OSNews

"More important than this, is some of the technology which Apple is showing off in the Xserve product. Finally, Apple is up to speed with their internal bus speeds."

If Im Not Mistaken ... Is The FSB Still at 133mhz Per G4 ???

Correct Me If I'm Wrong...

WooHoo First Post !!!

and a very good discussion of the decisions apple had to contend with. with respect to the ata drives, i have read articles which discuss the use of ata drives in future higher end storage systems. i am not quite comfortable with them yet, but redundancy and a back up plan should take care of that. i do use apple servers, and i am looking forward to getting an xseve this summer, though i would like to see the raid system first. and, since we have a small design office, i am considering using these as rack mount workstations like the intellistation r (discontinued). these really need to be evaluated in terms of apple's market, which is not ferrari's but more like subaru's with the wrx--fast, inexpensive, reliable... and, having witnessed the potential horrors of windows servers, i have to say that apple's servers tend to work very well in heterogenious environments, and never seem to outright fail.

Advancements in computing have rendered ECC obsolete. There is no reason to have ECC RAM as long as your computer uses SDRAM DIMMs or newer style RAM.

The ATA busses in the Xserve are ATA/100 for a combined theoretical peak of 400MB/sec.

The RAID in most Macs is software RAID, but Apple has included a hardware RAID controller for the Xserve.

All in all it is a good value considering that the G4 1GHz can run circles around the P4 2GHz. [Everyone thinks clock cycles are a measure of equivalency, but they aren't. That's just Intel's reality distortion field. No one would question that the Itanium is more powerful that the best P4, but it only runs at 800MHz.]

Check out the Spec stuff for the Itanium....for integer operations it's terrible. I think that the Sparc and the Itanium are the only modern general purpose CPU's that lack out of order execution, and the Spec int scores reflect this.

"thinks clock cycles are a measure of equivalency, but they aren't"
That's not the whole story. How many instructions a CPU can execute in one second is
(clock speed) * (instructions per cycle). So what if the G4 can do more instructions per cycle if it's clock speed is so much lower the the P4(less then half the fastest P4 by now). You can either increase clock speed or the number of instructions per cycle to increse your performance.

My $0.02

James

Taking things in reverse order:


> No one would question that the Itanium is more powerful that the best P4, but it only runs at 800MHz.]

I think that quite a lot of people would differ with this statement.

The fastest an Itanium managed in on the SPECint2000 cluster of tests was 395. It did quite a bit better on SPECfp2000, scoring a 715. By comparison, Dell managed to score 811 SpecInt2000 and 802 SpecFP2000 on a Precision WorksStation 340 w/ 2.2Ghz Pentium IV.

The scores for the Itanium are a bit old. Compiler improvements over the past year or so would have driven up its speed a bit. But not *that* much. Outside certain applications (primarily cryptography), Itanium hasn't done particularly well. Certainly it hasn't been able to make any real inroads into the markets where 64bit processors are needed. We'll see whether the much-delayed McKinley is able to pull the Itanium architecture out of the benchmark dungeon.

>All in all it is a good value considering that the G4 1GHz can run circles around the P4 2GHz. [Everyone thinks clock cycles are a measure of equivalency, but they aren't. ]

In that case Apple should fire all of its marketers, who have been saying this ever since the G4 got stuck at 400Mhz.
I think the "clock cycle" myth is a bit of a red herring.

The problem is that on most workloads the G4 really isn't that fast. While there are certainly exceptions, particularly applications that heavily benefit from Altivec's vector model (Photoshop and BLAST), the G4 doesn't perform that well.

And that makes sense. Remember, Motorola's primary market for the G4 isn't Apple. It is embedded systems. This means that power efficiency, heat dissipation, and probably cost are more important than raw performance. On the other hand, only a tiny percentage of Intel's new x86 CPUs go into anything more embedded than a laptop, so Intel spends watts like there was no tommorrow in pursuit of high performance for its pricy processors.

Here Anton Ertl summarizes a run of SPECint and SPECfp by C't magazine. This is quoted from the comp.arch newsgroup (see groups.google.com) The text of the actual C't article is available from http://www.heise.de/ct/english/02/05/182/

"c't 5/2002, pages 182-183 publishes SPEC CPU2000 results for various
Apple G4s and a Pentium III/1000 (all with 512MB RAM in single-user
mode). Here they are:

C compiler SPECint_base SPECfp_base
G4 800 gcc 242 147
G4 867 gcc 259 153
G4 1000 gcc 306 187
P3 1000 gcc 309
P3 1000 MSC 236

The Metroworks compiler produced slightly worse results than gcc. The
gcc version used was gcc-2.95.[23] on all machines. The compiler used
for the Fortran benchmarks was Absoft Pro Fortran 7.0.

The article lists the individual results for SPECint_base; The G4 1GHz
is significantly faster then the P3 1GHz on 176.gcc (444 vs. 286) and
300.twolf (522 vs. 340), and significantly slower on 186.crafty (334
vs. 444).
"

Note that a) the P-III is slower than the P-IV. b) gcc isn't a great compiler. But it was the fastest available for these tests on MacOS X. Using a commercial compiler, SPECint base on the fastest P-IV is more than twice Apple's score, and the Dell 340 is three times as fast on floating point.

GCC 3.1 helps out the G4. But it won't be in wide use until at least the release of MacOS 10.2.

Don't get me wrong. I like Apples. I bought a brand new TiBook a few weeks ago. But I didn't buy it because it was overwhelmingly fast.

Yours truly,
Jeffrey Boulier

One thing that this article seems to ignore which is very important is the software of the various systems, an area I believe that apple has a clear advantage. Most customers looking for an entry level server system cannot afford the most skilled system administrators and thus having reliable and easily configured software is essential.

By building OS X on a Unix core with well tested open source software (Apache, Samba, PHP, MySQL, Tomcat, OpenSSL) apple has created a rock solid, full featured server platform. Beyond this they add their own QuickTime Streaming Server and WebObjects software to increase the power of the system. Most importantly however is how easy apple has made it to configure and monitor such a system. As they put it the Xserve adds "legendary Apple ease of use that makes it possible even for non-techies to manage a large server farm", somthing that the other entry level servers cannot claim with the validity apple can.

It seems to me like the Xserve is showing what may be (to Apple's suprise) the strongest point of OSX, its ability as a server platform. By combining the power of a Unix server with the ease of use of the Mac may lead apple into a whole new market where the profit margins are much larger.

... but this stands out:

All in all it is a good value considering that the G4 1GHz can run circles around the P4 2GHz.

Nice try, troll. Next time, don't make your lies TOO bad. Maybe a half truth like 1GHz G4 = 2GHz P4 or maybe even the truth: 1GHz G4 ~= 1.7GHz P4. But 'run circles'? Too obvious of a troll.

> All in all it is a good value considering that the G4 1GHz can run circles around the P4 2GHz.

Sean, get real.
Please consult previous OSNews stories about some benchmarks done against dual and single G4s against P4s and Athlons.

until it ships why even have this discussion?!?!

Pull more crap like this and I'll stop reading this "news". Call it what it is--a 7 foot heap of speculation straight from the asses of zealots.

ugh

I recently got a new iMac G4. I did not get this because I wanted the fastest computer on the block, nor a screaming powerhouse. I'm still in college so I move several times a year and the computer's easily transported. I love the LCD screen (easy on the eyes), although I would've liked a higher max res. I got the top 800mhz model and my old computer was a 600mhz PIII. Now obviously it is faster than my old computer; but the difference is far from staggering. But this is also true of faster Intel machines. The only places I've gotten a real speed boost are in encodings and compiling. I got my Mac because I wanted to try something new and I'm glad I did. I've developed an attachment to Apple lately that I never had with the PC platform. That being said, I still feel that anyone severely biased over these things is a moron. The power of average PCs and Macs seems to be relatively equivilent, and bickering won't get you anywhere. People should just use what works for them and not get their panties in a ruffle when someone else doesn't get the same experience out of it.

