Slashdot Mirror
Virginia Tech Upgrade: PowerMac G5 to Xserve G5
An anonymous reader writes "Virginia Tech officially announced that they will be migrating their G5 Supercomputer from PowerMac G5s to Xserves.
According to the article, the Xserve G5s will reduce power consumption, heat production and decrease the system size by a factor of three. The pricing of the upgrade is still being determined, and according to Srinidhi Varadarajan, they are working on getting "very good homes" for the PowerMac G5s which will be replaced."
*looks under desk*.. I'm sure I could find room for, oooh... a couple of hundred..
Does anyone know what the university got in return for allowing Apple to film the installation and staff for the Xserve promotional videos? A reduced price upgrade may have been part of the initial agreement
they are working on getting "very good homes" for the PowerMac G5s which will be replaced.
Can EBay be slashdotted? I guess we'll find out now!
I am a viral sig. Please copy me and help me spread. Thank you.
Is it like adopt a G5 day down at VT? Is there a background check or can I just pick up my tower and beat it to death once I walk outside with it?
To adopt one of those adorable little...
Heh.
Screw it, just gimme a G5!
Shinma
would be a good home
I'll volunteer to provide a good home for one of those spare G5s. I know, it's a selfless act, but what can I say? I can't just stand idly by while distributed supercomputer nodes go homeless.
I'm really hardly suprised. But my real question is, did Virginia not realize that 1100 full-size G5's might cause problems? There is a reason such things as Xserves exist; they are for large-scale installations such as this one.
When it was first rumored that VT might replace its G5 boxes with Xserves, a friend of mine shared the idea that the pulled machines should be resold to the public, with some indication that they had been part of the cluster, perhaps a plaque or laser engraving noting that they had been included in the VT supercomputer. I bet those things would be bid up sky-high on eBay!
Blogging Weight Loss, Distance Education, and more at verlin.com
why not have a few more Xserves, I mean they already have the infrastructure for that much heat/power/room, so why don't they supersize the Big Mac?
Bob
Now for all those people who droned on and on about how foolish VTech were for not getting stripped down boxes, here's the reason.
Quantum materiae materietur marmota monax si marmota monax materiam possit materiari?
would be to reward some VT Computer Science majors.
get an A in any programming class, take home a G5.
It really amazes me that VT can afford to waste so much money on a super-computer. Really, could they have not waited the 4 months to get their brand new shiney G5 based super-computer and not wasted so much money on all those G5 towers?
I mean sure they can use them in computer labs and such but it still seems like a waste in the current education environment of tight budgets and program cutting.
All Im saying is that this money probably could have been put to better use else where than getting VT their brand new toy "right now" instead of waiting 4 whole months to get the hardware they really needed for the project.
"We Don't Need No Truthless Heros!" - Project 86
Allthough 1000+ 'refurbs' could distort the 2nd-hand market a bit and I'm not sure Apple would want to buy them back, why don't VT make some money out of this. I'm sure that a Mac with an official "In my previous life I was part of BigMac" plaque on it would sell for more than the price of a vanilla 2Ghz G5. With all the publicity over the last 5/6 months these things are collectors items. VT could come out of this with a profit! ;)
(Guess I should have phrased this in a 1... 2... 3... Profit format)
they are working on getting "very good homes" for the PowerMac G5s
"working"? more like "stealing" i assume... it would be interesting to know how many of these G5 will leave the house through the fire exit.
Of course, in the original implementation (which some of the VTech guys had actually ported to Gentoo to replace the emerge system!), I had to rely on ASCII graphics and animations to replicate the Quartz Expose graphics. But, running natively on Mac OS X and G5 hardware, it would be possible to strip out the text Expose animations, and write PowerPC 970 assembler to directly invoke the OpenGL version! That's not all though. Because the load placed on apt-get-expose would be much larger for a cluster the size of Vtech's, I had to code a parallel graphics engine, taking advantage of the AltiVec unit on each G5. This new implementation, coded in tight hand-tuned Python, actually replicates the entire DirectX 9 platform, but using AltiVec calls for enhanced speed. That way I can use the same code base for Mac OS X and Windows XP when win-apt-get is released.
