Terra Soft Releases Rackable Dual G4/1GHz

pinqkandi writes "Terra Soft Solutions just released their new GVS 9000, a dual G4/1GHz 2U rack system. With Mac OS X and Yellow Dog Linux pre-installed, along with the Black Lab clustering suite, these boxes pack quite a lot of power. $3,839 + shipping." Another user adds, "As an added bonus, now you know where to go to find Linux users who look like Cindy Crawford." Yes, well.

Comments

How about

[Treeluvinhippy]

A beowulf cluste..... oh right sorry.

Re:How about

[ThatComputerGuy]

A beowulf cluster of linux chicks that look like Cindy Crawford? Wow!

Re:How about

[Treeluvinhippy]

Ok to whoever modded me reduntant. Set your threshold to oldest first and you will see that I had first post! I know the joke is old but in respect to the thread I was the first to mention clusters. Or anything for that matter.

A linux user who looks like Cindy Crawford

[psergiu]

...and let the DDoS^H^H^H^H/. effect begin ...

Re:A linux user who looks like Cindy Crawford

[rjamestaylor]

Was fast for me at the time noted on this post.

Very interesting development - I know my company wants to move all our development and hosting to G4s...I'll pass this story upline.

Servers and one crazy beowolf cluster

[Stigmata669]

I am excited and dissapointed to hear this announcement. First, the idea that other companies are becomming interested in apple products as servers is very cool! Sadly, this machine is still too expencive to be practical in any accademic setting, and cheap PCs are still going to rule that piece of the pie... until some 500mhz streamlined sub $1000 rackmounts show up. Remember, the PPC archetecture simply generates less heat and requires less electricity to run, making it ideal for clusters. Come on apple, widen your spectrum!

Re:Servers and one crazy beowolf cluster

[phyxeld]

If these can run OSX, then, for all intents and purposes, mac clones are back. Right?

I realize these boxes don't use any apple (c) ROMs, but if OS X doesn't need them, then cloning is back, yes?

Could we possibly see some non-apple PPC OSX-compatible laptops in the future? Competition is the american way, remember.

Re:Servers and one crazy beowolf cluster

[pretoris]

No, these aren't clones per se. They're Apple machines that have been re-cased in a custom case. Basicly they buy a dual G4 from Apple and put it in a new case and install YDL.

Re:Servers and one crazy beowolf cluster

[phyxeld]

No, these aren't clones per se. They're Apple machines that have been re-cased in a custom case. Basicly they buy a dual G4 from Apple and put it in a new case and install YDL.

I find that hard to believe. Doesn't sound very cost-effective. (!)
I doubt apple would hookup a bulk rate for them either.

Re:Servers and one crazy beowolf cluster

[Graymalkin]

It's how it is done. They get them for the same rate as other VARs, instead of giving you a free printer they jam the components into a rackmount case. They even come with the Radeons in them like the G4 tower. They probably even make a few extra bucks ltting Apple have the cases back for use in refurbishing.

Not looking like cindy crawford

[Anonymous Coward]

Nope, that girl is good looking, not some used-by-the-industry hag like cindy crawford.

rack case ?

[patpro]

now we are sure a bi-G4 can sit into a 2U rack case, does anybody can tell if this kind of compliant rack case is available somewhere ?
I mailed GVS but I got no reply for the moment.

Re:rack case ?

[Anonymous Coward]

now we are sure a bi-G4 can sit into a 2U rack case,

Oh, man, she's bi, too? Where's that video?

Re:rack case ?

[patpro]

oups, sorry, I used french language here, in computer context 'bi' means 'dual' ;-)

and well, I don't know her enough to tell if she's bi :-)

Re:rack case ?

[Anonymous Coward]

Check out Marathoncomputer.com they sell empty 4u racks for g3/g4 stuff

YDL chick

[Tom7]

Well, that chick is a bad enough actress to be real...
(Or just a bad actress!)

Does your computer "define who you are"??