The article mentioned the following:

[i]The first thing to note is the outrageous pricing of the Compaq servers in this and future categories, which is literally more than twice as much as the next most expensive machines. I can not see any reason for their machines to start at over $12000--without any software installed no less! [i]

There are a number of reasons why server equipment is more expensive than expected but the biggest is redundancy. The compaq servers that are priced so outrageously have features such as dual power supplies, hot swap drive bays and most importantly on-site support. This is the sort of thing no company can do without.

Unfortunately your story made no mention of these sort of features existing on the Apple offering. Without them the only market for these Apple Servers are small companies that like to live on the edge.

All in all though, these servers will fill a demand I think.

Cheers
David

The fact is that people shouldn't talk about what they don't know. I'd bet that most of you don't even have a bachelors degree nevermind a masters degree in the field.

Benchmarks are never accurate unless comparing the same processor (just like clock speed).

SPEC is hichly influenced by intel in order to get good results for their processors (it isn't multiprocessor enabled which everyone is under the impression of).

Someone said that the clock speed myth was a red herring. Then why is intel marketing itaniums at 800MHz as their most powerful processor?

As much as everyone likes clock speed it can be misleading. The P4 has a 21 stage pipeline while the G4 has a 7 stage pipeline (shorter is better). One way of attempting an increase in clock speed is to increase the length of the pipeline. The problem is that each instruction has to go through each stage. This works beautifully if the computer is able to anticipate the instructions that will come after it, but most of the time this is not the case. The finished instruction will make everything in the pipeline irrevalent and the pipeline will be cleared and an instruction will start at the beginning. Having a full pipeline means that your processor is running most efficiently and a RISC architecture helps this. Intel's high clock speeds help to offset some of the problems of a pipeline since it is able to fill up faster, but its length is still quite long.

Processor design is very complicated and none of you understand anything about it (which is what Intel wants). GCC is absolutely atrocious when it comes to the PPC processor (there are many other compilers that are much more efficient). The fact is that people who know what they are talking about know the superiority of processors like PA-RISC, UltraSPARC III, PowerPC, Power4, etc. The problem is that even well educated people don't understand more than they hear in an Intel commercial.

Oh, and Jeffrey Boulier the Itanium is much faster than any P4. The problem is that people look to sources of information that are either partial or ignorent. YOU seem to read too many macosrumors articles (talk about ignorence).

>>what it is--a 7 foot heap of speculation straight from the asses of zealots.

<<

Maybe some people enjoy the heap, or don't think there is anything wrong with the heap. I just don't care ether way about the heap. But I will take a picture of it while looking at it.

On a whole I think the XServe's look very nice and from the article seem to compare favourably with x86 offerings.

From a marketing standpoint they do seem impressive. Dual gigabit ethernet controllers and 4 individual ata controllers.

This is where I am not sure about. Taking just the Ethernet controllers and ATA channels in consideration this equates roughly to over 600 Mbytes/sec (400 drives + 200 ethernet).

Now I am not sure how Apple BUS architecture is designed but if it's anything like general x86 setups there is a north bridge and a south bridge. The south bridge usually linking into the disc controllers, pci bus and onboard ethernet devices while the north bridge talks to the CPUs memory controllers and AGP. Now if apple do use similar chipset laytous to x86 is the interconnect between the north and south bridge large enough to cater for all these bandwidth intensive devices (4 x hard drives, dual gigabit ethernet controllers).

Also I read somewhere on apples page that while the Xserve has DDR ram it said somewhere else that is has a 1 GB bus to ram. Does this while the Xserve is equipped with DDR it can't take advantage of it? Does this mean that the Xserve while impresive specwise is more about marketing hype then technical superiority (ie. the wow factor hardware like gigabit ethernet cards let down by inferior bus architecture)?

If anyone knows aything about these issues I would be interested in knowing.

"Operating system costs, even for many non-Windows options, can be anywhere between $700 and several thousands of dollars."

argh! is this osnews?
who pays anything for a proprietary server os when you can get a free one for free?

Whether or not GCC is an 'atrocious' compiler for the PowerPC is largely irrelevant--it's the only one in wide use on the Mac platform, and for all practical purposes it's the "Apple anointed" compiler. The performance of Visual C++ on Windows is more important than the performance of ICC from a user standpoint for the same reason: the OS and nearly all the applications are made with Microsoft tools these days, except for the handful done with Borland's. It's cruel to say that ICC's chief use is for compiling benchmarks and entertaining engineers (and giving OS News editors and readers argument fodder), but it's probably not too far from the truth, either.

Back in the mid-80s most people who really knew processor design would tell you that Zilog's Z80, and Hitachi's enhanced HD64180 version, had better design than Intel's 8/16-bit 8086 line (up to the 80386SX). When the Z80/64180 was at 4-6 MHz and the x86 was at 8-12 MHz, the 64180 in particular could still be faster. When it was 6-9 Mhz versus 20 or 25 MHz, well... you could live with it. When it was still 6-9 MHz, maybe 12 MHz, versus 66 MHz, forget it.

Of course clock speed isn't everything. But it's disingenuous to suggest that CPU design can come close to making it irrelevant. An engine design that's brilliant and even groundbreaking from an automotive engineering standpoint may still get its ass kicked by a Mustang GT in a race. Likewise, Motorola's elegant, efficient CPU design makes their CPUs more elegant and efficient, but in the real world, inelegant, inefficient designs can be faster.

Isn't this like comparing pictures of burgers in front of McD's and BKing.
They both are made to look attractive.
They both look like Burgers.
but one is flame broiled.
how do you compare.
Besides they both suck.

To "gz".
Apple systems do not use the same system layout as generic PeeCees.
The past generation of G4's was using something called the MaxBus architecture.
What it is technically I don't know, although from what I've gleaned it's a more efficient way of distributing bandwith resources then the usual N/S/PCI setup.

I'm hoping that this new generation of systems are based on Hypertransport as well as RapidI/O.
But no mention of this is anywhere in the Apple Literature, and since Apple is a HyperT and RapidIO "Syndicate" member I would think they would advertise these technologies;
so it's an open question until we see which side of the toast has butter on it.

Just like everything else in this discussion.

To Sean.
Please don't.
I love Mac's, I love PPC, but no, we ain't faster, just because you say so.
G4 is good but not that good.
If it was there wouldn't be constant discussion on it's merrits; which implies that G4's are not showing a clear advantage over P4's.
Motorolla suxs and Apple needs to deal with them now.

PS: G5 is only a distant dream, which like many dreams might never happen, so fellow MacEvangelists, don't go there, as you will get sh*t on your face.

As long as Apple is dealing with the sinking ship which is Moto's CPU division, we got to be praying for an act of God to make PPC's desktop cpu future look rosy.

PPS: PC users; K7/8 will not help any, because it will take longer to move all software/people to a completely new Arch, than there is time left in the future dev of CMOS technology, (which incidentaly is supposed to stop being usefull at the same time as the Aliens are supposed to invade (2012)).
I think Apple should contract AMD to make Apple branded PPC's that will fulfill Apple's needs; but this would also take to long, so I see IBM re-becoming the prefered CPU vendor sooner than later.
Ciao.

