architecture is "64 bits" doesn't require that the implementation be 64 bits throughout, e.g. a particular implementation could do the 64 bit adds in two parts to reduce the gate count, or a particular implementation could recognize a particular instruction sequence to effectively be a 128 bit add and do it in one step if it had a 128 bit adder (this isn't as likely with RISC architectures like the PowerPC, but on the x86 platform chips often both break down complex ops into multiple simple ops and recognize instruction sequences as a single internal op)

Mark: Linux, you see, can't run any of the really cool Mac software, nor can it run the OS X UI, Quicktime stuff, or any of that... and all of that is what sells to most of Apple's market. (And, of course, it'll be able to run Linux just fine, too, if anyone cares to.)

I don't much care that it's "64 bit" in itself, either (I own a "64 bit" computer already, and it's a doorstop. But that just highlights, of course, that the bit-width doesn't matter much for most things), but the mondo-fast FSB, 8X AGP, and all the various I/O stuff (PCI-X, 800mB/S FireWire, 802.11G) are enough to engage even my jaded geeklust.

Sigivald:

I see your point, and I agree: it is unique software that differentiates the Mac, and I am a little confused by steps by Apple to step away from that uniqueness towards more open systems.

I guess I am a little confused by my friend who swears by his Mac, but runs only Photoshop, Microsoft Word, and Microsoft Internet Explorer. I am not completely sure what the premium he paid over the cost of a run-of-the-mill WinTel box bought him.

Despite my occasional anti-Redmond rumblings, I am a grudging fan of what those economies of scale will buy you.

But I should stop now before taking any more steps towards a Windows/Mac/Linux three front religious war. That way lies madness.

"But Jobs has gambled -- and won -- before, so I wouldn't put my bets down either way just yet."

Well, I would. :)

Jobs "gambles", gets a big PR splash, indications are things are incredible! Then the press wanders off and fails to notice the next year, with the next big splash that, gee, yeah, Apple didn't sell as many machines as predicted, things didn't turn out as lauded, but hey! look! shiny things! Clear cases! Nevermind the decreasing percentages! Nevermind lagging farther and farther behind in speed!

Now, I say this from a sysadmin perspective. Newer's not always better. Faster often isn't. Its the overall _system_ you have to look at. For someone like James Lileks, the lack of speed is made up for by the UI. Personally, the UI on the Mac slows _me_ down, and dislike it. (And if you really want to see me mad, lets discuss the monstrosity that is/was Mac networking, at least, pre-OS X). But that caveat is important, too. I know of 1 OS X machine in our department. Asking around, about 30% of the department used to use one at one time or another. Now 1 person does (who is a UNIX admin, and only because X finally came out).

Jobs gambles, and he wins a few thousand in a big fanfare, but what people are missing is that _overall_ he's losing the $20 per hand/roll/spin... and overall is losing the whole bankwad. That just doesn't make the press as much as the "big wins". (Many of which turn out to be... shall we say.. overhyped...)

Den Beste's got a good take on the Apple situation, IMO. Right now the only place they can expand is laptops.. and the network effects hurt them badly there.

http://www.pbs.org/cringely/pulpit/pulpit19990715.html

Interesting comment on the iMac introduction - and one that didn't get any more press than that... So the sold a ton the first few weeks.. but did they make any money?

Addison

1: it's the fastest computer on the block now - and slashdot has a good discussion of the benchmarks used, how these tests were compiled and the fact the the new macs are faster anyway.

2: the new g5's can address up to 9 gig of memory - which is tremendous if you are playing with video or pictures.

3: the bus is significantly faster, meaning much faster throughout of information - again multimedia applications are huge winners here.

4: they are significantly cheaper (!) than the Intel/Windows based equivalent. and faster. and prettier. and easier to use.

5: I don't come to this site for my computing news. slashdot is a good start...

More info on Edmond Hack's post.

Let's not forget that the Cocoa environment in X descends from NeXT, which ran (runs, I suppose) on Intel CISC processors. Apple may be moving to new hardware in the future.

There is also the possibility that Intel will develop a RISC processor that will run X. Intel has RISC processors, but I think they're used primariliy in networking products.

"Linux, you see, can't run any of the really cool Mac software..."

That may be, but there's an article at Slate about how Linux is about to surpass Apple in market share and may be eating some of Apple's lunch among anti-Windoze computer buyers.

http://slate.msn.com/id/2084727/

Price counts too, and if it does what you want...

anon, the dual-2GHz machine is a nice price-performance match for a similarly-equipped dual AMD Opteron system (which is also 64-bit albeit mainly server-targeted until the Athlon64 comes out in the fall), but the new single-processor Mac workstations are somewhat overpriced compared to similar PC offerings.

Let's wait and see what Apple tries to charge for the XServer line before tooting the horns, though.

Hmmm... lots of incorrect info here. Let's clear up some things:

* A 64-bit chip like the G5 operates on data in 64-bit chunks (as opposed to the 32-bit chips currently used). The difference is important in a couple of different ways: simple transfer of data around the system is faster, it's faster to load and do math on 64-bit numbers (most numbers in typical programs are 16-bit or 32-bit, but there is an increasing trend to double-precision numbers for things like rendering or high-end graphics, etc) and address space (the number of unique positions in memory that the processor can specify - 4 gig with 32-bit processors, 18 billion billion bytes with 64-bit -- not that you could afford that much memory).

* The chip is only as good as the rest of the system and Apple went all out. The gigahertz front-side bus is amazing, they can actually transfer 8 gigbytes a second from RAM to the processor - a huge increase of their old architecture (probably less than half that). There is a bus for both processors in dual processor systems, too, which is a big win. The I/O subsystem is similarly speedy, with very fast paths to disk and graphics systems.