Re:YDL chick

[baruz]

No, it doesn’t. But then, she doesn’t say that, does she? It’s a “reflection” of who she is—I’ve certainly watched the video enough times to be able to know :.) I think this is true of anyone who uses a Macintosh. Some people have to use Wintel machines, say, at some office, perhaps, and can say that their use does not reflect their personalities. As someone who drags a TiBook to every site I go to even though the Rational product I support runs on Windows NT/2000, I can definitely say that my machine reflects my personality.

But how about rackmount servers from Apple?

[hcdejong]

Rumors on "Thing 2" (a rackmountable server from Apple) have been floating around for some time now. And there's a petition underway.

And 2U is not enough (or too much...). With blade servers being the new wave, even a 1U unit may be seen as taking up too much space.

It's a start...

[frankie]

...but Apple really needs to get on the ball and put out their own 1U "RacMac" G4 line in the $1500 range. The G4 mobo already has on-board Gigabit ethernet. You don't need SuperDrives in a vector cluster.

The market is there, if only Lord Steve wants to do it.

Not new, not from TerraSoft.

[navywife]

OR, for those without the need for pre-installed YDL, you can order it directly from GVS with OS X, and save $400. It's good to see these units gaining exposure, but I think the editors of Apple(MacSlash)dot (what else to call the two sites) would better serve their readers by publicizing the source of this HW.

As far as these being "NEW!" The availability of these was noted way back here in a discussion of Mac clustering.

As long as Apple(MacSlash)dot are posting exactly the same news day-to-day, it seems appropriate that I post identical commments to both sites...

why so expensive? ... allowed?

[kossico]

The reason that these are so expensive is that Terra Soft must first purchase the G4s from Apple, and THEN do their modifications or whatever and then sell them to their customers. As we all know, there's no "clone" market allowed.

Or am I wrong?

Superdrive???

[Betaman]

What good is a super drive with Linux?? Seems odd they would force you to buy essentially a $450 useless component if you got the Yellow Dog linux version. Even if you got the OSX server version DVD-R burning does not seem fast enough for server backups, especially automated ones. That is the biggest flaw right there.

Re:Superdrive???

[johnpaul191]

it's odd they don't pull them and sell on the used market. might not be as salable now, but when only top-o-the line G4s had superdrive, a lot of resellers would pull then in favor of a CD-RW or DVD and then resell them to other users. if you custom build a G4 these days from Apple, the Superdrive is about $200 more than the next cheaper optical drive.... granted saving $200 on a $4000 rackmount device might not be too much, but it's the principle of not needing it. then again maybe with OS X DVD burning it is a valid way toa rchive data? i dunno. i'm still hoping Apple makes a rackmount device at somepoint. as a longtime Apple user (since my ][e) i think it would make the boasting about OS X's power more desirable (and affordable?) to the super users.

Re:Superdrive???

[discstickers]

Sure DVD-Video burning isn't real-time yet, but burning data DVDs from the Finder would be much, much faster. Apple quotes it as 2x. Not that fast, but faster than tape.

Why Apple?

[Buskaatt]

Hm, these aren't apple products, and while they run MacOS X they the story says they're being sold by a Linux company. Why is this at apple.slashdot.org and not the mother site?

$3,839 ???

[Perdo]

That ought to make a screaming fast system, but it doesn't. For god sakes, Apple's performance claims are pure marketing FUD. Get two dual AMD 1u systems with 6 times the power for the same price.

Re:$3,839 ???

[Lars T.]

Of course SPEC doesn't account for SIMD. Not even with the super-duper Intel compiler.

Re:$3,839 ???

[paradesign]

that benchmark inst Altivec enabled, its like running a mile with only one working leg. give me real results.

Re:$3,839 ???

[Anonymous Coward]

Well, of course you can use SIMD! It's entirely up to the hardware vendor to provide a good compiler that is able to use SIMD. If you do that in a good way, you will definitely benefit from it in SPEC.

The Intel compiler actually codes for the SSE instruction set, but in general it only gives you 10-15% extra performance, since most advanced algorithms aren't perfectly vectorizable.

