|
Who says Apple is Doomed? I DO! |
Apple's demise
|
|
Ok, so where are the SMP Spec results? How have the compilers and operating system been enhanced to take advantage of SMP? What applications (other than fricking Photoshop) will be available that can take advantage of this second processor? How exactly is a stupid cooperatively multitasking operating system supposed to use a completetly asynchronous resource like a whole other CPU? Where the *F* is OSX?!?! Did Apple work with any Linux vendors to make sure they could take advantage of the second processor?
Update: Just as a point of clarification, the "dual G4's" are not capable of "SMP" using OS 9. It is strictly a client-slave arrangement (much like a graphics cards plays a subserviant role to the main CPU.) Although OS X beta has been released, (SMP based) benchmark results are not forthcoming (big surprise ...)
Tech-Report has a nice little reaction to the whole thing.
| GMP Performance | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
The page also shows estimated scores for the Motorola G4-500 as being basically identical to the G3. This makes perfect sense since the library does not use any of AltiVec or MMX or SSE or 3DNow!
This benchmark tests raw integer ALU speed, roughly in "operations per second" normallized by the theoretical performance of an Alpha 21264-750Mhz (it should be noted that the real Alpha tests do not show scores of 100.)
Update: (12/21/00) there are new scores with actual runs for the G3 (as opposed to projections):
| GMP Performance | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
The G4 did post some gains (probably due to over projections of the Alpha results) over the previous version of the benchmark but both x86 processors gained even more. I mean holy cow look at that Athlon! Its basically the fastest processor on earth for multiprecision adds and multiprecision shifts. (Note that the IE browser has difficulty with table rendering -- the Alpha's 93.5 score in lshift is *lower* than the Athlon's 100.)
The mac consistently loses to the PC as shown by the chart below (higher is better).
| STREAM Performance | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Notice that you cannot divide out the Mhz here to give some sort of dellusional number to represent how fast the architecture is. The benchmark is very memory intensive, and as such is not very clock rate dependent. Of course you might complain that we are really comparing 133Mhz RAM versus 100Mhz RAM ... but the fact that it has available for some time in the x86 market and is not in the PPC market is kind of the whole point.
Update: (06/16/01) Some updated results:
| STREAM Performance | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Of course the real story here is the dramatic improvements made by Intel. Focussing just on the Apple results, we see that they are still playing catch up to results from 12 months earlier -- a standard Athlon now has nearly 2x the performance.
This is easily reproduced (with Quicktime alone -- not any other PC application).
They wrote the following:
|
Oh how I enjoy people's estimation of my age and intelligence (which is commonly followed up by people correcting my spelling mistakes.) Anyways, the last sentence is the culprit I was looking for. Their David Koresh is outright telling his fanatic following to come throw stones at me. Notice that they clearly didn't bother to read past the first few entries of this page. I'll just tersly respond to this:
Here's the letter (slightly modified for technical reasons) from my informant who blew the whistle on mactoday.com (to the author: if you would like your name published, just email me and I will):
|
I must say, I definately appreciated this letter. Its refreshing to know that even among Mac users, there are reasonable, clear headed people who will at least make a reasonable effort to understand what I've written. I'm going to respond openly here, just with the intention of clearing up what I may not have made as clear as I probably should have:
The "Gates bought our company" comment was actually not written by me. Some time ago when Apple did a complete redesign of their web site, they had a three section image which is shown on the top right of this web page (the cookie, the shopping cart and the screwdriver). At about this time, Microsoft surprised many be investing $150M into Apple. Shortly thereafter on rec.humor.funny a joke was posted with the three captions shown. I thought it was so hilarious I compositied and posted the image here. I suspect either my use of the image as shown, or the rec.humor.funny post may have motivated Apple to change the images on their website shortly thereafter. Yeah, its old. If I get inspired by something else maybe I will go change it. Any suggestions?
Of course, I'm not attached to my os.
Yeah the ihateapple.com is quite militant. But I justify it as a counter balance to sites like mactoday.com which are just as militant in the opposite direction. The truth is that I only added it a few days ago as a kind of serrendipidous (sp?) side effect of searching for the cult that was encouraging the flames I've received recently.
Indeed every once in a while, I get the urge to clean up this anti-Apple web page like my anti-Microsoft page. But I consider this page to be a very part time kind of thing. I just can't justify the effort to really clean it up. This may seem contradictory considering how large this page is, but you have to understand that it was written up in fits of energy over a very long period of time. I have almost certainly put 10 times the work into my anti-Microsoft page.
As to my C/C++ discussion ... well unfortunately a detailed analysis of languages takes kind of an enormous amount of work. Its also a religious kind of thing. No matter how well I did such an analysis, you get morons who are one side of the religious fence who will just feel like flaming me without hearing me out (sound familliar?)
Interestingly, recently I did a mini-analysis of Scheme versus C (just analyzing a single benchmark proposed by proponents of Scheme). There was a claim that Scheme was 21 times faster than C -- I showed that even theoretically it could not be more than 3 times, and that in the example given it was highly unlikely that it would even beat the C code performance under comparable implementation conditions. I put many hours of work into my analysis, but the Scheme proponent who I was aiming this at (Bruce Hoult, who is a PowerPC proponent as well -- go figure) decided that, even without reading through it, that my analysis was flawed. His wording was meant to attack my credibility, which he is fond of doing on comp.arch.
I still want to do it (in the hopes that the *entire* audience wouldn't be a bunch of Bruce Hoult's), as I feel the C and C++ war is an unresolved question in the minds of many. I just need the time to do it and a big boost of energy.
But in this sea of stupidity is finally the voice of reason.
Read all about "Processor Wars" on holymac.com
I am guessing that Paul Demone's article has caused a stir on some forum somewhere, a raging flame war has ensued and my web page has been cited. Anyone feel like confirming or denying it?
My other theory which is more conspiratorial, is that some more organized Mac advocates have decided to target me. Some organization is telling their cult to mail in to me to tell me what they think of my page.
Either way, don't expect me to take this page down any time soon.
You mean Apple will not die because they've made translucent cases, and hired back Steve Jobs? Well, I will admit that Steve is an excellent speaker, though why Intel hasn't sued him for libel is beyond me. Making a see through box is not "transforming" -- its called a "fashion statement". Its a one trick pony that doesn't have any longevity (its too easy for PC manufacturers to do their own colorful computers.) The only reason why Apple might have the best selling kind of computer of some particular category is because they have a monopoly position on the MacOS market. Collectively, the "Wintel" manufacturers stomp all over Apple in any category. Apple is experiencing a resurgence right now (this page was written while Gil Amelio, aka Dan Quale, was head of Apple) however, so far Apple has not demonstrated any interesting long term strategies other than "OS X" which is still an unknown quantity at this point. (Just as Windows 2000 and the ultimate impact of the various linux distributions on the PC are unknown quantities.)