* This particular 64-bit processor can run 32-bit apps natively, without any changes; they built compatability in. If you want to take advantage of the 64-bit address space or use any other 64-bit features, you have to recompile, but developers at the conference this week are having success converting their apps in only a single recompile - call it a half an hour.

* Being able to install up to 8 gig of memory is silly for most of us. We would be swimming in bytes with nothing to show for it. It's a different story for the workstation market, however. If you're doing DNA sequencing or rendering movies over at Pixar, more memory is always a Good Thing, no matter what. As the last decades have shown, things that start out being "only for professionals" tend to slide down and become toys for home users (imagine the home user able to edit movies - a whole DVD worth - all in memory at the same time. We'll be there before long).

* Part of this week's announcements (and I can't remember a computer company ever releasing this many big new things all at once, it's amazing) was a new professional grade video codec designed for their "friends" over at Pixar (hint: Steve Jobs runs both companies). It's a very high-end codec that, unlike MPEG, has no noticable artifacts at pretty good compression rates and no-interframe compression, which makes it much faster to zip back and forth in a movie.

All of this sounds like a laundry list of discontiguous tech toys, but there's an underlying theme that isn't hard to see: all of this is aimed directly at Apple's core constituencies: creative professionals and academics. These folks have been left out in the cold a bit by Apple in the last couple of years, and this is just the sort of stuff to pull them back in.

People who obsess about market share just don't get it. Apple's market share doesn't matter as long as they sell boxes and stay in business. Porsche doesn't have big market share (it must be way less than 1 percent), but they still make the cars and they are still cool. Apple will be fine as long as they continue to "dance with the ones what brung 'em".

A small segment of tasteful home users will also say loyal to the Mac and they gotten their goodies lately too: better game machines, nifty video conferencing (and a neat little camera), better iDisk functionality, iPod improvements and updates, nifty web browser, Apple music, and new ease of use features in the OS.

As far as I can tell, Steve has Apple focused exactly on the right stuff. It would be totally wrong for any other company, but Steve understands Apple and its business perfectly.

rob wrote:

* A 64-bit chip like the G5 operates on data in 64-bit chunks (as opposed to the 32-bit chips currently used). The difference is important in a couple of different ways: simple transfer of data around the system is faster, it's faster to load and do math on 64-bit numbers (most numbers in typical programs are 16-bit or 32-bit, but there is an increasing trend to double-precision numbers for things like rendering or high-end graphics, etc) and address space (the number of unique positions in memory that the processor can specify - 4 gig with 32-bit processors, 18 billion billion bytes with 64-bit -- not that you could afford that much memory).

There is nothing inherently faster about working with 64 bits, it's in how you use them. PCs have been using SDRAM and DDR-SDRAM with a 64-bit wide DATA path for a long time now, but the address-path (which governs the maximum quantity of memory space you can install) was still 32-bit (as well as the core's primary processing hardware). OTOH, the SSE SIMD instruction set can work on 128-bit numbers for certain types of (SSE) operations.

It's also NOT faster to "load and do math on 64-bit numbers" in situations where 32-bit is all that you need.

It's a complex issue but, even as a simplification, it is not really correct to state that 64-bit is generally faster.

"It's a complex issue but, even as a simplification, it is not really correct to state that 64-bit is generally faster."

No, you've got it exactly backwards. While many particular operations aren't any faster on a 64-bit machine, some are (like 64-bit integer math, which includes address calculations in 64-bit apps, having 64-bit registers to hold more when doing string operations, etc). This means that while any specific bit of code might not benefit from a 64-bit word size, some code does and when you mix up a lot that doesn't with some that does, the 64-bit processor will be generally a bit faster.

As a check, we can also say this must very certainly be true or all of the CPU vendors wouldn't be rushing to develop 64-bit CPU's. If there was no advantage, the investment would be wasted. QED, ipso facto, ha-cha-cha.

Rob, I'm happy to defer to you on engineering if you'll defer to me on business. Apple cannot afford to be the porsche of the computer industry. It needs to amortize its R&D costs over a significantly larger customer base than does Porsche. Apple is certainly going to lose money on these new developments for the first year or so they're out, and the number of people who switched to a PC architecture during the long, hard winter of the G4 has severely impacted their market, particularly as the recession has grossly hurt industries that consume creative work, like advertising. This means that, as appealing as the new stations may be, it's going to be a while before they ship in large numbers. Nor can academics especially likely to purchase shiny new $3000 computers on a whim, particularly when the data-driven fields are often already locked in to a Wintel or Unix support architecture. Mac is dominant in the English department, but that doesn't really help.

Which is not to say that this may not help Apple win back the high-end market they were losing. But at these prices, that's kinda dangerous. I'd estimate that the cost of these machines is closer to $5,000 than $3,000, and with Apple's high manufacturing costs, it'll be a long, long time before they can make it up in volume. I don't think they're going to get there, which means they'll have won the battle and lost the war. But I could very well be wrong.

I can't believe the level of scepticism I'm seeing here!

Let me begin by pointing out one thing. In the post above this one, you say "the data-driven fields are often already locked in to a Wintel or Unix support architecture", but fail to notice that OS X is in fact a flavor of Unix! For any experienced Unix hand, there is virtually no switching cost involved in moving over to a new Macintosh. I've ported software from Linux/Unix to OS X, and let me tell you that in most cases, it requires no more than setting a few compiler flags, and going "./configure, build, build tests, build install".