1)
If the bottleneck was in the processor, and not in the memory system, then we can reasonably expect these G4's to perform on par with a mid-1.5GHz Pentium chip on normal operations, and maybe up to a 2GHz Pentium on certain optimized floating point operations.

Completely wrong..
Altivec helps a g4 1 gig keep up with a p4 1.7.. if your useing floating point operations it wont come close not a chance. You mean if they used vectorised altivec code.. this is differnet from "floating point operations" which implies using the fpu .. which ath and even p4 kick g4.

Benchmarks show this is part of the "reality distortion field".
Some lightwave benchmarks?
http://www.blanos.com/benchmark/

After Effects Showdown: Mac vs. PC
http://www.digitalvideoediting.com/cgi-bin/getframeletter.cgi?%...

GFlops even a k6 (BEFORE Athlon) 350 mhz machine can get up to 1.4 GFLOPS!
http://aggregate.org/EXHIBITS/sc98.html
As to apples marketing .. be afraid .. be very afriad.
Another real world benchmark with multimedia (u know the thing that allways mac goes faster at? WRONG AGAIN)
http://www.ableton.com/forum/viewtopic.php?t=538

If anyone can actually find me a benchmark other than distibuted.net which relies on 1 instruction that altivec does but not SSE then pls post it here.
EVERY REAL WORLD BENCHMARK ON THE NET SHOWS that a g4 1 gig is similar to about a 1 gig Athlon or 1.4 gig p4.. the dual gig 1 g4 (being 2 processors not 1) is of course not as good as a single p4 2 gig + or athlon let alone dual proc for pc.
2)Since the G4 in the dual processor Apple workstations, was recently defeated in multimedia benchmarks by a dual 1.5GHz Athlon, it will be interesting to see how the Xserve will benchmark against the single and dual processor Pentiums with their new memory subsystem.

So do some dam benchmarks.. get one .. run apache and linux .. and apache and windows and apache in osx.. youll see how the ddr ram cant really save apple. The pcs will win.. u might be able to find one of the photoshop benchmarks .. say the 5.01 (the one made up by apple marketing department) to make the g4 win! u might!!

Unfourtunatly ppl do not understand.. most code is normal CPU code .. so similar Mhz between g4 and pentium 3.. with p4 and ath they seem to give more cpu performance per clock than g4.
Altivec is better than SSE but not as good as SSE2 (altivec dosent support 64 bit numbers) and ends up performing similarly per clock to 3dnow and sse.
Lets get to the point.. altivec is about multimedia only and in multimedia benchmarks when 3dnow or sse2 are used clock for clock they are very similar.. so would u rather a 2 gig machine.. or a maxed out 1 gig.. because the g4 technology has allready been pushed so far.

3) In terms of raw performance there is a great deal of benefit to be gained from using the DDR RAM.

not reall look at a benchmark for gods sake.. sure it helps a athlon 2 gig .. a bit.

4)it could provide a truly impressive disk I/O performance advantage for the Xserve, especially since the four disks act as one large disk via software RAID software built into the OS.

what like same RAID that was invented for SCSI?

5) Its a good price and built in gigabit is good.

Thats all.. oh but web serving and networking arent going to use altivec now are they?

Dan will you set up the free OS for free, for me ? ;-)

Thoems

I find Xserve overpriced. For half the price of the single processor version you can build a dual Athlon 2000+ MP with as much RAM and hard drive capacity in a U1 rack. And it will kiss Xserve ass all around as far as processing power go. And with FreeBSD, it won't be much different from console-mode MacOS X

Here is some RC5 benchmarks posted a few months ago from a technophile on one of the old Mac tech sites...

RC5 CPU Comparison:

Processor________________MHz___CPU___Key Rate

PowerPC G4_______________1000___2___21,129,654
DEC Alpha 21264___________725___8___11,536,680
AMD Athlon Thunderbird___1533___2___10,807,034
Intel Pentium 3__________1333___2___7,559,280
Intel Xeon_______________1000___2___5,835,597
Intel Pentium 4__________1800___2___4,870,420
Sun UltraSPARC III________750___2___2,977,968
Sun UltraSPARC II_________450___2___1,458,333

So here is another look at another benchmark to add to the too many benchmarks we already see. RC5 is another well respected benchmark in the industry as well. To be honest like the other Mac users said in this thread, I also didn't purchase my Mac due to some story about it being a speed demon, I went due to other reasons. I will say that Macs (as well as Sun boxes) will outshine PCs clock for clock from my own experience and daily use of the machines at work and at home. I don't care either way. I am more interested in the total package and Macs appeal to me more in this sense than PCs. I own both platforms, so I can speak without prejudice and/or lack of experience. A good number of Mac users have PCs as well and people don't seem to realize this and try to come to some conclusion that Mac users don't use PCs because of some inferiority complex or whatever they can think up as an argument, zealotry and/or excuse, that is not the case. Not everyone will agree on what is best and what is not... guess what I don't care, I use a Mac and nobody is going to change my mind otherwise, end of story!

Note that a) the P-III is slower than the P-IV.

Actually the P4 is the first Intel CPU to be slower per clockcycle than it's predecessor. IIRC a PIII 1 GHz performs about the same as P4 1,3 GHz - raw power, no CPU-uniqe instruction sets utilized.

Apple and Motoral are both part of AIM, right? The "I" is IBM. ... So, why is Apple "married" to Motorala as its sole source of processors? Does IBM just flat-out only make POWER chips? Couldn't they be talked into providing G4/G5 equivalents? That's something I've never really understood about the workings of that alliance.

I always enjoy friendly discussions on these boards. This is the first time, it comes from something I wrote. I figured I'd respond to some questions/statements all in one e-mail:

Sean: The ATA busses in the Xserve are ATA/100 for a combined theoretical peak of 400MB/sec.

No, the maximum theoretical bandwidth of the ATA system in Xserve, as quoted by Apple is 266 MB/sec, as shown on their website: http://www.apple.com/xserve/storage.html

Sean:The RAID in most Macs is software RAID, but Apple has included a hardware RAID controller for the Xserve.

Apple does not have a built-in hardware RAID on Xserve, but will ship a hardware RAID module later this year. To quote Apple's documents on the Xserve disk system: "Apple Drive Modules integrate seamlessly with Mac OS X Server, delivering software RAID support and built-in remote monitoring capabilities."

Sean:All in all it is a good value considering that the G4 1GHz can run circles around the P4 2GHz

Comparisons using mainstream software applications, not more abstract benchmarking system, do not hold up this performance claim in most operations. This is especially true in non-AltiVec enabled applications, or applications where AltiVec acceleration makes no sense.

tantalic: What about the software?
Without actually using all the different pieces of software on each of the systems, I did not feel it a point that I could investigate very deeply. I therefore mentioned the software, but did not do any in-depth reviews.

strobe:until it ships why even have this discussion?!?!
From looking at raw information, it is possible to project if there is any hope of the Xserve meeting expectation. This was the purpose of me writing this article. Many times I state that benchmarks and real world performance will be the final deciding factor, but that just looking at what the system is made of can give us an indication of if the Xserve has any prayer of living up to its hype.

David McPaul: There are a number of reasons why server equipment is more expensive than expected but the biggest is redundancy. The compaq servers that are priced so outrageously have features such as dual power supplies, hot swap drive bays and most importantly on-site support. This is the sort of thing no company can do without.

This argument doesn't hold water. First of all, the low end configuration of the systems are not double or triple the cost, as it is once you start adding the additional processors and hard disks. Second of all, ever machine reviewed had hot swappable drives, hardware monitoring and other features for minimizing down time. I'm not sure about redundant power supplies, but if that was the reason for the extra cost, then it should have shown up in the bare-bones configuration as well. Lastly, the support contracts are not standard on Compaq or any servers, and are not in the prices that you see reviewed. Again this was done out of fairness, since there is not enough information on the websites for me to make a valid comparison.