This is Steve Job's propaganda and its simply untrue. Every demonstration he has done has been rigged in some way, and his conclusions are gross mistatements. The performance of a top of the line PowerPC G4 (450Mhz) running MacOS is nowhere near the performance of a top of the line x86 (either Athlon @ 800Mhz or Pentium !!!B @ 800Mhz) running Windows. The funny thing is that at the previous MacWorld, Steve ran a demonstration showing the G4 450 running some bogus test where they projected that a comparable Intel CPU would need to be running 800Mhz. Several months later, Motorola has been unable to up the frequency and Intel just goes ahead and ships this chip that Steve said would take "some time" for Intel to deliver. So even in the most extreme cases of rigged tests, the top of the line x86 CPUs still beat the top of the line G4. And don't give me any nonsense about Motorola coming out with something better "soon". Intel and AMD have plans too, and they will be leaving the poor PowerPC architecture in the dust.
Absolutely not true. Its just that Steve's demos are consistently fraudulant. The last head to head benchmark that Apple has done was against a 500Mhz Pentium !!! at a time when Athlon 650's were available and the Motorola CPU he was using was not yet available.
Its very simple -- the PC platform has more choices for hardware expansion and not all configurations are robust. Different vendors with different emphases will have different degrees of "problems". For example, HP Pavillions, typically are not the fastest configured x86 boxes but have a reliability and ease of use that would easily rival a Macintosh. On the contrary, GateWay and Micron are much more aggressive, constantly trying to win performance wars -- as such they commonly choose configurations which are not as robust, but which deliver greater performance when its working. Its called consumer choice. With the Mac, you have only one choice. Apple's choice. Now Apple has quite simply chosen to sacrifice performance for stability, which makes perfect sense these days since no configuration of a Mac can match the performance of a high-end x86 box.
Untrue. Numerous informal polls indicate that upwards of 30% of all computer users play games on their computer. Apple knows this which is why they are trying to revitalize their platform for games (adoption of OpenGL, endorsement from id software, etc) -- but they have a serious uphill battle. They are competing with AGP 4x, and CPUs with monster floating point capabilities that will not be matched by the G4's piddly AltiVec instruction set.
No, it indicates that the game market is saturated and that there are a *lot* of games out there. Walk down the aisles of any software shop. There will be a very large section just for games. Then walk down it again in 3 months -- most of the games will be different. Everyone is just playing different games. At this point you are just making things up about reality.
That's true. Apple has woken up to the reality that competing against PC's head to head on standard consumer applications is a no win scenario. They have simply picked yet another niche (graphics design being their previous niche) and are touting it as the next big thing. Only a loyal Mac die hard is so easily swayed by this. Update: Apparently such firewire + digital camera solutions have existed for PC's starting pretty much at the same time frame. The big difference is that its such a niche market, that is barely gets any air play in the PC arena where more credible features are used to influence your PC purchase. That sub $1000 Mac is a piece of garbage outside of the fact that they've bundled it with its video editting capabilities (Update: a demonstration on ZDTV's "Fresh Gear" showed the iMac platform to be very non-realtime and very non-linear when performing some standard video composition effects.) They are selling you a "pre-obsoleted machine" into which they are going to try to build this video editting market. I don't think this latest ploy is going to go nearly as well as their flourescent cases.
FireWire is "too early". There simply are not enough devices to justify it. USB is simply a better solution for today. FireWire may inevitably be adopted by PC manufacturers as well, but its not because they can't put this into PCs cheaply today. Its simply the realization on the PC manufacturer's part that this is not a justified cost for today. For example, USB-II looks to seriously marginalize the utility of FireWire. The Mac market relies heavy on this dillusion that Mac's are somehow better than PC's because of thier useless doodads. I'm sure Apple feels that adding FireWire is necessary, not because there are enough devices out there to justify it, but because it makes an aritificial stab at the PC market.
You know ordinarly, I would not be so quick to judge what you know or don't know, but only by admitting to being a "helpdesk person" have I formulated the opinion that you are a "Mac geek, that knows nothing of PCs". Sorry, but my experience with "help desk people" is that they generally don't know what they are doing -- they are typically about 2 years behind the general knowledge curve regarding the computer industry. Just for reference, I am software developer that works closely with hardware devices. I am reasonably well known for by "programming optimization" ability as well as my understanding of hardware from a performance point of view. In the past, I have informally helped "Tom's hardware guide" in understanding performance issues. Kind of puts things into perspective doesn't it?
Sounds like baloney to me. I don't think any video card vendor would in their right mind release drivers that disallowed the use of the most popular software application on the planet. Your comments don't seem to be in tune with "reality". This sounds more like what you as a die hard Mac user wants to believe rather than what the reality of the situation is.
See my comments above about choice. |
5/11/99 While on this decidely anti-Apple web site it will not surprise anyone that I basically claim that Apple is lying with their performance claims versus x86 CPUs, it turns out that the internet news community on comp.arch has chimed in with a similar sentiment.
Now the folks on this newsgroup are probably amoung the most technical savvy in terms of being able to compare and assess the value of a CPU architecture. The people are well known in their circle, and the topics discussed are usually well grounded in terms state of the art processor design.
A thread started innocently enough with the following message:
|
Nan Tu: Hi, I'm badly in need of some material in web site or advertisements that make wrong statements about some arch's performance, such as emphasizing on one part of performance while ignoring other parts. The more egregious, the better. Have you seen such things, many thanks!! |
Now there is a fairly even representation of processor architecture folks on this newsgroup, as well as bold claims made by most processor vendors. So one would naturally expect this kind of thread to degenerate into a lot of back and forth between each processor proponent's opinions about other vendors claims that are on shaky ground. But that is not what happened at all. There seems to have been only one theme to follow -- Just how badly has Apple been lying in their advertised performance claims for the past few years.
In the 22 messages that followed this one they all had basically the same thing to say:
|
Terje Mathisen: The classic example is the PowerPC vs x86 ads some years ago, comparing predicted performance of future PPC's, including some that never was delivered, against the then current Pentium chip. The ad very clearly implied that at the time of the ad, PPC was twice as fast as the best x86 (NOT!), and that this difference would increase exponentially. Mike Smith: How about Apple's use of BYTEMark numbers to claim that the G3 was -- twice? three times? as fast as a Pentium II? Anthony D. Tribelli: [...] The advantage to Apple was that the PC binary (for ByteMark) had not been updated in years. Even when new the PC binary was troublesome, the base PC system (score = 1.0) was a Compaq Pentium 90 of some model but the binary was optimized for 486 rather than Pentium. Many years later, this 486 binary was still sitting on Byte's website. A problem of equal magnitude is that the source is simply naive. It tests the compiler as much as it does the CPU. One can get a 2X speed difference on a Mac just by using two different compilers. When the best available PC compiler and best available Mac compiler were both used (at the time of the original Bytemark-based ads - I don't know about current compilers/systems), Mac showed a 25-35% advantage. Arizona: Here are the scores from ZDNet, from their review of the iMac. They mention the apparent dishonesty of Apple marketing with regard to this benchmark. Personally, I consider it out and out lying. No one is going to tell me that Apple didn't try recompiling the code, and didn't know that the scores were totally inaccurate.