I see hardly any appreciation on here for another extremely important fact, namely that the CPU address space has been extended from 32 to 42 bits, i.e. 4 terabytes. This makes the Macintosh the only currently available mass-market operating system to have done away with this irritating barrier once and for all.

Itanium is not and will never be a mainstream product, while Opteron is targeted firmly at the server market (Athlon 64 is just so much marketing as of present). Even if either processor were aimed at the mainstream, the fact is that there are no consumer-oriented operating systems available from Microsoft to take advantage of them, nor is there even any indication as to when said operating systems might appear.

For those of you who use your computers only for mundane tasks like word-processing, writing email and internet surfing, the move to 64 bits is likely to be of no consequence, but then again, you aren't the sorts of people to need new machines anyway. Those of us who have to deal with large data sets, as is the norm in scientific computing, will be salivating at the prospect of getting one of these machines; finally, an affordable, fast 64-bit Unix machine!

Anyone who thinks "2 GB should be enough for everybody" is simply naive, or ignorant of the history of personal computing. Anyone working in engineering or computational biology will be singing hosanna on hearing of the new Apple machines. 4 years ago, carrying out singular value decompositions on large sparse matrices, I had already hit the ridiculous 2 GB memory barrier that PCs suffer from, necessitating my purchase of a DEC Alpha at 4 times the cost of the top-of-the-range G5. Adding a second processor to that Alpha ballooned the cost up to about $16,000. Even now, when I no longer have to deal with large data sets, my desktop machine requires at least 1.5 GB of RAM to hold the various developer tools I require for day-to-day work.

As for the business argument, I think it is easy for those who know little about computer engineering to overestimate how much the new hardware development will cost Apple. The fact is that Apple has moved more and more towards using off-the-shelf components and industry standards, rather than doing in-house development. PCI-X is an industry standard, the G5 is simply an IBM Power4 with less cache and a SIMD unit added, while the fast CPU interconnects are Hypertransport links developed by AMD. The standard OS X compiler suite is GCC, and even the underlying operating system is based on BSD 5.0 (which is as fine a base to build on as they come). In light of these facts, where exactly are the onerous financial burdens supposed to come from?

No well-informed observer can dispute the conclusion that the new machines have just given Apple its' best chance to truly compete for marketshare in years. All manufacturers of Unix-based scientific workstations ought to be worried, and Sun Microsystems especially so.

Jane, you know worlds and worlds more about economics than I do, there's no way I would go there.

BUT, I've been following Apple closely for about twenty years and there are a couple of things that should be brought into the discussion if we're going to it right: one million iPods, five million songs, new video camera, PC software for iTunes, etc.

Apple isn't just about Macs any more. All of these are at least potentially high-margin sales that feed directly into the bottom line (well, I don't know about the music store, it's hard to estimate Apple's costs on that).

As for the costs of the new machines, I can't say. I do know that folks on Slashdot have said that the G5 chip itself is considerably cheaper than equivalent Pentiums (maybe IBM is doing a little loss-leader here for their own purposes, I don't know). The rest of it is just a slick case (interestingly, it's a lot cheaper to bend aluminum than to create plastic moulds) and a lot of industry-standard innards. My guess is that Apple is bringing these machines in at very low margins, but I really doubt that they're taking a $1000 loss per box. Apple tried giving up margins for market share a couple of CEO's back and it did them no good at all. I don't think Steve has ever sold a product at a loss in the past.

It's possible that you over-estimate their R&D costs, too. The acquisition of NeXT (and, really, it was the other way around, but no one noticed) is where nearly all of their current OS R&D came from. When you program the Mac these days, most of the objects you use have names that start with "NS" as in "NSEvent" or "NSApplication" and, of course, that's NS as in NeXT Step. Apple continues to tweak it around and add stuff, but the big base of underlying goop came from NeXT, BSD Unix (Berkeley) and the Mach kernel (Carnegie Mellon). Microsoft is struggling to get to where Apple is (64-bit, rendered desktop, snazzy UI, etc) in their Longhorn release, due out in 2005. Apple skipped all that pain by letting itself be assimilated by NeXT and embracing open source. My guess is that Apple will have spent a tenth of what Microsoft will spend to achieve the same ends. If Apple had continued down the path it was on (remember Copeland, the OS that never was?), they would have crashed and burned. Apple can't take Microsoft on in straight R&D, they just don't have the cash. Instead, they take short cuts. Can they keep finding short cuts? I don't know.

There's one other reason that I'm somewhat (but not toweringly) optimistic about Apple. Since about 1984 there has been an endless stream of "analysts", journalists and pundits of every stripe who have predicted the imminent demise of Apple. So many people have been so wrong so many times that there must be something about Apple that defies easy analysis. I can't think of another company that has been so relentlessly downplayed, yet continued to function, year after year.

I think Steve keeps Apple right on the edge because he likes it that way.

As I say, I don't count Apple out. But selling a major product at a loss is not a winning strategy regardless of whether other lines are profitable. . . unless it acts as a loss-leader for the smaller products, but it's a hell of a loss for not much lead.

rob wrote:

* The chip is only as good as the rest of the system and Apple went all out. The gigahertz front-side bus is amazing, they can actually transfer 8 gigbytes a second from RAM to the processor - a huge increase of their old architecture (probably less than half that). There is a bus for both processors in dual processor systems, too, which is a big win. The I/O subsystem is similarly speedy, with very fast paths to disk and graphics systems.

Umm, doesn't that just move the bottleneck? DDR bandwidth isn't at 8 GB/s and the DDR SDRAM is still nowhere near 1Ghz. Mushkin's top offering manages 433Mhz effective(216ish DDR), 3.5 Gb/s bandwidth.