Man you should see them down at VTech now. Those guys are loving being able to manage up to 1024 individual apt-get sessions at once, and with a simple tap of the F9 key, select a package upgrade from anywhere on the network. It's an example of how the new Apple, open source, and the ingenuity of the original apt-get developers can combine to produce something bigger than just a fancy window tiling animation. apt-get show desktop out.
I'd Like a G5..or a couple hundred.. Where are they going to sell them?
Imagine a beowulf cluster of th-- oh wait. DAMNIT!
--
The last digit of pi is four.
when i heard about this the first thing i thought was "they just it up and running and now they are doing an upgrade?" i'm not in the cluster world does this happen often? does anyone else think that it came too soon? Or is apple giving them another deep discount to keep an Mac based computer #3 on the supercomputer chart?
You can bet that the VT is HEAVILY supported by mac marketing devicion. Even the original prices look suspeciosly cheap, and the machine was just ready at the right time to enter top500.
And questions about missing ecc ect were evaded,ect...
I suspect that the machine until now was nothing more than a apple marketing stunt and has not done any real computational work besides network/topology testing an a top500 linpack run.
HI O WISE PRINCE. WHT TOOK U SO DAM LONG?
I would like to let V. Tech know that I am more than willing to accept some of those poor, forgotten G5's into my loving home.
There, they can spend the day happily puttering away in my new (soon-to-be) G5 cluster, working merrily at finding ways to improve the human condition, advancing understanding of the universal truths, and produce superior pr0n...
Whew! This water sure is cold!
Not a G5 expert here...
I wonder if there is a processing gain acievable by doing this or of the motivation is purely power dissipation and space. If so, at the end of the day it seems like the power bill delta over the usable life of the computer wouldn't make the expense of the upgrade worthwhile (especially considering VT has an on campus power plant of their own). Wouldn't it make more sense to wait around for the 'next best thing' instead of the same thing in a different package? If it ain't broke, why fix it?
But I guess they want a super-computer the football team can be proud of...
I want what their eBay user ID is....
1
668.5
Trust me, the university is not letting anything out of their hands that can't be obsoleted first. It's a state school so they have a pecking order. My first bet is a large majority ends up at the Empo' followed by professors (who are also looking to build a smaller farm), faculty, staff, other state schools, and if we are so fortunate (and this is really a long shot) you can scoop one auctioned[PURCH].
This is really amazing to me. They spend a chunk of coin
on the system, barely get it up and running, don't actually use it for anything, and then run out and upgrade it 4 months later. Sounds like a typical home computer geek!
But really, shouldn't somebody be better managing the money
there and saying 'hey go do something worthwhile with your new toy, then ask for more money'?
How is this Insightful? The Lead in stated the reasons as a less power consumption, less room needed, and less heat produced. Last I checked trying to save money on Electricity, Cooling and Floor space was simple good use of students tuition dollars.
E. Bay.
--- Ban humanity.
Or a supercomputer that the football team can spell. "G5" is shorter than "Pentium".
--- Ban humanity.
+4 Interesting?!? Did any of you moderators actually read more than 2 lines. +4 Funny maybe...
If I seem short sighted, it is because I stand on the shoulders of midgets
"No offense, but after running thermal dynamics equations, your, how do you call it, 'leet' Photoshops skills are somewhat beneath me. I'm looking for something that will stretch my thinking, not bore me to tears. I don't think I'm right for you. Perhaps a Blueberry iMac would be more your speed. Yes, a beige G3 with 64 megs of ram and os 8.1 should handle your AOL sessions just fine. I'll continue my search. Thanks for your time."
"I had to rely on ASCII graphics and animations to replicate the Quartz Expose graphics. But, running natively on Mac OS X and G5 hardware, it would be possible to strip out the text Expose animations, and write PowerPC 970 assembler to directly invoke the OpenGL version!"
C'mon.. If a post has more than 20 lines it automatically gets modded as interesting, without people taking the time to read it?
They could put them on ebay and advertise them as a piece of computer history. Seeing as they were part of a world class super computer
http://Lenny.com
I wonder if the lower power PowerPC 970FX used in the Xserve has superior performance to the ordinary 970 used in the PowerMac G5...
;-)
It would also be interesting to know if the 970FX has suitable energy saving modes and a low enough power consumption to be used in a G5 PowerBook
Were not talking about a public high school here.