The first line is the numbers that Apple got from an executable that they found on the web, reportedly compiled for the 486. The second line is the result that ZD got with an up to date compiler in September, 1998. It looks like to me that the P2 was actually twice as fast as an iMac. If you check the SPEC, Apple and Intel websites, you will see that the same thing is happening with the latest G4 machines: Apple claims overwhelming superiority, but is actually below average. [...] |
|||||||||||||||
In its desperate attempt to stay no more than 200Mhz behind Intel and three years behind the rest of the industry (read: SIMD) Apple prematurely announced -- oh I don't even know why I bother. Lets get right into it with a brief outline of their lies.
Appearing on Apple's website at the time of the launch:
"The first supercomputer on a chip"
This was actually claimed and delivered upon by a little known company named Chromatic Research in 1998. This claim is typically based on the idea of having DSP-like calculating functionality (essentially SIMD) that delivers performance similar to Cray super computers of yesteryear. AltiVec (now called "Velocity Engine") is nothing more than catch up to AMD and Intel, each of which have have evolved similar SIMD functionality in products that have been actually delivered to customers for years now.
I have assurances from university researchers who are familiar with Cray supercomputer performance levels, that the K6-2 (which was shipped in May of 1998) also had similar functionality and performance levels to a single CPU "Cray-3". To call themselves first is just a lie.
Just to put things into perspective, the main use of SIMD technology today is to deliver higher performance games as well as higher quality video playback. To this end, x86 based PC's have game performance that is simply not even approached by the Mac platform and most mid-range x86 PC's can deliver DVD playback which meets its highest quality standards. I don't believe the Macintosh platform has any opportunity here except to attempt to catch up to where the x86 PC already is today.
The image above is very suggestive that the Pentium !!! is a two way execution engine, while the PowerPC G4 is a 4-way execution engine. This is an extremely gross exaggeration. When using ISSE, the Pentium-!!! can generate 4 results per clock with fewer instruction bytes than "Velocity Engine".
The Pentium-!!! can run two SIMD calculations at once, while the "Velocity Engine" really only has one SIMD calculation engine (and thus is not even superscalar in its "Velocity Engine").
On non-branch instructions, the Pentium-!!! is a 3-way execution engine, while the PowerPC is only a 2-way execution machine.
"The new PowerPC G4, architected by Apple, Motorola and IBM, is the first microprocessor that can deliver a sustained performance of over one gigaflop. In fact, it has a theoretical peak performance of four gigaflops."
That's nonsense. The very nature of the Intel and AMD SIMD instruction sets allows them to easily exceed 1 GFLOP in most situations. In truth the theoretical maximum of 2GFLOPs for a 500mhz P-!!! or K6-2 or Athlon is very easy to come very close to on real world code. Its part of what is leading to continually higher and higher frame rates of games like Quake. (I wonder if Apple is at all interested in running timedemos head to head with comparable PCs.)
"Velocity Engine"'s ability to hit 4 GFLOPs requires that back to back MADDs are peformed on every clock (counting a MADD as two operations). No other combination of "Velocity Engine" code will deliver that level performance.
Update: A deeper analysis of the PowerPC G4 architecture shows the 4 GFLOP claim to be an outright lie. The "completion queue" limits the processor to 6 AltiVec instructions out of every 8 instruction sequence without stalling. So even stuffing the remaining two instructions with single precision FP-MADD instructions leads to an absolutely peak of 3.5 GFLOPs. Of course constructing such code which corresponds to a real world algorithm is basically impossible to come by
"The Power Mac G4 comes with an ATI RAGE 128 graphics accelerator."
Well good for you. As of 08/31/99, the ATI Rage 128 was two generations behind the state of the art. nVidia announced the availability of the GeForce 256 graphics accelerator which adds 50 GFLOPs of performance to an x86 enabled PC, in addition to an unparalleled level of graphics performance. This obsoletes the nVidia TNT2-Ultra which in turn obsoleted the ATI Rage 128. In Job's otherwise well calculated drive to subsume existing PC technologies, he failed to realize the incredible volatility of the graphics industry and has ended up with an obsolete lemon.
Other things to note: At 450Mhz, SpecInt/FP95 are estimated to be 21.4/20.4. The Int score is simply not competitive with Intel or AMD, and the FP score (while ahead of Intel) is still well behind where AMD's Athlon is at. Given that the G4 is a RISC chip its staggeringly pathetic that they cannot match the floating point performance of the latest x86 chip. Regardless of how well the Athlon's x87 FPUs are, the instruction set itself should keep it from reaching the performance of comparable RISC chips (indeed, CPUs like the Alpha 21264, and PA-RISC have no trouble trouncing the Athlon at FP.) Nevertheless, it was brave of Motorola to release these numbers. They've probably grown tired of Steve Jobs' fraudulant Photoshop demos.
The primary use for a SIMD instruction set such as AltiVec is for games and multimedia. What this means is that unless the G4 suddenly improves substantially in these areas, there will really be no point to it. Hands up all of you who think that that G4 is going to compete with the AMD Athlon on games performance?
It has also been disclosed that the new G4 now supports "speculative execution" (where the processor executes instruction before it knows for sure that the code path will be taken.) Oooo aaah. Sounds neat doesn't it? Guess what folks this technique has been used in x86 processors for years. Both the K6 and P6 cores perform fully speculative execution (meaning there are no caveats or strings attached -- the degree of speculation is as deep as the instruction buffers will allow, which is very very deep.) It goes without saying that the Athlon also executes speculatively. At best this is catch up on the part of Motorola.
I am not alone in laughing at the new G4 (especially versus the Athlon.) Take a look at http://www.ugeek.com/readercomm/091999/comm101.htm for another clear headed opinion about the new G4 versus contemporary x86 processors.
Here's some more from John Carmack's .plan file:
|
Apple's new G4 systems. The initial systems are just G4 processors in basically the same systems as the current G3. There will be some speedup in the normal C code from the faster floating point unit, and the Apple OpenGL has AltiVec optimizations, so framerates will improve somewhat. The limiting factor is going to be the fill rate on the rage128 and the bandwidth of the 66mhz pci bus and processor to main memory writes. The later G4 systems with the new memory controller and AGP will have better performance, but probably still limited by the new 3D card. After Apple gets all their driver tuning done, it will be interesting to try running timedemos at low resolution to factor the fill rate out. Apple has a shot at having the best non-geometry accelerated throughput, but it will still be tough to overcome a K7 with an extra hundred or so mhz. On a purely technical note, AltiVec is more flexible for computation than intel or AMD's extensions (trinary ops), but intel style write combining is better for filling command buffers than the G4's memory streaming operations. |
I'd say Carmack is being optimistic about the G4 systems. Trinary ops do not translate to any real world performance gain that cannot be overcome by hardware register renaming and ordinary software based scheduling techniques (although Carmack is careful not to claim any performance advantage from this). The Mhz differential with the "K7" is actually 200Mhz. The x86 OpenGL implementation has been performed by ex-SGIers (Michael Gold, Mark Kilgard, etc) whose degree of optimization is unlikely to be matched by any Apple implementation, so I don't know why Carmack thinks the G4 has any chance of even matching x86's at the same Mhz on Quake 3. In any event, he does confirm that the rage128 is just too slow, that 66Mhz PCI (as opposed to AGP) does not cut it, and the current memory controllers in the G4 are pretty bad.