Oops, I guess it's possible they redesigned the memory subsystem to work with a couple of memory controllers in parallel to side-step the bottleneck.

I don't understand why Jane thinks that Apple is selling these machines at a loss. They aren't. This is the high end of the market, the margins here are very good per box. IBM developed the chip, not Apple, and the 970 is an adaptation of another chip they sell. As has been noted, Apple has moved more towards off-the-shelf parts than ever in its history with these machines. Panther is simply a new release of OS X, part of its planned evolution, and its development costs are spread over the entire product line (including iPods, which sport a mini-version of the operating system). By all accounts I've seen these machines ought be bottom line winners. What's your evidence for thinking Apple is selling them at a loss?

"I see your point, and I agree: it is unique software that differentiates the Mac, and I am a little confused by steps by Apple to step away from that uniqueness towards more open systems."
Mark Woodworth

The consensus in my Comp. Sci. dept. is that Mac is pretty much going to dominate the desktop side of Unix. Mac works without needing to jack around all the time with settings, but simultaneously gives me the Unix stability and features that I want to be able to handle work that I bring home. The Open Source movement is a built in development constituency and Apples own apps are distinctive and innovative enough to remain separate in the open source (and commercial) crowd.

Apples main weakness has always been the hardware vs. price. If they can bridge the performance gap they're looking pretty good.

64-bit will be nice for extremely large hard drives, a little bit of video processing, and some miscellaneous things, but it mostly doesn't matter. It doesn't increase the flow rate when you double the diameter on a hose, does it? It's mostly about the pump.

The concept of people using macs as servers is pretty funny, I must say.

"Itanium is not and will never be a mainstream product, while Opteron is targeted firmly at the server market (Athlon 64 is just so much marketing as of present). Even if either processor were aimed at the mainstream, the fact is that there are no consumer-oriented operating systems available from Microsoft to take advantage of them, nor is there even any indication as to when said operating systems might appear."
That's funny... I could swear I had read something about a 64-bit windows version... Oh yes, I did, gee, why didn't I think to look on Microsoft's very own site?

Microsoft has now introduced Windows® XP 64-Bit Edition Version 2003, with support for the Itanium 2 processor, to meet the demands of technical workstation users who require large amounts of memory and floating point performance in areas such as mechanical design and analysis, 3D animation, video editing and composition, and scientific and high-performance computing applications.

Like OSX 64-bit, most applications require only a simple recompile to work on Win64.

Then of course there are the 64-bit Linux versions...

As for your little canard about Opteron and Itanium2 not being aimed at the consumer market... neither BOXX (which sells "digital content creation systems for the design, entertainment and digital film industries ... featuring dual AMD Opteron" processors) nor HP (which sells dual Itanium2 workstations) seem to agree with you.

Incidentally, you are correct about Itanium not ever being a mainstream product... it was essentially a proof of concept and released in order to allow OEMs to develop around it. Itanium2 and its descendants are different, however.

Finally, PCs are not limited to 2GB of memory, and have not been for some time. 4GB is the "normal" limit (since at least NT 4.0, AFAIK), and there are ways to use even more than that -- with a 32-bit chip and operating system.

"I don't understand why Jane thinks that Apple is selling these machines at a loss. They aren't."
Could it be because Apple's quarterly returns consistently show a loss before the profits from cash investments are added in?

That may change if IBM can ramp their yields high enough and keep the R&D cash flowing; otherwise it will be the 500MHz G4 all over again.

I don't want to leave any false impressions; I certainly think the new G5s are a breath of fresh air technically; definitely competetive with current best-of-breed x86 PCs, though not IMO faster across the board. Whether they will have any effect on Apple (re)becoming a non-niche player is another story entirely.

Rob wrote:

No, you've got it exactly backwards. While many particular operations aren't any faster on a 64-bit machine, some are (like 64-bit integer math, which includes address calculations in 64-bit apps, having 64-bit registers to hold more when doing string operations, etc). This means that while any specific bit of code might not benefit from a 64-bit word size, some code does and when you mix up a lot that doesn't with some that does, the 64-bit processor will be generally a bit faster.

And can be countermanded by situations where you don't need 64-bit numbers but have to work with them anyway, increasing the memory overhead vis-a-vis a similar 32-bit system operating at the same clock speed, for a small performance drop. There are still tradeoffs.

As a check, we can also say this must very certainly be true or all of the CPU vendors wouldn't be rushing to develop 64-bit CPU's. If there was no advantage, the investment would be wasted. QED, ipso facto, ha-cha-cha.

Cute, and logically fallacious. "Faster" does not have to be the only justification for moving to 64-bitness -- you realize this?

One of the primary advantages is getting rid of the memory barrier. I believe somebody earlier referred to ways of getting more than 4GB of memory on 32-bit systems, but this is mainly a work-around approach with associated trade-offs. The way it is done is through Intel's "PAE," which uses a 36-bit scheme to potentially address up to 64GB of RAM -- BUT it is limited in that the application itself (as compared to just the OS) has to be aware of the feature, AND it can only access the additional space in 4GB chunks. Each process running under that app can grab a 4GB chunk of the extended memory space for itself, but the system can't use the additional memory as a "flat" space, i.e., any starting point accessible in any quantity for any needed purpose.

Also, you take a performance penalty for using PAE (I've heard up to 30%, but haven't personally seen any hard tests), which further limits its usability. The primary application where it has been beneficial is in low-end database servers, where cacheing disk contents into RAM (for a notable performance boost compared to accesssing directly from disk every time a request is made) meshes nicely with allocation in 4GB chunks, and provides a performance gain in excess of the penalty.