David McPaul:All in all though, these servers will fill a demand I think.
I agree 100%

in your phd program. but, the point is correct, that specific comparisons are necessary, though processor specific is a bit much. the whole point is comparing different processors.

Someone said that the clock speed myth was a red herring. Then why is intel marketing itaniums at 800MHz as their most powerful processor?

Intel hasn't really been marketing itaniums as their most powerful CPU yet, in that the initial run (which was extremely low volume) was more oriented at porting applications.

As much as everyone likes clock speed it can be misleading. The P4 has a 21 stage pipeline while the G4 has a 7 stage pipeline (shorter is better).

That is a overgeneralization. I can build a single cycle core but it's performance will be crap.

One way of attempting an increase in clock speed is to increase the length of the pipeline. The problem is that each instruction has to go through each stage. This works beautifully if the computer is able to anticipate the instructions that will come after it, but most of the time this is not the case. The finished instruction will make everything in the pipeline irrevalent and the pipeline will be cleared and an instruction will start at the beginning. Having a full pipeline means that your processor is running most efficiently and a RISC architecture helps this.

This is misleading. You assume Intel/AMD's chips are still CISC and 1 CISC instruction per pipeline segmen. However, this hasn't been true since the Pentium Pro cores. While the chips appear to be CISC, they are not internally. The pipelining is done on the RISC instructions needed to perform the operation.


Processor design is very complicated and none of you understand anything about it (which is what Intel wants). GCC is absolutely atrocious when it comes to the PPC processor
GCC is atrocious when it comes to any processor, and there exists compilers that regularly outperform it on any platform. However, I think apple is trying to achieve portablility, thus the big push for GCC.

The fact is that people who know what they are talking about know the superiority of processors like PA-RISC, UltraSPARC III...
Indeed, but the problem is those are unaffordable for the most part on the home system, tis why the argument keeps coming back to the horrible x86 line and motorola bumbling with it's G4/G5 chips. As much as I'd like to buy an ultrasparc machine, I don't see dropping a grand for a UltraSPARC IIe as a great investment. Nor do I see shelling out for a underperforming, undercached G4 as a valid one either.
I don't like the x86 processor, and I'm not bound to it by software, however, at this current moment it is the best investment (unfortunately).

There's no point in me saying whether G4 or P4 is faster. But the fact of the matter is a system built with x86 hardware is cheaper than Mac hardware, and if you play it right you can get as much performance if not more out of it. Clock for clock, the G4 may be better. But clock for clock, the G4 is more expensive. However, the cost to maintain an x86 based servre may be more than the cost to maintain a Mac server. But for the people who are capable, they see it as x86 with the advantage.

Many people are saying SPEC is not a decent benchmarking tool, many more base their opinions on Photoshop benchmarks, which are completly ridiculous because Photoshop does not show consistent results even when testing over and over on the same machine. Rather than using technical factors to evaluate CPUs, I like to look at how much bang I can get for my buck. I think it's great that Apple invests so much effort into designing their systems, that includes efficency, design of enclosures for quietness, and what not. But if they would but the same amount of effort into cranking out more power, we can honestly say that PowerPC is the dominant platform.

I'm a little offended by Sean in his statement about most of us not having Bachelor or Masters degrees in comp sci. You don't have to be a geninus to know what to do with your money. And if Sean says you do, then please explain why hundreds of thousands of employees with masters degrees and maybe more, working for a company that has billions of dollars, can screw up so bad like Microsoft?

#1 Even the "free" one's aren't always free
#2 Look at the net stats. Since the overwhelming majority of servers are *not* Linux, and the big ones that are probably have service contracts behind them, I'd say quite a few people pay for proprietary server OS's.

The ethernet cards are not on the same bus as the PCI bus and the hard disk bus. The listed bandwidth for the PCI/ATA bus is 566 MB. This can be seen in the graphic at:

http://www.apple.com/xserve/architecture.html

Their downloadable PDF file goes into some raw number descriptions too. I found it very informative.

Sean, based on "Author's Response #1" and the documentation it provides refuting three of your assertions, I wouldn't brag too much about your educational superiority to everyone else here. Nobody knows everything, and that includes you.

I also noticed that you can't spell "ignorant" (your version is "ignorent"). Maybe you should have studied more English in college.

Sorry all, but I can't stand arrogant, insulting academic snobs.

Xinet looked at some servers, including Mac OS X server (on a REGULAR G4, mind you, not on the new XServe rack-mount):

http://xinet.com/benchmarks/benchmarks.2002/

FWIW: The mac does pretty well.

Glenn Sweeney: "Altivec is better than SSE but not as good as SSE2 (altivec dosent support 64 bit numbers) and ends up performing similarly per clock to 3dnow and sse. "

Actualy the new G4's have a 128bit engine which is still better than SSE2, no matter how you look at it.
Yes it does seem like the P4/K7 at their current speeds beat the current G4, but if they where running at the same speed it wouldn't look to good for the intel world.

Also Glenn: "Lets get to the point.. altivec is about multimedia ..."

No Altivec is about the whole experience, and is very important in the markets that Apple serves, much more so than SSE is to intel users.

altivec isn't comaprable to SSE in any way really because not only is it's general computing power double that of SSE2 or 3dnow, it's design also doesn't require a cycle jump like that of the Pentium IV/III.

To betcour: "overpriced"
Why do people like you not get trough your head that you can't compare a real car to a Kit Car.
I can also build a space ship from russian parts that I can buy on eBay for very cheap.
first of all I looked and looked and couldn't even find a cheap 1U rackmount case anywhere. 2nd all of the energy necessary to power your dual Athlon puppy, could power 4 of these Xserves.

PS look at these interesting benchmarks, not very positive for the Mac but important none the less.
specificcally check out the osx tab.
http://xinet.com/benchmarks/benchmarks.2002/

I always hoped that the benchmarks would show the Xserve putting out comparable, or even better benchmarks. If these short benchmarks are any indication, it sounds like it might very well do just that!

Howdy:

Re: ealm
Actually the P4 is the first Intel CPU to be slower per clockcycle than it's predecessor. IIRC a PIII 1 GHz performs about the same as P4 1,3 GHz - raw power, no CPU-uniqe instruction sets utilized

I think this is where we get into the "reverse clockcycle" myth. While the P4 doesn't get as much work done per clock as the P3, especially if you are dealing with older code, it clocks far faster. So despite it being less efficient per tick, the P4 IS the faster CPU.

Consider the original IBM POWER, which capped out at 60Mhz or so, but could manage a ridiculous amount of work per clock. A vastly less efficient UltraSPARC-IIe is the faster CPU "simply" because it can run at 600Mhz.

SEAN> Benchmarks are never accurate unless comparing the same processor (just like clock speed).

I've thought that the whole point of benchmarks was to allow someone to compare different processors, architectures, operating systems, compilers, etc.

I have a Wang VS 5640 and a SPARCstation 20 sitting on the shelf next to me. I have a COBOL program of which I would like to run multiple iterations. It compiles on both systems with minimal effort. It takes eight minutes to run on the Sun and forty-eight on the Wang. The output produced is the same.

Should I refrain from saying that for an application like that COBOL program, the Sun is much faster? After all, I'm comparing two completely different systems; the Sun uses a RISC SPARC running a Unix operating system, the Wang's CPU is CISC with a somewhat 360ish instruction set running the decidedly non-Unix VS OS. Without the use of that COBOL benchmark, how can I compare their aptitude for my workload? On looks? Based on the fact that Wang is a way cooler name than Sun, and had marketing slogans like "My Wang works wonders"?