.25 cents a head (or what ever it would come out to) to recoop the costs, which to say your school has a top what ever super computer in it, isnt a bad deal.
This is a college, where unless your on a full scholarship, someone is footing the bill for you to be there.
If nothing else, tuition might be raised
One of our local college's here, University at Buffalo, recently (within last year) got a 1000+ node Dell PowerEdge cluster for DNA sequencing or something like that for the medical lab. Don't know if it got on slashdot, but it was definately in international news.
I dont think anyone at UB (north or south campus) is complaining about the money it took them to build it.
It may seem like a waste to upgrade a system only four months old, but the reduced power consumption will save some dollars in the long term. By ditching the towers, they also save a boatload of space...
Where they can use some some of that extra money to purchase more nodes...
To put in all that extra space...
How many more nodes would it take to surpass number 2 on the list? Or possibly give number 1 a run for its money?
I think VT may be on to something here.
I like big butts and I cannot lie.
I wanna be the first kid on my street to have 1/1100 of a supercomputer!
BOGO would be good deal too...
They can toss them about the network and use them as a distributed compile/render/$distributed_task farm.
Absolutely an impressive school. 10 years ago, they were a joke. Now they built a national reputation via their football team, so people have heard of them, and projects like this put them on the map. When I looked at schools, they never entered the equation. If I was looking at engineering schools today, I'm sure that I would end up applying there.
This is a school with great self promotion and is going to go places. Unlike places like MIT, they don't sit on their Laurels, they are exploding.
I expect that in 20 years, they'll be considered one of the elite engineering schools. Kinda neat to have your college degree appreciate in value because the school gets better. I can't imagine that you don't get a decent engineering education at any engineering focused school, and this research project is a brilliant PR stunt.
Alex
Some people are like "Why would they upgrade from dual 2 Ghz G5 desktops, to dual 2 Ghz G5 rackmount servers??"
They won't be dual 2 Ghz G5 rackmount servers.. VTech is going to do the same thing they did when the G5's were released.. get first dibs on new inventory as soon as the new rackmount servers are released - the new 2.3Ghz rackmount servers.
Apple knows what's in it's product pipeline, and I guarantee you that they are in talks with Virginia Tech about offering their new xserves that are *yet* to be announced. You honestly think that Virgina Tech had no idea about the nee G5's prior to Steve Job's and his keynote? They are planning on upgrading their supercomputer, and they are going to be making it FASTER, and Cooler (bad pun.. I know)
Apple's marketing line is going to be: "Look, Look Not only is the 3rd fastest computer on earth powered by our G5's, but it also is run on our new XServes.. You need mission critical hardware? No problem. We build supercomputers!"
-Buddha wears grass shoes
This was *more* than worth it for Virginia Tech, academically, scientifically, and economically.
They spent $5M to instantly catapult themselves to the forefront of high performance computing, which was successful. Now they're replacing the entire cluster with ECC on the cheap, and will be doing real work with it in no time. This is a coup for VT, plain and simple. No one will be #3 again on the Top 500 list for anything close to $5M anytime in the foreseeable future. (The Top 10 will soon be populated with even more $100M+ clusters.) Virginia Tech's gamble will pay off many more times over for Virginia Tech, the people of Virginia, and the federal taxpayers who helped pay for it. As you claim to be a professor (which I doubt), it surprises me that you're too dense to realize that. Remind me to steer clear of your "classes".
They became the #3 most powerful supercomputer site in the world, #2 in the US, and #1 in education - and the first academic site to break 10Tflops - for a pittance, and in accordance with all rules set forth by the Top 500 organization - and now can attract much more grant money to do even more research and become an even bigger contributor, instead of taking years and millions more dollars to do it.
The Top 500 list has always been about hype! Wake up! Bravo to Virginia Tech. The only "pity" here is that you're so ignorant and shortsighted.
...how loud the Xserves are compared to the G5's? I can't imagine the decibels in a room full of them. One thing they don't mention in the article, and possibly another reason to upgrade to the Xserves, is the use of the Server Manager software. This software doesn't work on the PowerMac G5's because it doesn't have the sensors built in that the Xserves do. Not being that keen on cluster arrangements, I wonder if they have another product in place now that does the same thing with the PowerMacs?