Subject: Apple sues sky: claims color scheme stolen from iMac From: "H.W. Stockman" Newsgroups: comp.sys.intel Apple Computer interim supreme commander Steve Jobs announced today that he was suing the sky for infringement on the stylistic design of the iMac. "The translucent blue color is a clear ripoff," said Jobs. Apple also announced its plans to sue numerous manufacturers of toasters. "The rounded corners and approximate size of the so-called `food-appliances' are clearly inspired by the iMac," said Jobs, as he gazed lovingly into a mirror at an image he referred to as "God". Also cited for possible style infringements was the bottler of Blue Nun Liebfraumilch. "The bottle is almost exactly the same height as an iMac, the name `Blue' is clearly an infringement on our color scheme, and the bottle is also translucent. Don't tell me that's a coincidence." http://www.news.com/News/Item/0,4,40716,00.html?st.ne.fd.gif.e |
The following poll was taken on an Apple oriented site (which uses HTML code that is not Opera compatible -- sigh):
The results, when I checked (08/01/99), were as follows (html changed for syntactical correctness):
| MacBC Question | |||||||||||||||||||||||||||||||
| |||||||||||||||||||||||||||||||
| Click Here to create your own FREE Poll Now! Companies: Advertise on freepolls.com. 2000 FREE Impressions! | |||||||||||||||||||||||||||||||
As of 08/02/99, when Intel releases their 600Mhz Pentium-!!!, Motorola/Apple will be 150Mhz behind. (The AMD Athlon is supposedly also due out soon, and will reach 600Mhz at introduction.)
Update: Holy cow! AMD surprised everyone by introducing a 650Mhz version of the Athlon with greater per clock performance than any Intel based CPU. So that puts Motorola 200Mhz behind the current x86 leader. (This is starting to look like game set and match soon at least for the PowerPC processor.)
Oh while I am here I might as well make fun of Apple's high end G3 specs: P-!!! Xeon 550Mhz with full speed 2MB L2 cache is available, Xeons support up to 4GB of actual on board memory, ATI Rage 128's are soundly beaten by: nVidia TNT2, TNT2 Ultra, Matrox G400Max, and 3DFX Voodoo 3000, which are options on any x86 based PCs -- in addition to other super high end cards like "Wildcat" and "Oxygen" video cards that are not available for the Mac.
450Mhz Pentium's are kind of low end by today's standards. So making comparisons to it are kind of silly.
It appears as though Apple has settled a class action lawsuit regarding some disputed charges for technical support. You can read about it on Apple's site.
05/05/99 I recently got to meet an employee from Xponential who was intimately aware of what happened. First of all a recap: back in 1996 Xponential had developed a 400Mhz PPC part that they were able to crank up to 533Mhz in a technology demo. This was back when x86's and PPC's from IBM/Motorolo were running at no more than 300Mhz. The inside scoop was that the 400Mhz processors really only performed like a 250Mhz part (mostly because of BUS limitations) in real world tests, though it demonstrated fantastic performance in the now coveted "Adobe Photoshop tests" that Apple pushes so much these days.
Anyhow, as is well known, there is a serious market for PC's which are quantified purely by clock rate. That is to say, even if the CPUs didn't actually perform better, they would still have sold with customer perception being that they indeed were faster. Somehow Apple, under Gil Amelio, missed this -- egged on by IBM who was promising some fantastic new PPC part that would blow away the Xponential and the then current 604e's that Motorola was shipping (IBM never shipped this part). Xponential lost Apple's business, most of the investors pulled out and soon went belly up -- though not completely.
After selling off its patents (which had implications for Intel on Merced's intellectual property) in a highly publicized auction (which S3 bought for $10million from which they incredibly extracted a broad patent cross licensing agreement with Intel -- a gigantor company at least 10 times their size.) Xponential decided to take one final go at it by selling their CPUs to the Mac clone manufacturers (you remember them right? Power Computing, UMAX, etc.) Just as the clones were poised to go public with their highly clocked Xponential driven boxes, Jobs stepped in and pulled the MacOS licences out from under them.
Why does this matter you ask? Its very simple -- there is a single chip model in this industry that has awed all who have understood it. -- the DEC Alpha. Although a marketing failure, there is no doubt of its technological success. The first generation was an in-order single execution fully pipelined ho-hum processor, the second generation was a 4-way super scalar architecture, and the third generation is a fully out-of-order super-pipelined multi-ALU monster chip. Throughout its lifetime, the Alpha has always had unbelievably high clock rates and has been either the fastest or second fastest CPU on the market (even though Digital's silicon fabrication process has always been at least one generation behind Intel's.)
What the Alpha model proves is that indeed having a very high clock rate is incredibly important, because what you learn in designing a processor like that you can carry with you even as you improve the architecture from generation to generation. (The mistake that PPC advocates always incorrectly and irrelevantly cite is that architectural performance can be more important than clock rate -- what this misses is the fact that clock rate scalability is in fact just as important as architectural performance, furthermore, having a high clock rate does not preclude having a good architecture -- ergo the 21264 processor.) This is something that the Xponential folks clearly understood. Apple showed their myopia in not having enough vision to try to support the Xponential guys. With the Xponential technology, by now they may have had 700Mhz+ cpus with improved architectures that really were twice as fast as x86 CPUs. That kind of umpf could have significantly expanded Apple's product line and made it far more compelling than the x86 boxes out there today.
So why can't Motorola simply pick up where Xponential left off and build some high clock scalability intellectual property and bring this idea to reality only a few years later? Its like all technology -- blink and you miss it. With the purchase of Digital, the Alpha team members have basically disbanded, however, many of them have landed at Advanced Micro Devices and Intel. To put it mildly, its too late. Both Intel and AMD have demonstrated 1GHz parts -- and not just some technology BS, but actual working parts (both companies are committed to shipping such CPUs by the year 2000.) Motorola doesn't have a prayer of catching up anymore. The disparities in the clock rates today between x86s and PPCs are only going to get worse -- much, much worse.
Apple is currently faced with a crisis. It may not hit them today or tommorow -- but the fact is that neither Motorola nor, IBM can keep up with Intel and AMD which are going to be going like banshees with their incredible clock rates and incredible CPU architectures, all in affordable PCs. Apple will soon have to face the fact that their PPC boxes will be serious piece of crap compared to the contemporary x86 box. And no, AltiVec will not save them.
I hear rumours that Apple is looking to switch to the x86 platform -- but how do they do that without admitting that they've lost? As usual they look like they are totally missing the boat -- the companies Connectix, VMWare, and possibly TransMeta could probably solve their problem by developing technology for high quality and high performance emulation of PPC code on an x86 processor. This could allow them to develop an x86 based version of OS X, while retaining backward compatibility with the Mac. But I have not heard a single peep from anyone indicating that anything like that is going on.