Sean> SPEC is hichly influenced by intel in order to get good results for their processors

Yes. They are also highly influenced by Sun, IBM, SGI, AMD, and scads of others. Remember that Intel doing well on SPEC is a very new thing. Aside from the Pentium Pro surprise, they were a perpetual bottom feeder until quite recently. This may have something to do with Intel being able to spend tons of money on CPU design and state-of-the-art fabs.

There are some interesting discussions about the politics and decisions of SPEC on the comp.arch newgroup. Check out groups.google.com.

Sean> (it isn't multiprocessor enabled which everyone is under the impression of).

Hopefully this just means the poor benighted masses who don't read OSNEWS. We know that SPECrate is the multiprocessor equivalent.

Sean> the Itanium is much faster than any P4. The problem is that people look to sources of information that are either partial or ignorent.

Most of what I have read indicates otherwise, including from the Intel-dominated SPEC consortium. It is certainly possible that it's innaccurate. What specific applications do so well against the P4? Why hasn't the Itanium done very well in the marketplace so far, given its tremendous speed? Is everyone just waiting for McKinley (the osbourne effect)?

Sincerely Yours,
Jeffrey Boulier

Seabass,

You say that the new G4s have a 128-bit engine, but does it do double precision floats? The person who said that the G4 doesn't do 64-bit numbers isn't entirely accurate. I guess what he/she meant to say was: "The G4 can't do double precision float numbers." Because that was true when the G4 first came out, I don't know about the new crop of G4s. In fact, the AltiVec extension had 128-bit registers from day 1. It just means that you can store up to 4 32-bit floats into each AltiVec register, it doesn't mean AltiVec handles double precision floats.

Has anybody verified the pricing on the Compaq DL360 G2 mentioned in the article. According to the HP website, a DL360 with dual 1.4P3's, 512mb ram and an 18.2gb disk is only $4k. That's still more than the XServe, but only about a third of what the article mentions.

Nothing that I can see shows any change in the AltiVec units themselves. Just like MMX and 3DNow, they all do 32-bit floating point only.

It's funny, the G4 actually wins awards in the embedded world because of it's speed being so high! Embedded does not automatically mean slow, Ericsson are using G4s in GPRS equiptment (base stations not the phones) and these will need a lot of raw power. They had previously done a multibillion dollar deal to use Alphas. Then Compaq killed the Alpha - Do'h!

However put a G4 into an Apple and suddenly it's swimming in treakle, why?
I think there is firstly the memory bandwidth issue (which has now been taken care of) but also OS X seems to cause the system to slow down badly, the compiler will have an effect also (SPEC marks are strongly effected by compilers).

I'd like to see some benchmarks run later in the year when 10.2 is out with some gcc 3.1 compiled code, I think the G4 will not do so badly then.

>PS look at these interesting benchmarks, not very positive for the Mac
>but important none the less. specificcally check out the osx tab.

I did this is one thing it say:
>On a per-processor basis, it was the fastest processor Xinet tested,

thats not very positive?

It also said this:

>By itself, the UFS filesystem did not perform well, but
>adding a RAID masked this behavior and greatly increased write >performance.

Part of OS X holding the system back as I suspected...

DrP wrote:

Sorry all, but I can't stand arrogant, insulting academic snobs.


Don't worry; I don't think he's an academic. You'll note that our "esteemed" friend never claimed to have a degree, either, and I can't imagine he would, since he made such blankets statements as "shorter is better" for pipelining. For the "non-academics," anyone who saw the Macworld NY this past year or read the Ars Technica article a few months before that would have seen exactly where he got his information. The funny part is that he managed to mangle a simple concept while chastizing everyone else for being stupid.


As for the Itanium, I know Intel's going to push the new version soon enough. However, a few weeks ago one of the chief designers there came to my compilers class and was showing how some of the features hadn't paid off very well (granted, it was a somewhat cursory introduction, but still...). So I don't know if it's exactly fair to assume that the previous generation of Itaniums are necessarily faster than the current generation of P4s. At the very least, our "enlightened" friend didn't provide any evidence for his claim, which is hardly surprising.

Wow, the marketeering hype is flying fast and furious here! Lots o' allegations and spin, but nothing that would really benefit someone who would use something like that. So what is the real deal?

Well, despite the weasel words and mumbo-jumbo, you're not getting any bona fide UNIX system -- you're not going to be able to install and run your favorite AIX programs, for example. So what's the point? The bottom line is that you're either stuck with what software Apple has to offer, or if you go the free software route you're paying more money for a platform that's less capable than a commodity x86 box.

Given the choice, I choose to spend less and get more. What kind of bonehead would want it the other way around? People who waste money stocking data centers will find the difference gets made up out of their salary after they're fired. So what is it that would keep me from being out on the street if I recommended one of these boxes?

Generic services like HTTP and FTP are easily done, but Linux/x86 wins hands down for the most bang for the buck. So what's left? Application services? I don't see any for Mac/OSX! Where are they? About the only thing that makes this box attractive is that it could replace the hodgepodge of Macs that litter a typical Mac shop's "server room".

When it gets down to brass tacks, the only reason someone's going to buy one of these things is because:

1. They buy Apple stuff as a knee-jerk reaction.

2. They buy stuff for people who buy Apple stuff (as a knee-jerk reaction).

Not terribly exciting, is it?

I'm going to have to take the blame on those Compaq prices. I went to their site, and sure enough was able to get the numbers I listed. There's just one problem, I was looking at the "3 Pack" configuration...read three of them. Didn't see that. Chalk it up to late night working after a long day. Scary thing is I double checked my numbers and made the same mistake twice! So, the new price range for the Compaq's on the mid range server are $4077-4852, and on the high end $6420-8200. That doesn't change the final conclusions, however. Sorry for the confusion.

I think a major part of the equation that is missing is the SCSI RAID system Apple announced but won't ship for a while. Jobs mentioned they will release a high performance SCSI RAID in a separate cabinet, I saw a picture somewhere, it looks about 3 or 4 units tall. If you go poke around on the Apple Store, there are interesting options to add high speed SCSI cards, obviously they're primed to just plug in massive storage units. I think most people will just set up their own low cost RAIDs from whatever suits their application.
Also there are interesting options for dual gigabit ethernet, and ISTR reading of fiber channel options coming soon. I want it!
I've been playing with MacOS X for servers and I think it has absolutely great features. The Perl implementation is excellent. I've seen many great MacOS X sites running PHP, JSP and other tricky web services with ease. Apple has great ties to the open source BSD community and there have been some recent improvements to many core programs, even GCC itself, that will lead to some awesome improvements in the short term future. And don't forget: no server taxes.

Wow, the marketeering hype is flying fast and furious here!

You've done such a wonderful job of summarizing this board that you even summarized your own post.

Let me humbly summarize some points that other people have made here and other places, about why people might want one of these things.

1. Price. (Hardware) I defer to the article.

2. Space. They're tiny. And unlike so many x86 servers, they won't overheat at max load. (My lab's first rack of dual P2's was effectively nonfunctional due to heat concerns. Yay segfaulting perl scripts!)

3. Price. (Administration) Windows is expensive. Hiring a Linux or BSD administrator is expensive. Obviously Mac OS X won't always be cheaper to administer, but will it be more expensive?

Reasons not to buy an XServe:

1. Doesn't fit your needs. There is no magical uber-computer that fills every need and costs $0.42. That doesn't mean that the XServe is useless.