Apple doesn't provide any compiler that automatically vectorizes & generates SIMD to its users, so IMHO this is a fair reflection of the performance a user can expect on normal code.

Re:$3,839 ???

[Lars T.]

Sure you can use SIMD! But hardly any compiler does (because it's hard for a compiler to see when they can use it, esp. when using C(++)), including AFAIK the Intel compiler. At least c't hasn't reported that anything had changed from version 4.5 in mid 2000, when it only used around 20 prefetch (not SIMD) instructions in all of the SPECint95 suite. Remember SSE also contains non-SIMD instructions.

And if you think that "this is a fair reflection of the performance a user can expect on normal code", because the norm for programs would be to be simply compiled with the "normal" compiler, you also have to use the results from the MS compiler, not from gcc or Intel.

One more thing, while c't said they used the Absoft Fortran compiler, they did not mention wether they also used the "VAST-F/Vector - preprocessor which automatically inserts AltiVec instructions in your code."

Benchmarks?

[nullard]

For god sakes, Apple's performance claims are pure marketing FUD. Get two dual AMD 1u systems with 6 times the power for the same price.

Please, we all know that SPEC is designed to compare x86 compatible systems. SPEC tests integer math and perfectly formed sets of instructions. The G4 is a flop monster and recovers more quickly from branch mispredictions than x86 processors.

Using SPEC to compare a G4 and an x86 is like taking a Porsche and a rocket drag racer to the race track. The rocket car will win on the 1/4 mile -- it was designed for it. The Porsche will win on any track with a curve in it. The G4 has much better real-world performance than x86.

Re:Benchmarks?

[Graymalkin]

What the fuck are you talking about? SPEC runs functions on both the integer and floating point units that are taken from a wide array of common funtions. There's a bunch of crap flooding the waves about SPEC but the whole point of SPEC is you can do anything you want to optimize performance EXCEPT alter the actual source code. SPEC doesn't use SSE or 3DNow! instructions just like it doesn't use AltiVec. The G4 is a good performer but this crap about SPEC benchmarks has become ludicrous. C't did a shitty benchmarking job but it isn't the fault of SPEC or the tests. If you want to blame x86 bias look to the compilers. It's tough to find a proper PPC C compiler let alone a proper Fortran PPC compiler.

Re:Benchmarks?

[Anonymous Coward]

SPECbench is definitely NOT optimized for the x86. It's a collection of scientific software programs that have been voted to be very good representatives for high performance computing.

Actually, before the days of the Pentium III and Athlon, the x86 processors absolutely sucked at SPECfp (the floating-point part of SPEC), and everybody wanted Digital Alpha CPUs.

The most important reason why x86 processors (and Linux) is so common in scientific computing today is that they improved significantly a couple of years ago, and now provide by far the best performance/price ratio.

The current SPECfp performance leader is IBM with its Power4 CPU. At more than 1100 SPECfp2000 it is almost 10 times faster than the G4. Did you know that this is actually also a PowerPC CPU?

Get a clue: _EVERY_ single manufacturer _EXCEPT_ Motorola provide good to very high performance on the SPEC benchmarks. Just as the G4, most of these processors have vector extensions (SSE/SSE2 on intel, 3DNow on AMD, VIS on Sparc, MVI on Alpha, etc.).

The only difference is that all processors except those from Motorola provide excellent performance even when you compile standard portable C or Fortran code.

Of course it's possible to handcode nonportable multimedia instructions, if a significant amount of your users would benefit from it. This is the reason Adobe has done it - 50% of their earnings come from PowerPC users. But how many companies do you think will do it for 5% of their users?

As a longtime Unix user I really like OS X, and my next laptop will almost certainly be a Mac (since raw speed isn't as important there), but this type of stupid Motorola-defending will actually hurt OS X in the long run.

What do you think the first impression of OS X will be for a scientist who has just paid a lot of money for a computer he/she expects to be much faster than his Athlon, and then compiles his code and finds out it runs considerably slower instead?