Why 64-bit in the consumer sector? When you consider that an OS frequently will reserve up to 1GB of space for its own activity (Typical for a Windows or Linux machine running under very high load -- you may not have 4GB of physical RAM installed in a machine but the virtual memory can still go that high) and you have additonal transaction overhead for use of the remaining 3GB, the maximum amount of space that remains on a typical system for use by applications is about 2.5-2.7GB. Anymore it is not terribly difficult to burn off that much space with an advanced photoshop session.

That is going to be the single greatest gain from 64 bits. Other forms of "faster" are going to be dependent on the application at hand, and on how quickly the industry transitions over to 64-bit software.

"It doesn't increase the flow rate when you double the diameter on a hose, does it? It's mostly about the pump."

Ummm... then why don't I see fire trucks driving around with 1" garden hoses?

If the pump can't pump enough water, then a bigger hose won't do anything. But if it can produce the volume, a bigger hose will certainly allow it to flow. It's all about chasing bottlenecks - and from the look of things, Apple has done a pretty good job of chasing them with this beast. If the applications demos are even close to accurate, this thing has hella bandwidth...

So it's amateur hour now, is it?

"That's funny... I could swear I had read something about a 64-bit windows version... Oh yes, I did, gee, why didn't I think to look on Microsoft's very own site?" ...

Have you ever seen a copy of Windows XP for Itanium 2 on sale anywhere? Have you heard from any vendors expecting to receive shipments anytime soon? I suspect not ...

"As for your little canard about Opteron and Itanium2 not being aimed at the consumer market... neither BOXX (which sells "digital content creation systems for the design, entertainment and digital film industries ... featuring dual AMD Opteron" processors) nor HP (which sells dual Itanium2 workstations) seem to agree with you."

Sigh. Since when has "BOXX" (whoever they are) been any sort of mainstream vendor? Look, I'm not just making this stuff up; AMD themselves have said that Opteron is meant for the server market, with Athlon64 to be targeted at the desktop. What is more, there aren't even any chipsets available for Opteron that support AGP! How is that supposed to fit in with your claims? Who'd buy a machine without an AGP slot as a desktop?

As for HP selling Itanium 2 machines - have you actually tried pricing one of these machines? And how many Itaniums of any generation do you suppose have been sold to date? Hint: AMD plans to overtake that total in the first month of Athlon64 sales. So much for a "mainstream" chip. Heck, even Intel's developer roadmaps and press releases contradict your contention - they've clearly stated that they see no need to introduce a mass-market 64-bit chip in the near future, and that the successor to the Pentium 4 will also be a 32-bit chip.

Then there is the issue of software availability to consider. If you actually think that all there is to porting to Itanium is a simple recompile, either you've been writing some very clean and forward-looking C++, or (more likely) you don't know a damn thing about Windows software development. For any codebase that's more than 2 years old, there's going to be a painful effort going through everything to fix problems like pointers being cast to int, ULONG or DWORD, or code that assumes LPARAM and LRESULT will always be 32 bits.

Then there's another obstacle to Itanium 2's success - the fact that IA32 code runs like a dog under emulation mode. If you think every developer out there is going to rush to buy new compilers just to target a platform with near-zero market penetration, you've got to be smoking some strong stuff.

"Finally, PCs are not limited to 2GB of memory, and have not been for some time. 4GB is the "normal" limit (since at least NT 4.0, AFAIK), and there are ways to use even more than that -- with a 32-bit chip and operating system."

You have it completely wrong. You seem to think that because a chip can address 32 bits of memory, applications can make full use of the entire memory range. Unfortunately, things just don't work that way! Modern operating systems that make use of virtually mapped memory have to reserve some of the address space for their own private uses, so that programs can do things like accessing graphics memory by writing to a virtual address, and so forth. As a result, the top half of the memory space usually isn't available for applications to use.

I have my copy of "Inside Windows 2000" in front of me, as well as the complete Platform SDK (available at MSDN), and both sources clearly say that Windows 2000 Pro/XP can only make use of 2 GB under normal conditions. At best one can squeeze in 3 GB, but this requires the OS itself to make do with only 1 GB of address space, not a good idea at all under normal circumstances.

IA32 actually supports 36 bits, using Physical Address Extensions (PAE), but it's a horrible, extremely slow, complicated kludge, and is only an option if you're willing to shell out for Windows 2000 Advanced Server or Datacenter Edition, scarcely the ideal choices for desktop users. Even with these systems, individual applications won't be able to use more than 4 GB of virtual address space, and the chore of manually handling memory paging under PAE makes this an unrealistic option for run of the mill software developers. I don't know of any products that use PAE other than DBMSs.

To sum up, you don't really seem to know much about computers, beyond what the average fanboy picks up on Slashdot. A little more restraint on your part is in order.

"Ummm... then why don't I see fire trucks driving around with 1" garden hoses?"

Because the friction/whatever on forcing that much water through a 1" hose is too high. It's easier just to scale up the size of the pump chamber than the rotational speed, too.

And on the "fixing bottlenecks" angle - memory access time is a lot more critical than memory bandwidth, and 64 bits doesn't change your access time at all.

"memory access time is a lot more critical than memory bandwidth, and 64 bits doesn't change your access time at all."

The reality is more complicated, as it depends on how much locality there is in the calls that are made to memory. Highly local, and therefore predictable, memory access (as is usually the case with high-speed video) is relatively immune to the effects of latency, as once the initial fetch is done, the latency will be masked by pipelining. The same is true of dense matrix calculations - we know we'll be working on the next row, so we can request it be fetched ahead of time.