Jobs has proven that he is a resourceful guy as well as a master at marketing (perhaps almost as good as Gates himself) but what will he do when he is faced with a really hard problem like this one? We shall have to wait and see.
At the MacWorld expo in San Francisco recently Steve Jobs has finally made announcements from Apple that I could not laugh my ass off at. This is what Steve Jobs announced (that I care about):
OS X server - its vapourware no more, its an actual product.
OpenGL & ATI Rage 128 support.
Conectix Virtual Playstation product.
Dropping SCSI in favor of fire wire.
Whoah. Finally -- after years of useless baloney out of Apple, some actually credible, real products.
First some comments: (1) PPC's are not faster than x86's -- Adobe and Byte just have interesting concepts on how to write fair cross platform applications. (2) The ATI Rage Pro in the iMac's is also in some PC systems, and is generally considered among the very slowest cards in the PC market (nVidia's and Matrox G200's have no trouble in trouncing the Rage Pro). The Rage 128 is very fast, however it will be available very soon on the PC's as well -- clearly Apple just did a positioning deal with ATI for product lauch, that will not deterr the PC's adoption of it. Also, some PC power users will probably like 32MB's of ram or more in their versions.
I'm not sure why Jobs wants to continuing misrepresenting the facts like this.
Anyways, but when we cut through all the crap, and baloney that Jobs feels he must be a part of, he was left with a few technologies that there can be no denying is actually good for something.
OS X represents an answer to Be, Linux, and NT all at once. Now, I don't know many details about it, but if it sprang from the UNIX-like NextStep, then it probably is a serious no-nonsense OS.
OpenGL means that all the id games, and lots of serious 3D graphics work can move to the Mac if so desired. The pressure for other vendors (consoles, digital TV's and so on) to move to OpenGL so that they can develop on a Mac or a PC will grow, and the competition to MS's proprietary and controvertial Direct X will be more credible.
SCSI is stupid (too expensive for the irrelevant benefits it provides). Glad to see Apple dropping it. My reading on fire wire is that its state of the art right now.
The new G3 form factor is actually quite nice (as opposed to the iMac which is just a joke). Being someone who lugs my PC around every now and then -- that actually seems like quite a good idea. The handles seem like a winner to me. (Every time I lug my PC around, I'm scared I'm going to drop it.)
Ok, so does thing mean Apple is going to start taking serious inroads into the "Wintel" market with these machines? I don't think so. They are still limited to a market of people who don't understand why the Mac OS is a piece of crap, but they are moving in the right direction. If they want to launch a credible attack against -- well at least Microsoft, they need to take a few additional steps:
They need a way to lose the Motorola dependence. The PowerPC is a dead end. I'm not saying that they have to go x86, after all they could try for the Alpha, but just lose that ridiculous PPC processor. It has no future. 400Mhz towards the later half of 1999 will be useless -- there are a couple serious CPUs that will be in the 600Mhz-800Mhz before the end of the year. Anyone who understands the current state of architectural scaling of CPUs knows that the future is in the clock rate, not in any claims of "better CPU design" (Pentium-II's and CPUs from AMD are actually better designed than Motorola PPCs anyways.)
Make sure they are able to support lots of graphics vendors (ATI Rage 128 won't be the fastest forever -- ATI is notoriously mediocre at delivering a consistent well designed architecture that scales for any appreciable amount of time as are most other graphics cards.)
Well, they'd better bring down the price of OS X server. I'm sure they are limiting the volume so that they can use the initial customers as unofficial beta testers (that's what this industry has become) but $999.99 is a bit much for an OS even if it is for servers. Remember that the competition is MkLinux (which costs significantly less).
Apple just may be making a real serious come back.
So should I eat crow? Well in my defense you have to consider the facts: after the introduction of the Macintosh, there has been, in my opinion no serious technology innovations from Apple. That would be a decade of languishing while Bill Gates hobbled together stuff to catch up and leave the Mac in the dust. I've had absolutely no real feedback that suggested that anything would happen to really change that. Perhaps, I should have given Apple's iCEO (or at least the team he brought over from Next) more credit.
The fact is these developments have all come to a head very recently which is about as quickly as Jobs and his Next team could have reasonably pulled together their substantive changes. (As opposed to translucent boxes, slashing prices, fraudulent performance claims, and slapping a disk drive and internet search engine together are about all they could do immediately.)
And the situation has even gotten better for the Mac line ... I have recently gotten information under NDA that indicates that Apple will be taking a much more radical step to ensure the future of the Mac line. I am very sorry that I cannot disclose this to anyone until its public info ... Hmmm lets go see if its on that moron "Makido"'s site ... Well I can't find it. (That guy is such an idiot -- Intel lost their market share to Advanced Micro Devices, not Motorola/Apple, Intel's reference compiler comes with VTune which is about the same price as a compiler, and Gil was clearly swindled out of his job by Jobs who it looks like is on a mission to "save Apple") Looks like you will have to wait for it.
So I'm playing Apple's pride and joy "Quick Time" because my alternative is Real Player, which I haven't registered, so I can't save Real Player files. It turns out that unless playing in the foreground, the "Quick Time" player refuses to play audio. Uhhh ... there's absolutely no reason for that, AVI's can play concurrently in the background no problem. I suspect Apple is just artificially limiting QuickTime to keep it from appearing to function any better on Windows than it does on the Mac (where I'm guessing it has the same problem -- but only due to an OS limitation since there's isn't really any such thing as a concurrent task.)
Anyways, so that's not such a big deal, but in that case I would like the option of forcing the Quick Time player to the foreground (Windows apps can be programmed to have a "force to foreground mode".) But there is no such feature.
What the hell? No wonder Apple is so eager to push their "standard" onto Windows. That way they can ruin the whole OS experience through active limitations in their so called standard. Bastards.
At the recent Microprocessor forum (and following the Intel developer conference) Intel gave the details of the KNI architecture.
At the "Makido" site there is an article attempting to detract KNI in favor of AltiVec. Here is my rebuttal:
Remember that MMX + 3DNow! has been around for quite some time now. Your so called "rush to market" has Motorola and its VMX/AltiVec vaporware practically twisting in the wind. These technologies are actually here, ready to be used and showing tremendous results (John Carmack says Quake on PPC is the same as Quake on a P-II (theoretically), which we now know is a lot slower than the 3DNow! implementation.) Intel is late, and so is Motorola. In that sense both KNI and AltiVec are failures.
Now specifically onto KNI vs. AltiVec. Your "3 register operation" nonsense isn't worth the bits you wasted on your web site to talk about it:
PPC 750 + AltiVec (issue rate of 2 instructions per clock):
AltiVec_MUL reg0, reg1, reg2
Katmai (issue rate of 3 instructions per clock):
mov reg0, reg1 ; This instruction is essentially free.