Re:Benchmarks?

[Anonymous Coward]

...but this crap about SPEC benchmarks has become ludicrous. ...If you want to blame x86 bias look to the compilers. It's tough to find a proper PPC C compiler let alone a proper Fortran PPC compiler.

It seems rather suspect when a benchmark relies on a language that very, very few applications programmers use. The only reason I have a Fortran 77 book is because the CS department was throwing it out a few years ago. Also, being proud that a benchmark never takes advantage of special features of a chip is akin to being proud you've never used the overdrive on your car. I'd rather know what a chip will really do for me when I put it to use with both code that takes full advantage of the chip and code that doesn't. Benchmarks in general are shady things, but at least with something like Quake fps tests there is an actual task I do which is being tested. SPEC doesn't seem to have any mundane analog for what I do as a user (I don't compile kernels particularly often, which is what this test seems closest to simulating). And because of its generality it doesn't reveal the true speed of a processor in a controlled environment. I'd prefer seeing MMX vs Altivec for a number of different algorithms if we're going for pure lab speed testing. As it is, SPEC really doesn't seem like a particularly remarkable test for most anyone.

The FUD calling the marketing FUD

[Anonymous Coward]

For god sakes, Apple's performance claims are pure marketing FUD.

That benchmark is pure FUD. It's many, many problems are pointed out on slashdot, macslash, and probably best, at appleturns.com. The tests:

• Use only a single G4
  
• Are designed to occasion a minimal number of failures (if any) in the pipeline, thus making CISC chips looks more effective than they really are in normal use.
  
• Do not use altivec.
  

I know you're a troll, so that's all the more time I have for you.

Re:The FUD calling the marketing FUD

[Anonymous Coward]

* Use only a single G4

Yes, that's right - they compared a single G4 with a single Pentium III, since there is no single-cpu 1 GHz G4 sustem. What's your point? Of course they could have compared a dual Pentium 3 against a dual G4, but that would have required an autoparallelizing compiler. There are a couple of them for x86, but none for OS X as far as I know, so the G4 would have fallen even further behind.

* Are designed to occasion a minimal number of failures (if any) in the pipeline, thus making CISC chips looks more effective than they really are in normal use.

Interesting claim. Could you substantiate it, apart from a rumor on a Mac site? The Spec tests include a quantum chemistry calculation, a molecular dynamics simulation, fluid mechanics, a small part of the gcc engine (compiler), etc. Apparently we scientists must have spent the last 30 years optimizing our research results for x86 without knowing it. We should probably recall John Pople's Nobel prize since his Gaussian software uses the same algorithms.

Do not use altivec.

No, that's true, but that's Apples fault since their compiler doesn't produce it: This is exactly the performance you'll see if you compile high-level programs on the PowerPC. Also, most scientific programs require double precision and/or 64 bit integers for accuracy reasons, which rules out altivec.

By the way - gcc doesn't produce SSE code either, but apparently the Pentium did well even without SSE.

I'm sure the G4 is an excellent CPU for Photoshop and video editing, but a "supercomputer" that can't deliver in double precision and sucks on compiled high-level code is just a joke :-D

briQ

[jmcmurry]

Not exactly what you were thinking of, but you can buy a briQ from Terrasoft. Teeny little PPC computer that can fit in a drive bay (!). The 500Mhz G4 is $1985, the 500Mhz G3 is $1485.

This was news to me.

no rack case available...

[patpro]

here is the reply from GVS about empty rack case for G4 systems :
- - -
Thank you for you interest.

Unfortunately because of warranty and prices that will involve you do the
conversion and other parts such as controller, restart and raiser cards.

For a limited time we are offering the conversion of you current G4 to GVS
9000 2U or 4U system for $499.00.
- - -

I don't know the cost of a trans-atlantic shipping for a G4 (+ insurance), but I guess it's chipper to sale my current G4 on the 2nd hand market here in France and to buy a brand new rackmounted dual-G4 from GVS