With regards to applications with high nonlocality of reference, I don't see that Apple could have done better than they have here. Dual-channel DDR running at 400 Mhz gives you the best of both worlds in terms of latency and throughput, given the state of the art.

What mustn't be overlooked is the dramatic increase in FSB bandwidth, which means this chip can take full advantage of the fast memory at its' disposal, unlike the G4. Also, bandwidth should scale linearly with the number of processrs, thanks to HyperTransport and the point-to-point system controller, ensuring much better scalability than with Xeon boxes.

As for the 64-bit address space, you're right - the only benefit is in terms of addressable memory, not bandwidth, but this really is a major benefit.

All in all, Apple seems to have come up with a very clean, well engineered new machine.

Wow...only one comment that mentioned Sun?

Apple's are shiny and they taste good.

Sure, there's a bit of a gain in memory bandwidth from 64-bits, but you could get the same effect by just doubling the clock rate on a 32-bit bus. Yet what you see in the press (and here) implies doubling the bit size magically transforms it into a 100% faster processor.

Still not as fast as intel, and costs way more.

So it's amateur hour now, is it?

Must be.

Have you ever seen a copy of Windows XP for Itanium 2 on sale anywhere? Have you heard from any vendors expecting to receive shipments anytime soon? I suspect not ...

Look

What is more, there aren't even any chipsets available for Opteron that support AGP! How is that supposed to fit in with your claims? Who'd buy a machine without an AGP slot as a desktop?

Look. AMD's AGP tunnel has been available since Opteron's launch. If you knew what you were talking about, you would have said there were no motherboards that included it, but as you can see, that's false too. Margins are highest in the server market, so that is targeted first, but as the supply of Opterons increases the workstation market will continue to get more attention.

This makes the Macintosh the only currently available mass-market operating system to have done away with this irritating barrier once and for all.

You mean besides Windows and Linux. lol

Itanium is not and will never be a mainstream product, while Opteron is targeted firmly at the server market (Athlon 64 is just so much marketing as of present)

Macintoshes will never be mainstream either :) Opterons start at below $300 and can be aimed at whatever market a vendor wants to target them at, and the Athlon 64 will be released in August - the same month Apple's G5 boxes will ship. That being said, you can get a fine dual processor Opteron box for $2000 right now.

Those of us who have to deal with large data sets, as is the norm in scientific computing, will be salivating at the prospect of getting one of these machines; finally, an affordable, fast 64-bit Unix machine!

Look. All dual Opteron, all below $3k, all available NOW, unlike Apple's. No AGP, but scientific computing doesn't require it. Need a big name vendor? Go to Polywell.

All manufacturers of Unix-based scientific workstations ought to be worried, and Sun Microsystems especially so.

You're right. Itanium will kill Sparc in the scientific computation market, and probably everyone else too.

For the AGP Opteron system, the link was broken. See here.

As pointed out by boo, Abiola is an idiot. Windows has offered 64-bit versions since the Itanium went into test release; so has Linux.

This kind of ignorance helps contribute to the common opinion that Mac zealots really don't know anything, which is sad, because the platform itself really is kind of cool.

Why do I think that Apple is selling this at a loss? Because they have, from what I can tell, spent the last 2-3 years developing this thing, which is a major undertaking. IBM may be selling their 970's at a loss in order to screw Microsoft, but it seems somewhat unlikely to me. And if Dell can't bring in their high-end workstations at under $3,000, it seems vanishingly unlikely to me that Apple can. Even if Dell's margins are $2,000 a workstation, Apple's much smaller production runs over which to amortize their R&D make it likely that they are taking a loss at the price points they're showing.

As for "THere ain't no 64-bit windows", Abiola, you're assuming that though Apple has changed, Microsoft won't. Historically, a very bad assumption to make -- just ask Netscape investors. How much do you want to bet there's a 64-bit Windows workstation out there competing with Apple in under a year?

"As pointed out by boo, Abiola is an idiot."

Is that the best you can do in terms of cheap insults? That's the thing about anonymity; it lets gutless punks like you say the sorts of things they'd never dare say face-to-face.

"Windows has offered 64-bit versions since the Itanium went into test release; so has Linux."

Tell me something I don't know already. My point, boo's link notwithstanding, was that 64-bit Windows isn't widely available. You can get it with overpriced Itanium 2 machines (look on HP's website), but you don't exactly see it in your local Computer Shopper, do you?

Come September, the only mass market, consumer oriented 64-bit platform will be the Macintosh G5, and if things go right for AMD, a beta port of Windows Server 2003 will be released by Microsoft at the same time. Athlon64 will run 32-bit windows fine, but what is the gain in doing so? Might as well buy a Xeon then.

As for Linux on Itanium, here's a newsflash for you: I've been using 64-bit Linux for various platforms for years, but there is no way that it can be considered "mainstream" in any normal usage of that term. You will not see 64-bit Linux on Itanium 2 in any consumer distribution channel at any time in the near future.

Finally, if you want to be better informed about Intel's own thoughts on the future of the Itanium platform, you could do worse than read this interview on CNET.

"How much do you want to bet there's a 64-bit Windows workstation out there competing with Apple in under a year?"

I'm sure there now will be, but this is precisely why I'm glad Apple is back in the game - it puts pressure on Microsoft not to sit back and coast along on License 6 revenues.

As I've said, given the number of COTS components being used in the new G5s, I doubt that the production costs are anywhere as high as you fear they will be, but this isn't really something that can be settled one way or another until we see Apple's quarterly results in December.