KNI_Mul reg1, reg2 ; SW swaps meaning of reg1 and reg0
To counter the "few registers" argument, the P-II architecture has direct load forwarding meaning that redundant loads (with other unretired loads) take 0 clocks in the same way as above. I.e., the extra "loads and stores" you are talking about are irrelevant. Since the need&$47;desire for 3 operanded calculations are very rare it barely scratches the extra 50% decode bandwidth of the P-II, while saving lots of otherwise usually unused silicon.
The "too few registers" argument really doesn't hold much water, since the real performance bottleneck of modern out-of-order architectures is limited by the ALU performance.
The same holds for the "multiply accumulate" issue. Under KNI, you simply follow the multiply with an add instruction. While this does occupy two out of 6 decode slots leaving 4 (versus AltiVec's 1 out of 4 leaving 3), it does achieve the maximum unit latency. And with SIMD instructions, latency starts becoming a lot more important than raw throughput.
Also remember, that since the x86 architecture uses a variable length instruction set, the average code length is a lot smaller than similar code on the PPC increasing instruction bandwidth.
Your "ugly" comment has no technological relevance.
Intel, having spent a long time on KNI, and having rumoured to have had higher precision, as well as other features, has ended up with a ALU performance that is essentially identical to 3DNow! So the question is, with such a long development time, how is it that they were not able to improve on 3DNow! ?
Simple -- the architecture was optimized for highest real performance. Not baloney paper and pencil performance like AltiVec. The internals were simplified to make sure that Katmai could reach and exceed clock rates of 500 to 600Mhz. The folks at AMD/NexGen simply arrived at this same "sweet spot" architecture a lot sooner than Intel (AMD has other issues limiting their clock rate though). Intel was beaten by a design team with more foresight -- they were not out-designed. AltiVec is just being beaten much worse (from a time to market point of view.)
Now given AltiVec's rather "aggressive spec" with all the extra silicon required for 3 valued alu's multiply accumulates, log and exp approximations (*laugh* -- who the hell needs frigging logarithm and exponential approximations?!?!?; that is such as ridiculous joke!) how the hell is AltiVec going to achieve its performance goals?
There are two choices for Motorola: (1) They simplify their internal architecture to something similar to Katmai/K6-2 (via double pumping like KNI for example) thus achieving no more performance but allowing for higher clock rates than are currently available in PPCs, or (2) They implement a highly aggressive low latency, fully pipelined and fully parallel architecture but become restricted by clock rate.
You can see how the AltiVec architecture has started to come apart at the seams when you look at its "reciprocal" and "reciprocal square root" approximations only give 12 bits of accuracy and lead to a first iteration accuracy of 1 ulp in only 95% of the inputs (3DNow! is 14/15 bits and is exact for 85% of inputs and within 1 ulp of all remaining inputs.)
In addition, I believe Motorola has announced a die size for AltiVec that indicates that it is very small. This leads me to believe that they have tried to cut corners, and therefore have gone with approach (1). (Fast, large register designs usually require lots of die area.) Also the fact that IBM has announced processes that will take PPC to 400Mhz and beyond indicates that they cannot afford to let AltiVec hold them back.
Sorry, but I just don't see AltiVec making a serious difference versus 3DNow! or KNI. It'll all be about clock rate, and it looks like Intel is winning that race. (If you ignore Kryotech's refrigerated AMD K6-2 500Mhz parts.)
Update: Ok, so I've read through the MicroDesign resources explanation of AltiVec. Rather shockingly, it does appear to be a solid 2 times faster than KNI/3DNow!. Its basically a single issue, fully pipelined, single pumped, 4 wide SIMD unit.
So what's the catch? Simple: It won't be available until Q3 '99 and it will only be running at 450Mhz in a dual issue CPU. x86 CPUs should be well above 600Mhz by that time frame and be triple issue. In other words, Motorola has wasted this huge amount of time designing this AltiVec monster that will drag down the overall performance for sake of a theoretical architectural coup. (And they have missed the entire SIMD market between '96 and Q3 '99 -- ridiculous.)
Hiding behind some claim that Mac users don't want it in its current form, Jobs announced that the next generation of OS's for the Mac will be Mac OS X (meaning version, 10 I suppose.) He then gave some song and dance about how Rhapsody, in its full implementation, will be available some time after that. (Keep in mind that the BeOS, which was an alternative to Next Step, is shipping in Beta form for both x86's and PPC's right now.)
(As you can tell, I've grown so apathetic about what is happening in the Apple world that I didn't even bother to report this when it became news ...)
06/11/98 IBM has sold their interest in the PowerPC processor back to Motorola. This should come as no surprise since IBM has been aggressively licencing Cyrix, AMD and Centaur designs and probably makes far more on them than the PowerPC which they invested so much more money in.
(05/19/98) VMX has been renamed AltiVec reportedly 162 instructions with SIMD capabilities. They target 32 128 bit new independent registers. Following the PPC architecture as I know it so far, this means if they use one unit (likely) they will be able to compute 128 bits of result per clock with a granularity of 32 floating point, or 16 or 8 bit integer. This is no more than AMD's K6-2 throughput which performs 2x64bits of result per clock.
Such a decision to go with 128 bits doesn't make a lot of sense to me, since you are not ordinarily going to encounter that much parallelism in data per operation in ordinary algorithms (except for Photoshop of course; it makes me think Motorola's CPU division is nothing more than a hardware division for Adobe.) More to the point, typical multimedia algorithms do many simple operations (add, sub, average, multiply, saturate) on parallel streams of data, then funnelling them (with an accumulate or select function) to a narrow data stream to perform more complicated functions (reciprocal and inverse square root.) But since AltiVec is orthogonally 128 bits, that means it has to commit all 128 bits of its compute power to each operation, even if its just a divide on a single 32 bit quantity. This lack of flexibility versus the K6-2 (which uses two units which target 64 bits each, but which can be assigned to different operations) makes it seem inefficient by comparison.
Motorola has left the door open to the possibility of having multiple AltiVec units in a PPC, but I find it hard to believe they would do that (instead of seperating their Load/Store unit, which would be much more valuable to them in terms of real world performance.) But then again so has AMD and Intel.
One difference they seem to be very proud of is a permutation instruction which is indeed very useful (as long as they make no attempts at patenting it, since the MPACT media processor, as well as other DSPs, clearly predates it.) On the other hand, they do not offer any SIMD 32 bit integer operations (which are present in the MMX design available from all x86 CPU vendors -- their integer SIMD width is limited to 16 bits.) Also since the new K6-2 instructions (called 3DNow!) are few in number (21) they are very simple to understand and programmers can get to work with them in and understand optimization immediately (I've invested a total of about 2 days, just studying them, and I understand them. Update: I've now spent considerably more time with 3DNow! and I must tell you it rocks -- it certainly beats developing on vaporware.)