2. Macs make you puke. Understandable, but no self-respecting geek would make technical decision based on emotions.

Sorry to be flamy. I get that way when I run into kneejerk posts that invoke "brass-tacks." (Talk about stuck in the past...)

Hmm, I smell trolls from both camps, but this one caught my eye which hadn't been discussed yet...

Well, despite the weasel words and mumbo-jumbo, you're not getting any bona fide UNIX system -- you're not going to be able to install and run your favorite AIX programs, for example.

Can you prove this? Linux itself isn't a bone-fide Unix system either, but it apparently has a lot of attractiveness lately in the business market. Also, just because Apple slapped on their own proprietary GUI to a modified Mach3 with BSD interfaces (calling it 'xnu'), doesn't mean you can't recompile quite a few apps. I have a rather sizable swarm of non-OSX apps running daily under XFree86 4.2.0 or the CLI. Sure you can't do a 1-to-1 port of some low-level apps, but that is true with the multitude of 'Unix' systems out there. Heck, I didn't expect TinyFugue to compile straight over since the code is so old, but it works like a charm.

So what's left? Application services? I don't see any for Mac/OSX! Where are they? About the only thing that makes this box attractive is that it could replace the hodgepodge of Macs that litter a typical Mac shop's "server room".

Hmm, Application Services? Which ones are you looking for? I can probably point you to a place where to look or someone to complain to for not being on the ball and supporting their customers.

When it gets down to brass tacks, the only reason someone's going to buy one of these things is because:

1. They buy Apple stuff as a knee-jerk reaction.

2. They buy stuff for people who buy Apple stuff (as a knee-jerk reaction).

Hmm, you have a point... that is if you are referring to the big guys. This thing isn't really aimed at the big guys, but the small guys who can't afford the staff required to keep a real Linux box supported with the more expensive IT guys. Apple's key pushing factor is ease/cost of support, not for blowing everything else away in performance/dollar factor.

Overall, I wouldn't mind one, but that is because I don't have the time to keep Linux or Win2k under my watchful eye and constantly tinkering it for my users. A little ease of use does go a long way to some people, even if there is a trade-off.

Since you like raw power and are used to something other than what Apple is now offering, be my guest to use what you want. Apple is offering a choice, not a demand. You can take it as it is or leave it. Just don't bash people like me for our choices in computing, especially since I haven't done the same, and have given money to both major commercial players (OSX/WinXX), and worked with the major free player (Linux).

(I'm speaking here as a SysAdmin, primarily of FreeBSD and Sun boxes, who uses OS X on a TiBook for most things. This is why I wouldn't buy an Xserve for general purpose use.)

No ECC RAM. This is one of the biggest omissions IMHO. As far as I'm concerned, without ECC it's not even in the running.

Only 2GB of RAM. 2G isn't much, these days, particularly if they want to be moving into the Oracle market. I would have expected 4G minimum.

No SCSI option. I'm far from a SCSI bigot, but at the end of the day if boatloads of disk I/O and random-access disk patterns are what you have, SCSI is faster. If all the local disks are going to be used for is booting the machine or perhaps some low-end fileserving, then ATA disks are fine - but it would have been nice to see an option on the high-end machine to swap the ATA drives for SCSI drives. I consider this particularly relevant to Apple's apparently upcoming foray into Oracle territory.

No hardware RAID. Again, I consider this a rather large omission as it *seriously* limits the amount of "useful" disk space available. AfAIK OS X doesn't do software RAID5, so you're limited to either a RAID1+0 (with some space subtracted from each drive for the system, since OS X won't be able to boot from a RAID0 - or a RAID5 for that matter, should it be added to Jaguar) or an optimised-for-failure RAID0 (again, losing some space to the system). Even one of those halfway-hardware-RAID chips put on many PC motherboards would have been sufficient, as it at least manages to make all the devices appear as a single drive to the OS, but ideally they would have used one of the existing "real" IDE RAID cards like 3ware make. From where I'm standing, these things top out at around 200 - 220GB of usable space. Other people may be prepared to risk the fourfold increase in risk by using RAID0 over all the drives, but I wouldn't be.

Apart from those things, I think the Xserve is an ok deal, depending on your needs. The places I expect to see them popping up are:

* Data-processing clusters, in which case my RAID requirements above are largely moot (the ECC comment is still very relevant though). They'll be a good deal for this, assuming the required processing benefits greatly from Altivec. If not, a bunch of PowerEdge 1650s are a better deal.

* Low-end fileserving. Eg, to the small group of Mac users we have here, or for a small company. In this case the large amount of storage for the relatively low price is pretty good - although my above comments about the RAID aspect should be taken into account. Additionally, the (undoubtedly simple and excellent) management tools will be a real winner here.

For general purpose use though (eg file & print serving to a range of different machines) I'll stick with my PowerEdge 1650s running FreeBSD. On that note, I'll just point out a few things about the 1650s (my personal favourite 1U machine) that I see a lot of people making these comparisons neglecting:

Price. They start cheaper and mostly stay cheaper.

Processing power. For some things, the 1Ghz G4 is going to be faster. For most things, the 1.4Ghz P3 is going to be faster. For the most things, however, CPU power is largely irrelevant as the tasks are IO bound - in which case the SCSI on the 1650 gives it the edge.

Dual power supplies. Not essential, but nice to have.

Up to 4G RAM and ECC. It might be slower PC133, but even that is going to be faster than swapping. Plus it has ECC - essential for any non-toy server IMHO.

Hardware RAID. Very important - bumps the "usable" amount of disk space up to about 200G (3x73G drives) which puts it in the same ballpark as the Xserve in terms of "usable" storage. Given the 128MB of cache included, also nullifies most of the overhead of RAID5.

Free slots. Even with hardware RAID and dual GB ethernet, the 1650 still has two 64 bit/66Mhz slots free. I'm not quite sure what Apple are thinking with their combo AGP/PCI slot...

Support. The standard support with the 1650 is four hour onsite support. To get that level with the Xserve costs an extra US$950. I'm not sure what Apple's "standard" support is (anyone ?), but bringing the 1650 back to "3 years next business day" knocks nearly US$1800 off the price.

I banged together a 1650 that I consider equivalent to Apple's high-end offerings to compare prices:

1650 w/dual 1.4GHz P3, 2GB RAM, 3x73GB 10kRPM drives, RAID controller, 3yrs SILVER support (4hours onsite) : US$7873

Xserve w/dual 1Ghz G4s, 2GB RAM, 4x120GB drives, Applecare premium: US$7799

1650 advantages: more expandability, much faster drive subsystem, ECC RAM.

Xserve advantages: faster processors if you use something benefiting from Altivec, potentially easier administration.

Personally, I think the 1650 is a better machine for most tasks. But Apple has done fairly well with the Xserve as a first go. I look forward to the second generation which will hopefully address my concerns above.

One other problem which I haven't seen mentioned too often is the filesystem. If you have a 480GB volume (foolishly assuming RAID-0) and you are the least bit concerned about downtime, you are probably going to want a journaling filesystem to avoid the painfully long fscks you will have to endure if the volume were ever to go down hard.

SGI offers XFS for both Irix and Linux. On the Sun side of things, there's VxFS, offered as a 3rd party solution, but contrast that to Apple, which has no journaled filesystem solution in any respect. Your choices are UFS or HFS+. The UFS side of things doesn't even offer more modern conveniences such as soft updates.

Worst of all, Apple has Dominic Giampaolo, yet they haven't produced even rumors of having a journaled filesystem. How do they expect people to use these as storage servers when their filesystem choices are sub-par?

Actually Apple does support multiple file systems and is flexible to different configurations...