Apple's pricing does serve to put real pressure on Sun Microsystems; just look at the prices on the Sun Blade 2000 workstations, or even HP's Itanium 2 machines, and you'll see why I'm excited.

Jane,

As for just how cheap these boxes are, I think you're off a bit. Here's the opinion from Ars Technica, which is as unbiased a site about Apple as I can find:

"All that having been said, I still have reservations about the prices of the machines. I had personally hoped that they'd be priced at about half what they are (yes, I know, it was never going to happen), and I'd have no complaints at all if they were $500 cheaper. But with the bottom-end G5 coming in at $2,000... well, these really are pro machines at pro prices. I know Mac users hate to hear this, but non-pro consumers who aren't in the 64-bit workstation market can get serious PC hardware at those prices, especially if you either build it yourself or go with one of the smaller vendors like iBuyPower."

And I don't agree that they've been working on this box for the last 2-3 years or that it has cost them any more to develop it than any other new product they bring out. But I can't prove it, at least not easily here. So let's just let the real proof come forward as Apple releases their quarterly numbers. They are quite good about giving sales and margin information on their product lines.

Why not revisit this in six months and we'll see? For that matter, why not revisit some of the other disagreements from your readers as the facts replace all our bull****. I can think of a couple I'd like to see....

Abiola wrote:

Come September, the only mass market, consumer oriented 64-bit platform will be the Macintosh G5, and if things go right for AMD, a beta port of Windows Server 2003 will be released by Microsoft at the same time. Athlon64 will run 32-bit windows fine, but what is the gain in doing so? Might as well buy a Xeon then.

Well, the general idea is that with Athlon64, your 64-bit upgrade path is software only, with the inital hardware upgrade staying competitive. In addition, it might be possible to make partial 64-bit moves, with each app being moved to 64-bit as they get ported.

Also, there are questions coming out on how much the benchmarks were...tweaked. The SPEC scores for the PC don't tend to jive with what other sites have seens for their benchmarking and there are guys analysing the fine print that basically says the deck was somewhat stacked in Apple's favor. The memory bandwidth certainly is nice, though, even if it can't be fully stressed at the moment.

All that said, the new Mac is a solid professional workstation. I'm less sanguine about it's position in the mainstream consumer market. $2000 is a lot for a home PC, even if it does have all the new whiz-bang stuff. I agree with the guys at Ars, the consumer test will be when the G5 moves into the iMacs.

Just one more comment and I'm out of this thread:

People like to compare Mac prices to knockoff PC prices all the time to show how much cheaper PC hardware can be. It's true, Macs are more expensive.

The story is deeper than that, though. Apple's products are, for the most part, much higher quality. Apple wins over at consumer reports all the time in terms of out-of-the-box problems and customer service. There's no mystery, they're a premium brand. You can get coffee a LOT cheaper than a Starbucks, but people (not me... but that's another story) still go to Starbucks because they want the high-quality products that Starbucks produces. Other people think it's stupid to spend four bucks on a cup of coffee and stay away. Both sides are right, of course.

Here's a simple example of quality. Last fall, I got my wife an iBook. It worked fine out of the box and there were no surprises. A couple of days later, at night, we noticed it across the room. She had closed the lid, which puts it to sleep. A white LED on the front was pulsing slowly, from all the way off to all the way on. Instead of simply flashing at us (and my wife hates things that flash at her, whether it's the LED's on a clock or the 12:00 on the VCR), it was slowly pulsing, as if it were breathing. We both thought it was the cutest thing we'd seen all day.

Now, this is a very, very small thing. It certainly had nothing to do with my purchasing decision (I didn't even know about it). Still, I've used Apple's products for years and this sort of thing happens all the time and THAT has a lot to do with my purchasing decision. Now, this sort of thing isn't important to everyone (any more than a finely crafted latte is), but it IS important to ME. In certain categories, I like to buy things (and pay extra to a certain extent) that are very finely crafted. Some people get off on insanely great kitchen gadgets, I like high-quality computers and hawaiian shirts. Go figure.

Apple understands this perfectly well. They know who we are and they keep making products for us. For people who DON'T care about this sort of thing in this category, Macs don't make sense. We're both right, you see.

Rob has it right.

I used to be a Windows user, but I sold off my last Windows machine in '97 (long before the Switcher campaign).

I could get a hotshot Windows box for a lot less than $2K (and I am perfectly capable of building it myself). Instead, I will pay the $2K for a G5. I'll wait until next spring, when my budget has an upgrade to my main computer already scheduled. I've only budgeted $1800, but I can flex. :)

I've kind of reconciled myself to owning an orphaned computer, since Apple has been doomed for so very very long.

i'm with you, gary. while i'm now using a dell 2.something ghz xp box at the office, it doesn't frankly feel any zippier than my 450g3 at home. mind you, i only surf and email. but i've had macs since 87...

Jane,
can you elucidate your economic argument a bit? Surely the R&D is sunk costs; a great deal of it is also IBM's.

With all due respect to your economic chops, I'd back Fred Anderson over you on price setting for Macs.

One reason these are so price competitive is that IBM is charging much less for the G5's than Intel et al charge for their 64-bit chips. That the G4 & G5 were designed to work well in symmetric multiprocessing, whereas the P4 doesn't do this (you need a costlier, less common Xeon) also gives an edge.

The 64-bit issue is also somewhat moot, as the Altivec vector unit in the G4 and G5 has been capable of 128-bit instructions for quite some time, and the OS and 3rd party applications already take advantage of this. The extra address space is nice, but the additional on-chip execution units do not slow down existing code in anything like the same way that what is effectively running x86 emulation does for the newer Intel chips.