According to the tech docs on it (available here) they committed silicon to approximate log_2 and 2_exp functions. I can't possibly imagine an application wanting 128 bits of log_2 or 2_exp result per clock, but hey, its Motorola's die area to burn I suppose. Their 1/x and 1/sqrt(x) estimation functions generate only 12 bits of accuracy, which is 2 bits less than the K6-2 on first iteration. It is not surprising that they include a note regarding numerical instability that prevents it from generating consistent estimations within 1 ulp even after using Newton-Rhapson. They also do not support a built in iterator. So the Newton-Rhapson step must be hand coded by the programmer, into what appear to be 4 additional instructions, versus the K6-2's 3 additional instructions.
In all honesty, I think AltiVec and MMX+3DNOW are about equivalent (the points I make above are kind of nit picky), but the most important difference is, 3DNow! is already sampling in the K6-2, and will be released at the end of May, while AltiVec is slated for Motorola's version of G4 which is not being supported by IBM, and which is not going to be available anytime soon.
Stories such as the one on news.com don't quite get the picture. The AltiVec technology does not offer anything that will not be available from AMD (which will beat them by 6 months (Update: 15 months)) and Intel (whose Katmai processor is slated to released on roughly an equivalent time frame as the Motorola G4. (Update: SSE will beat AltiVec by 6 months))
A little while ago, Intuit announced that they would discontinue Quicken for the Mac. A devastating blow to Apple that would have seriously called into question the role of the Mac as a home computer (even more so that it already is.) But Intuit did an about face shortly after, and in fact has become a bundled package for Apple's new iMac (a translucent throwback to the original version of the Mac, priced at about $1300.)
What does all this mean? Very simple, Intuit has no serious commitment to Apple, and the deal that they were trying to work out with Apple for this iMac nonsense must have been going sour for Intuit, to the point where Intuit felt it had to threaten Apple to get a fair deal. It really looks like signs of desperation from Apple.
A company called CONIX3D has developed OpenGL drivers for the Mac OS. You can find about about it at:
http://www.conix3d.com/MacOS.html
Now the info they give that is important to me is under Performance. They claim:
|
"On a Power Macintosh 8500/250, 24 bit color, 110,000 light smooth shaded, z-buffered, polygons/sec, and 135,000 light flat shaded z-buffered polygons/sec. Plug-in renderer support allows you to choose the configurations you need for those writing a commercial application. The standard release comes with six standard plug-in renderers representing different color buffer and depth buffer configurations. Plug-in renderers have 50% smaller memory footprint and support 3D hardware plug-ins. Hardware plug-ins will be available very soon." |
These are NOT texture mapped triangles kiddies. (That "very soon" comment wouldn't be a Copland-esque "very soon" would it?)
On the PC, hardware accelerated OGL is available. A quick check of the nVidia, RIVA 128 (an affordable PC based graphics accelerator that is currently available) data indicates: 5M triangles/sec, 100M pixels/sec, texture mapped. Being somewhat knowledgable about such things, I know that turning on features such as "smooth shading" doesn't cost anything for this hardware.
While I realize this is a bit of an apples to oranges comparison, even giving any sort of reasonable lee way for OGL overhead indicates that this is just night and day. On the Mac, you just can't get anywhere near the performance of a PC for graphics.
The commercials that Apple is currently running, indicate that the G3 is twice as fast as a P-II. In his keynote address as Seybold, Jobs cited Photoshop (as usual; see above) and ByteMarks. Fortunately, Byte supplies the source code for ByteMark which has allowed Eric Bennet to run the tests on his own.
As you can see, the code compiled with the Motorola/Apple inhouse compiler that we know nothing about, is far and away the fastest performer, but we have no idea if the results are really correct, since it is only a binary, and the ByteMark output is insufficient evidence of correct internal operation.
Furthermore, the results are not approached by any conventional PPC compiler on the market today, regardless of how high the optimization setting is set. (The PPC Linux results are a little unfair since similar P-II Linux results were not collected.) Keep in mind that your conventional Mac applications are not likely to be created from Motorola's internal compiler.
If we dismiss results that cannot be duplicated in conventional applications we find that the PPC and P-II at 266 and 300 respectively are really about the same performance. That's pretty pathetic considering the P-II is being held back by its older "CISC" based roots. (The 300Mhz G3 is no match for the 400Mhz P-II of course.)
Twice as fast is a myth. The real question you should be asking yourself is, how well does the Mac run WinBench and WinStone?
Update: Here's an email I recently received from an insider.
Date sent: Thu, 30 Jul 1998 23:18:46 -0400 From: [Withheld] To: Paul Hsieh Paul Hsieh wrote: > As an aside then, what hell is wrong with the IBM and Motorola people? > They have gone through several iterations of the PPC including the 601, > 602, 603, 604, 620, and 750 lines. How is it possible that they've > implemented them all so poorly? Do they just suck? Not that you heard this from me, but I've heard a rumor that Motorola used the wrong compilers to generate instruction traces to be used for chip designs. I don't know which of the chips were designed by Moto, and which were designed by IBM. I do know that the Apogee compilers (the ones that Moto used) cheated on 023.eqntott, because [incriminating reason withheld] Note, too, that HP once cheated outrageously on matrix300; it will be interesting to see whether the benchmarks used to design IA-64 are an accurate sample of "real programs" or not. |
As further fuel to the first read this:
TechWeb reports that Pentiums are faster in real world activities, especially graphics
Here's a warning label I've heard applied to a CD, full of articles, HTML and what not I was reading about recently:
|
Sorry folks, because of the way Mac's treat ISO9660 disks (reads the directory structure in as a Braindead DOS structure), the long filenames and directories are warped into filena~1.htm, and the HTML expects filenames.html. You can still get to the archives by traversing the CD directories via the Desktop. |
Well, Mr. Jobs, your admiration for ISO doesn't exactly extend to your product now does it? This wouldn't be an artifical attempt at making Windows look like it doesn't support long filenames would it?
Here are some tech report calls into Apple, that I think are just hilarious ...
I would like to point out "PaperClip" is also an old piece of software that use to run on Commodore 64's ... but that's neither here nor there.
Announced 02/27/98. Just one question ... how much did that failure cost?
The following was posted on comp.benchmarks:
Subject: Re: ByteMark Bench mark show Macs are faster ? From: macghod@xxxxxxxxxx.net (macghod) Newsgroups: comp.benchmarks, [...] In article , XpackX@xxxx.xxx wrote: > Chip Mhz Spec95 Millions Size Power Etch > int fp transistors mm2 watts microns > =============================================================== > PII 300 11.9 8.6 7.5 203 .35 > PII 333 12.8 9.1 7.5 203 .25 > 750 233 11.0 8.1 6.35 67 5 .25 > 750 266 12.4 8.4 6.35 67 5.7 .25 I feel so lied to!! The g3 is supposed to be faster than the pentium, even up to twice as fast!!! |
Geez, being an Apple user, I figured you'd be used to being lied to. Anyhow, the G3 does appear to beat the P-II clock for clock, but at the top of the line Intel is still ahead of Motorola/IBM.
Here's some more info he sent me:
How about these real world tests...