Local: HFS/HFS+, DOS/FAT, UFS, UDF, ISO

Network: AFP, NFS, SMB/CIFS, WebDAV

I think they have covered a pretty good range if you ask me. I think the journaled file system thing will comes, don't expect it overnight since they just hired Dominic Giampaolo not so long ago.

Tsk, Tsk.

Silly thing to say in a field dominated by non-degree types. Good way to wind up with trouble. You can be a sheepskin snob, but I'd recommend keeping it under wraps.

My finest engineer regularly stuns highly educated R&D types, and he has a high school diploma. He gets paid well, and deserves it. I couldn't run my high-performance, elite shop without him.

The best degrees are sometimes the Baccalaureate from The School of Hard Knocks, followed by postgraduate work at The College of Getting the Shit Kicked Out of You.

Oh. And exactly 16 angels can dance on the head of a pin, as long as it's a big pin, and the angels are small.

I wonder if that will coincide with the release of OS X 10.2 or with the hardware RAID option?

why do people think apple's will have easier administration, it's not as if their new management console has been in a proven position for a number of years, i'm very hesitant to even consider a system that looks like a giant kludge, seperate channels for everything is simply over engineering a problem that's been sufficiently handled elsewhere for quite some time. Apple is struggling to keep up, they see a trend with the adoption of linux and squarely punch it in the nose with their latest offerings, too bad the price is so high otherwise it might actually be adopted more quickly. Then again I'm not sure that's a good thing at all.

Just for fun I ran a test with gcc 3.1 on an IBM Winterhawk, and compared that with the results of the same program compiled with the IBM's xlc compiler. The test program was a bytecode interpreter written in C, running 10000 iterations of the dhrystone benchmark.

At highest optimisation the xlc compiled interpreter managed 950..1000 dhrystones/s, and the gcc compiled interpreter was about 50 dhrystones/s slower.

Not bad, I'd say.

Bascule: On the Sun side of things, there's VxFS

As of Solaris 7, Sun's implementation of UFS comes with journaling as well. Just add "logging" to the options section of your vfstab's UFS filesystem entries.

I see that some folks have answered my post, so I'll try to answer theirs.

Anonymous lists several reasons why we should buy the Xserve. Anonymous quoted me, so presumably read my post. But Anonymous apparently didn't notice that such ostensible reasons have already been shot down.

Apparently Anonymous isn't aware that the 1RU form factor isn't an Apple invention. Sorry, but Apple is the Johnny-come-lately to a market full of 1RU boxes, with many quality boxes at better prices.

Anonymous doesn't appear to appreciate that when it comes to price, lower is better. The Xserve is more costly for what it offers, compared to the commodity x86 competition.

And then there's the RDF falsehood that claims that Apple products are somehow much less expensive to maintain than anything else. I read falsehoods like this one phrased in the "everybody knows that..." style, leading the uninformed reader to believe that the claim is a fact, by appealing to ignorance. Nobody wants to appear unknowledgeable, and so on the receiving end, folks will happily nod and stipulate to what they don't know to be wrong, which when you know nothing is pretty much everything. (On the giving end, there is no shortage of people who are perfectly willing to show the world just how little they know, as evidenced by the ceaseless amateur rumblings on CPU design, and wild speculation based on a single line of advertising copy.)

I have little doubt that the "easier to maintain" fallacy is rooted in the "easier to use" fallacy. Back in the days when MS-DOS ruled, Macs may have been easier to use in comparison to DOS PCs, but those days are long gone. These days it's nothing more than a myth, having never been proved or otherwise legitimized. I think that the reality is that Mac users have a lot of trouble recognizing the WIMP, desktop metaphor on any other product than a Mac. They fail to understand that 95% of people have no trouble adjusting to a slightly different desktop that employs the same metaphor. But that's neither here nor there in this case.

Administering network services is a completely different animal, and the old unsubstantiated claim about ease of use in one area certainly doesn't carry over to a completely different one. One could be a naive and foolish consumer, and simply believe whatever they are told. I, of course, will not take that route. Because of this, I am lead to question the assertion. And what I find is that there's nothing to back up the claim.

Tough luck, Anonymous. Perhaps one day you'll learn that plastic "flashy trash" is not the stuff of quality, and reconsider your slight of things that you consider to be past.


Krevinek wants me to prove that OSX is not a UNIX system. Nice try, but the burden of proof is on the claimant. Let Apple and all their shills prove that it is. They cannot, and I know why. If you don't know, then you would be better off keeping mum, and perhaps learning something, instead of waging a bellicose fight against reailty.

Krevinek goes on to say that Linux isn't a recognized UNIX system either. Well Krevinek, I know that, which is why I never claimed that it was. Lay off the red herrings, k?

If I'm going to plunk down a few grand for a Rolex watch, I'm going to want it to be a real Rolex watch -- one that I can take into a Rolex dealership to have repaired, one that I can sell for a good price if I want to. A fake Rolex doesn't offer any of those benefits. Similarly, Apple's misuse of the UNIX trademark doesn't help me one bit. As a consumer, I'm inclined to get my own best deal, and not to act as a charity for dishonest vendors.

Krevinek also promotes the "ease of administration" fallacy, using buzzwords like "tinker" to imply that bona fide administrators don't really know what they're doing. Well Krevinek, if you have to have Appple to dumb it down for you, then you're not one to talk, are you?

My conclusion remains the same. Once you strip away all of the sophistry and other fallacies that surround Apple products, you find that the Xserve is not a very good deal. If you're an untalented person who runs a shop of Apple products, then the Xserve is the natural choice. Most of the "benefit" is really the lock-in to Apple's proprietary way of doing things, though. Give to Caesar what is Caesar's.

For those who don't suffer from xenophobia, I believe that a cheaper NAS box or server appliance would do the same job better and for less. For those who can administer their way out of a paper bag, something less proprietary and more standards-based is the better choice. I try and try, but I can't see any legitimate market for this product other than those foolish enough to spend money on the Apple brand and nothing more.

Anonymous wrote "There is no magical uber-computer that fills every need and costs $0.42. That doesn't mean that the XServe is useless." No, there is no magical computer, just the quite real ones that all do pretty much the same thing. That means that price is the only consideration. the Xserve may not be useless, but when you can pay less and get the same value, you're a fool. If you're doing it with your employer's money, you're also fired.

Like with all things time will tell. There are going to be businesses out there who will purchase and make great use of the XServer and there will be those who wont. There are going to be those who will find fault with this unit no matter how wonderful it is and there are going to be those who will defend it no matter how bad it is.

Sybase and Oracle are releasing the enterprise version of their databases sometime this Fall for OS X. Borland already has PowerJ for OS X. OS X is the only version of UNIX that MS Office has been nativley ported to. Despite what many people belive, tt is my belief that we will eventually see small and medium sized companies switch over to this platform.

I remember reading about Apple's first attempt on
an UNIX-like OS, called A/UX. Anyway, it's hard
to imagine an Apple computer running an IBM operating
system

I've gotten a few e-mails claiming that the Network server mentioned in the beginning of the article probably ran A/UX instead of AIX. Apple made A/UX back when the Macintosh platform was based off of the Motorola 68000-series processor. This OS was never ported over to the PowerPC platform, which the network server was based off of. IBM on the other hand, does have AIX for PowerPC, so it was a logical choice for Apple.