"Surely the R&D is sunk costs"

Slight misunderstanding here about the concept of "sunk costs". These are costs that are already water under the bridge, so to speak, and should not affect your future decision making. It has nothing to do with how you end up paying for said costs.

Example: had Apple spent $100 million in R&D on a top secret project to wring the last bit of performance out of G4s before they knew they would have access to G5s, those would be sunk costs - Apple would have to admit that it was a bad investment and get on with life, rather than throwing more money into the project. But it would still effect their bottom line - just because costs are "sunk" doesn't mean you don't have to pay them.

What Jane seems to be saying is that in her opinion, Apple invested so much on these new machines that it will never make back its money at $3000 per. I doubt this, but absent a look at their speadsheets neither she nor I really knows one way or another...

Theclaim that Apple has put too much into R&D to be able to make money on the new Macs at the prices that have been announced sounds a bit presumtuous to me.

Apple has rarely been shy about charging a premium for their machines, especially their top of the line models. I doubt that pricing has been set to a level where they will take a loss.

Here's jane's argument...

Why do I think that Apple is selling this at a loss? Because they have, from what I can tell, spent the last 2-3 years developing this thing, which is a major undertaking.

Also consider this question: is 2-3 years an unusual amount of time for to prepare a new generation of processors and the computers that will use them ? I don't know but it doesn't seem too excessive to me. As a point of comparison just look how long it took to get OS X itself from being announced in 1999 to where a usable version (10.1, 10.0 was a glorified beta version, IMHO) came out in 2002.

Note how each generation of Windows, or Intel processors has often taken a long time to be developed, often with many delays along the way.

It also could be that R&D costs may be lowered or benefits from it increased by the uniqueness of Apple's situation with the Mac. They may have certain advantages in that they design both the hardware and operating system. If new hardware demands a change in the O.S., they don't have to go begging to Microsoft to make it possible. They can tailor their hardware to get the most out of their O.S, and vice versa.

Case in point: The cooling system (taking the temperature of the various zones and determining the needed fan speed) is controlled by the O.S. (if I understand correctly). Windows machines may have this in the future, but Apple was able to do sooner since they don't have to go through the red tape of having coordinate accross multiple companies' R&D departments.

This may also be a key to Apple being able to tighten prices without as much damage to the profit margin. Dell has relativly less control over the cost of their machines, they are dependent on their suppliers , who each have their own R&D costs to make back. Dell for the most part is only able to impove their prices and profit margins only by benefit of the scale of their operations.

Apple may not have as much in the way of volume as a Dell, but the fact that much of the technology is developed "in house" cuts out a lot of "middle men" that add to the cost.

Apple as of late has made a lot of use out of open source software, and open standards. Apple, unlike Microsoft, doesn't suffer from the "Not Invented Here" syndrome. This has essentially given them a degree of cost efficiency in developing parts of OS X, and many of their applications (most notably, their new web browser, Safari). Microsoft is more likly to buy out a smaller software developer than to make use of open source software.

In summation, Jane's concerns about Apple losing money on the new G5 Macs, are based on too little facts and a big supposition about time equaling money at a flat rate.

Economics is about choices, their costs, and their benefits. Apple's control over more of the product design gives them a wider range of choices to follow, giving them more opportunities to reduce costs, and increase the benefits.

Jane,

Just out of curiosity, did you actually read Apple's specs on the machines before commenting on them? Had you, even at a quick skim, I can't imagine you wouldn't have noticed that the new processor runs 32 bit applications natively and suffers no performance hit as a result. Your conclusions on previous Mac processors were also inaccurate, as is the majority of the rest of your argument.

Almost competitive with PC's? The top of the line machine is a thousand bucks cheaper than a comparable Dell, and it blows the doors off it both in spec benchmarks and application demos.

Apple selling the new machines at a loss? Can you imagine any CEO of a company in Apple's position doing that, least of all Steve Jobs? Another groundless claim based on no factual evidence. Apple has historically demanded a premium for their hardware, since day one. Why would they suddenly abandon that, now that they have a best-in-class product?

If you have something personal against Apple or the people who use their products, like some people notably do, I wish you'd just come out and say it. You're far too intelligent a person to make groundless claims like these.

Edmund,

It may interest you to know that Dell also disables hyperthreading for their processor benchmarks. Apparently it yeilds higher results.

Also, SSE2 most certainly was enabled in Apple's tests. The -mfpmath=sse flag enables SSE2 within GCC. The author of the "debunking" article was apparently ignorant of this fact.

Moving a typical application from 32-bit to 64-bit is more likely to make it slower than faster. Integer arithmetic is rarely the bottleneck in a typical app, it's ususally things like floating point math, pointer manipulation, and memory management. Floating point math is essentially unaffected by moving to a 64-bit architecture. But all pointers get bigger. This means that for many apps, all their data structures get bigger and take more memory. So the whole app takes more memory, making it more likely to swap, and you can fit less useful data in a single cache line. In general, this is going to slow things down.

The real rationale for 64-bit computing is not so much performance as address space size. Some applications need to operate on a *huge* data set, and in that case, the benefits of being able to address it all directly, and keep it all in RAM instead of on disk, outweight the space consumption of 64-bit pointers. Typically these will be large databases or scientific apps, but I can imagine pro video apps getting into this territory as well.

$3K for the top of the line machine is actually rather moderate for Apple. I remember the IIfx going for something like $6K. The top of the line 604 based machine was $3600.

And Apple NEVER sells at a loss. That's just silly, Jane.

Oh, yeah, and 2-3 years is a perfectly normal development cycle for Apple for a NEW machine.

Comments are Closed.