Apple Gateway
Power Mac Pentium
G3/266 2/300
Photoshop Open 0:08 0:06
Artistic - Fresco 0:44 0:54
Sketch -
Graphic Pen 0:27 0:33
Texture - Craquelure 1:04 1:10
Gaussian Blur 3:22 3:03
Resize 0:22 0:26
Premiere Make Movie
(w/scaling) 8:16 8:44 |
Ok, so the G3 wins, but only by the slimmest of margins, on a test rigged to make the G3 look good (i.e. Photoshop.)
Here is an article by David Poque written in MacWorld:
http://macworld.zdnet.com/pages/june.97/Column.3700.html
While I personally have no great love of the Windows operating system, the article written above is totally dillusional. Here are my quick counterpoints:
My understanding is that the Windows market share is currently more like 85%, and Microsoft has recently announced (Jan 98?) that total Win 95 sales has exceeded total Win 3.1 sales. Assuming similar piracy rates, that means that Win 95 holds greater than 40% of the OS market share, with Win 3.1 a close second. Apple on the other hand has been held to 6%, as far I knew. While my figures may be different due to when the survey was taken, either way the premise that Windows was on its way down has not been supprted by statistics.
NCs as Ellison envinsioned them are a fanciful dream (that have nothing to do with over priced computers from Apple, BTW.) While the market will certainly be there, they won't be filled with Newtons or cheap Sun based Java boxes but rather by sub-$1000 PCs.
While nobody is a really big fan of this fact, Windows has found quite a home on the internet. IE, MSN, NT, ActiveX and Microsoft's flavor of pseudo-Java are all fairly big business on the net (as bad of a taste as that does leave in all our mouths.)
Windows too expensive???? Give me a break, regardless of price, Microsoft will continue to play their bait and switch "Windows is free with every new PC" strategy, and consumers will keep falling for it. BTW, how much royalties are UMAX still paying to ship MacOS on their clones? How much will they pay for Rhapsody?
Finally, the issue as to whether Microsoft can be saved ... is this guy on drugs or something? Bill Gates is the richest man in the world for a reason; he's 1/3 owner of one of the biggest gargantuine companies in all of the US. He doesn't make his money on the OS by the way; he makes it from selling Microsoft Office. (Which ships on the Mac as well, for what its worth.)
Best Buy Corp, one of the nations largest chains of electronics stores, has decided to stop selling Apple computers, due to slow sales.
A Best Buy spokeswoman said that Macintosh sales accounted for less than 0.5% of Best Buy's dollar sales of computers over the last year.
They will continue to offer some Mac software (for now).
According to a PC World news radio report, people working on the next generation Newton have been told by Apple to start looking for employment elsewhere. Maybe Jobs is just seeing how fast he can drive Apple into the ground.
Although Motorola has not made any anouncements, they are reportedly working very hard on VMX. Rumors are that it is a coprocessor that performans matrix operations in parallel to the main CPU. Sounds interesting, but Intel, AMD, Cyrix, and Centaur have all aready announced their plans for second generation MMX which will perform substantially the same thing as what these rumors indicate about VMX. As to the issue of parallelism, remember that the K6 and Intel CPUs are now "POST-RISC architecture" CPUs which means that they drive all of their units, including their MMX, integer and floating point units in parallel. So we'll have to wait and see, but its clear that Motorola has a major uphill battle in front of them since they will be competing against the current MMX as well the next generation MMX, by the time they are ready to ship.
Here's a system Joe Ragosta (a die hard Mac advocate) put together:
$2250 266 MHz G3 minitower $200 Upgrade SDRAM to 96 MB $500 Quality 17" monitor $300 Ultra fast/wide PCI SCSI $400 4 GB ultra/wide SCSI drive (in addition to 4 GB drive with the machine) $300 Fast video card. Total $3950 |
Here's the specs on a machine I actually bought:
$875 233 MHz K6 minitower, with DVD drive $35 32 MB of EDO RAM $900 Awesome 19" monitor $75 X2 56K Modem $35 Diamond Stealth 2MB graphics card $15 Cheapo SB-16 $100 100 watt speakers Total $2000 |
You'll notice I saved $2000, from which I loaned my sister $1500 for a nice used motorcycle. You'll also notice that Joe got a couple of extras that have extremely questionable value. A friend of mine put a system together from scratch; similar to mine w/17" monitor, 33.6K modem, and CDROM drive and managed shaved $400 off the price. Here's my run down on the two systems:
96 MB of RAM - since the PPC is a 32 bit RISC chip, that means its using its RAM half as effectively as a 32 bit x86 chip. There are basically no PC applications that even notice memory beyond 64MB, and basically none that dont work or are unusable at 32MB. While I could have gone to 48MB to be equivalent to Joe's system, I would have actually derived very little benefit from it. (OTOH, I do plan on an upgrade to 256MB in the future; for programming and experimentation purposes.)
SCSI. No matter how you slice it, disks (and external devices in general) are slower than RAM. Having a DISK run twice as fast is irrelevant if your application is reasonably written. The only reason Mac-heads are so big on SCSI is because of the god aweful load times for Mac applications *need* every inch of disk speed just to be tolerable. On PCs this is simply not an issue.
4GB IDE hard drive + 4GB SCSI hard drive; Does this not just make your hair crawl? Why not just get a big 8MB SCSI hard drive and throw out the damn IDE drive if you are so keen on a fast hard drive?!??! Better yet, why bother with all this bollocks SCSI in the first place??
What the hell is a Quality 17" monitor? If you can't see 1600x1200 on it then what's the point? Personally, I'm very happy with the premium I paid for my nice 19" monitor. Its almost as good as the 21" monitor I have at work.
$300 is the price (expected; its not yet available) of a 3DFX VooDoo 2. I very much doubt that any fast video card for the Mac is going to even approach the performance of such a graphics card. Furthermore, you lose again on the Mac regardless of the speed of the graphics card because of its inherently backwards graphics architecture. My crappy graphics card will probably run circles around anything available for the Mac for precisely this reason.
Joe made no mention of modem, or speakers. Do G3's come with them or what?
The 266 G3 might outperform a K6 233 in some situations, but I would have to study that two chips carefully to know for sure. Either way the price differential to a P-II 300 is less than $500 and doesn't come close to justifying the tremendous price differential between a Mac based system and a Windows based system. (I plan to upgrade to a K6-3D 266 or 300 Mhz processor later on this year which will blow away anything from Motorola/IBM, and possibly Intel; I expect this chip to cost me about $400.)
Geeez ... $4000, I wonder if I could get a 500 Mhz Dec Alpha for that price. It would certainly be more compelling than a crappy old Mac!
So while Joe gets a machine that will give him eyestrain, can't pre-emptively multitask, has a lot less (desirable) installable software out there (I got Quake II, and Age of Empires as well; I also installed Gravity, 4DOS, WATCOM C/C++ and Opera which I have owned all along) I've got a Yamaha Virago for my sister and $500 extra cash in my pocket to boot.
Did Joe just put together a bad system? Well, when it comes to Appl