To make it simple, here is the text of platform support from the A/UX FAQ:

I found an A/UX FAQ, to answer that A/UX on PPC question http://www.faqs.org/faqs/aux-faq/part1/ :

G.02) What's the minimum system I need (CPU, disk and RAM) to run A/UX?
---------------------------------------------------------------------- --

A/UX 3.0 works on the MacII (with PMMU _or_ 68030 upgrade with FDHD ROM's
installed), IIx, IIcx, IIci, IIfx, SE/30, IIsi (with 68882 chip) and the
Quadra 700|900|950 computers. A/UX 3.0.1 (and later) adds support for the
Q800 and Centris Machines (the Centrises _must_ have the real 68040 w/FPU -
See Q&A #G.03). A/UX will run on the Quadra 610 and 650s (recall that A/UX
requires the _real_ 68040 chip!) with a little bit of work:

You should make a copy of the A/UX Install Boot floppy and then copy the
Enabler for the Q610|650 onto this copy. You then boot up from this floppy
and install A/UX as usual. Finally, you'll need to copy the Q610|650
Enabler onto the A/UX MacPartition (or whatever MacOS disk you will use
when starting up your Mac and booting A/UX); do this by first booting off
a boot floppy or boot CD and then copy the Enabler over. You do _not_ need
to make any changes to the A/UX System Folder (i.e. the System Folder used
under A/UX).

A/UX will NOT run on the PowerMacs, any AV machines, any PowerBooks (or
portables), the LCs, the Duos, the ClassicII, the Q605 or on the Quadra
630... It is recommended that you NOT run A/UX 3.1.1 on the II, IIx, IIcx or
SE/30 machines, since their MacOS-compatibility is unreliable under 3.1.1.

Recall that A/UX _is_ UNIX and thus contains some very hardware specific
drivers. It's for this reason (and not Apple not doing things correctly) that
A/UX won't work on newly released platforms. To support a new platform, at
least _some_ work (and possibly extensive work in some cases) must be done.

If you really want to cut it close, 8MB RAM and an ENTIRE 80MB hard disk will
just make it. You'll have little room for user files (unless you clear out
some space by removing /games and maybe /catman) and depending on your
workload, may suffer from low performance (due to swapping... you may even
encounter the infamous swap messages

A much better system would be 16MB of RAM and about 200MB of disk space.
This would give you much more room to grow as well as sufficient RAM to
increase your performance (assuming that you tune some kernel parameters).
All in all, more RAM is prefered: 20MB (or more) is ideal.

>>Anonymous doesn't appear to appreciate that when it comes to price, lower is better. The Xserve is more costly for what it offers, compared to the commodity x86 competition.<<

You obviously haven't done your homework!

>>Krevinek wants me to prove that OSX is not a UNIX system. Nice try, but the burden of proof is on the claimant. Let Apple and all their shills prove that it is. They cannot, and I know why. If you don't know, then you would be better off keeping mum, and perhaps learning something, instead of waging a bellicose fight against reailty.<<

And I guess you're about to tell me that Windows XP is a totally new operating system built on 'N'ew 'T'echnology right?! Your post was useless FUD with no factual statements other than MS-DOS ruled the PC desktop way too long, but I can imagine that you're an average zealot that would just say about anything stupid to get attention!

'nuff said!

hi:

I liked the careful separation of stuff that's known (like pricing and megahertz ratings) from stuff that's believed (like performance consequences) and thought the article very thorough overall.

However, there are two things that should be added: in defence of Sun and in support of Unix (whether Darwin, Linux, or Solaris.

1- Sun's V100 isn't really a blade server. There's one coming and you can see a prototype by pulling a CPU board out of a V880, but the V100 isn't it. In either single or
(unadvertised but extant) dual CPU models the V100 is an entry level toy designed for pseudo embedded deployments and so Sun can claim to be cheaper than Windows PCs. My expectation is that the blade server will have dual 1.4GHZ US3i chips and sell for marginally less than the Dell Poweredge but, of course, I don't know this for sure.

2- I think the author should have included the full cost of the base Operating System with the hardware. You cannot run
the box without an OS and people who want to compare working systems have to take OS costs into account. To make the comparisons valid the article should have used the cost of a box that works, otherwise you unfairly penalize companies like Apple that don't charge separately for it. Yes, that would have created more rows to distinguish Lintel from Wintel and shown Wintel to be a bad deal, but so what? it is a bad deal.

I was torn between putting the OS cost in and not putting it in. I decided in the end to leave it out to avoid criticisms of biasing the Apple solution by artificially inflating the software cost. I do state that entry level Windows options are between $700 (Server) and $3500 (Advanced Server). Believe it or not, I still got flack about the OS integration, even though I didn't put it in. If I had put it in, and assumed that they were all running as Windows as file servers, then none of the Intel platforms would have ever won. I guess either way is unfair, but at least this way we could study the hardware trade-off a little bit better. If I had the software configurations, I would have loved to compare the costs of those as well!

While I am impressed that this disscussion is rather calm, I dont understand why this is a PCvMAC. here a company(Apple) is introducing a new U1 server. All we really need to know is; Will Apples server have what people want at the right price. looks to me that the answer short of seeing any hands on reviews is "yes". You should look at this particlular Apple product more like any other compnay offering a server. There are several companies using different chipsets.

how am I going wrong looking at it in this fashion?

Speed, since you are so knowledgeable about how OS X is not a UNIX based system, please inform us why that is. So far you have not supported your claims anymore than Anonymous has his.

Also, regarding the costs. Has everyone ignored what Hank stated in his article? Add on the cost for the common OS and applications used in the server world to the PC systems and then see how all the systems compare.

A lot of you also seem to be neglecting what markets the Xserve is aimed at. You can't compare apples to oranges (no pun intended). Think about who the systems are aimed at and what purposes they will primarily be used for.

The article may make no mention of it, but Apple is offering several support plans for these servers. I believe they range from $950 (for 24/7 phone support and 4 hour max. on-site support) to $2799 (for the same basic service you get for $950 plus all kinds of priority support, tools, and training guides). I have a feeling these may have been left out because they were not included with the base price of the server. You can read more about it at http://www.apple.com/xserve/support.html.

I see that Anonymous and pt_bandit are unhappy about some things, but since they can't argue with the truth, they really have nothing to say. Sorry Anonymous, but your mudslinging isn't going to alter the fact that Apple products are a poor value. And if you're going to get angry at someone about that fact, get angry at Steve Jobs, not me.

Some have noticed that every Apple product announcement is accompanied by an Apple vs. PC flamefest. This is not by accident. Apple actually has people on the payroll whose job it is to organize their religious followers to shill in public fora like this one. Ironically, Apple could have spent that money on improving their product line. But what we have instead is a group of poeple who all shout "what about Windows" when they can't explain a weakness in their own product line.

Witness the second part of Anonymous' personal attack, one that tries to evade the fact that Apple is marketing OSX as UNIX when it is not. pt_bandit takes the sheeply tack of repeating the same issue that I had just disposed of. Ho hum. The thing is that I'm not selling any product. I have nothing to prove here!

OSX is not UNIX because OSX is not UNIX. If it were a bona fide UNIX system, there would be a way to prove that it was. But it's not, so one cannot disprove a non-event. For example, I couldn't prove that fairies don't exist. But that inability hardly means that fairies actually do exist! In the end, those who want us all to believe in fairies, will o' the wisps, and other mythical tales can shoulder the burden of convincing us (or failing to, as the case may be).

But what are the implications of all these marketeering shenanigans? Well first of all, Hank is trying to make Apple's product appear more valuable than it really is because of the free software that comes with it, claiming "standard Unix utilities" and other nonsense. The truth is that you can get Apache for just about any OS that you want! What is being marketed as some mystical trait of "Unix" is in fact just free software that anybody can obtain and use, on any platform. And since on the IA-32 platform there's TOTALLY FREE Linux, choc