Money not found! A)bort, R)etry, D)eclare Bankruptcy
joe, shmoe - from the middle of nowhere
where is it, where is the code, joe!? - for this 'apt-get-expose' or 'apt-get-osx' or the hyped up 'win-apt-get'..
Stop beeing a SCO, and tell us, be precise - mkay
I don't claim I know more than I know, and if you know you know more than I know, then by all means, let me know.
So what? He is a very funny and inventive troll. You know--the good kind. The one we would need a +1 moderation option for.
Hank! White!
right, so their system went online in November last year. so that's what - a 3 month old super computer being replaced by pretty much the same thing just to save on space and energyrequirements.
i mean surely they have enough space at the moment, otherwise their existing cluster wouldn't exist. so why not just scale it up using the Xserves rather than losing money on an investment that is so young???
It looks as if our hopes are dashed. Hi, That statement on Reuters should have read "found new homes" not "finding" Regards, Srinidhi Dr. Srinidhi Varadarajan Director, Terascale Computing Facility Virginia Tech -----Original Message----- To: Srinidhi Varadarajan Subject: Request for information regarding liquidation of the G5's Greetings Mr. Varadarajan, I am sure my email is only one of a slew of letters to grace your inbox following the press release (http://biz.yahoo.com/rc/040126/tech_virginiatech_ apple_1.html). I
sincerely hope this letter is no intrusion. I am writing in hopes of
obtaining enough information to keep up to speed with VT's liquidation
of System X. Hopefully, there will be an opportunity for myself and
others like me to finally purchase a G5 without breaking the bank!
Any leads you could give me would be greatly appreciated.
Thank you for your time,
Dylan Baxter
Computers are like air conditioners - they stop working properly if you
open Windows.
just gave this a run on my 2 home pcs and it's already queued up 2 cheap WiFi cards from eBay without hardly having to do any configuration. nice...
Whoops, forgot line breaks, heh heh. Anyway, that was a letter from Srinidhi Varadarajan, who is in charge of the project. -D
You'd think that someone could have predicted this back when the Big Mac was first being hyped. I just overestimated the longevity by a month. My bad.
Correct me if I am wrong, but I thought that Steve Job's himself had already said this in his Macworld Keynote. An excerpt from someone's notes:
Jobs talks about the G5 processor and Virginia Tech SuperComputer, who wanted "the first" 1,100 dual-2GHz Power Mac G5s. ("We pissed off a few people" getting them the first ones.") Cost them only $5.2 million and sending ripples through Supercomputer world. Jobs shows Virginia Tech Supercomputer video. It uses Infiniband networking; it took less than 3 weeks to assemble. Now in the top 3 Supercomputers. First academic machine to break the 10 teraflop barrier. The entire system runs on Mac OS X. Jobs says he expects to see a few more [Supercomputers] popping up hear and there
So VT is probably going to be THE FIRST to recieve G5 Xserve's.
Gorkman
Intel is their competition!
The article says the units are 1.75" in height. (Apple says 1U - same thing). That is the same height as the Dell's 1U units (1.67" according to Dell's site).
More servers in a smaller space has a lot of economic advantages, but have you ever tried to unplug anything in the middle of a rack stacked full of 1U servers? My hands don't fit. It takes about half an hour to finagle something in or out. The real reason our jobs are going over seas isn't because labor is cheaper. It's because they need the small hands of 3rd world children to do cable management.
Score:-1, not fawning over macs.
/.ers just ditch their old machine and buy a brand new one 6 months out because it consumes less power?
How many
Are we starting to see that this was all just a publicity stunt, or are we still blinded by the 'enemy-of-my-enemy' crap?
Apple's store has had dual G5 refurb's available for a while. Maybe they offered to swap. Anyone at VATech got some serial numbers?
Go to store.apple.com and click on special deals on the lower left.
Dual 2Ghz XServe Cluster Node: $2499
4 GB DDR 400 ECC SDRAM: $2115
36th month Mac OS X Server Maintenance: $249
36th month AppleCare Warranty: $760
Thats 5623 per node at Edu pricing before any volume discounts. This is also without the expensive networking hardware. If you use 6000 per node X 2000 nodes, that equals a 12 million dollar upgrade.
Add that to the 1100 XServes they have and you have a whopping 6200 G5 processors.
Priceless!
It sounds as if someone has a fun budget to play with. Is the PowerMac cluster too slow already or is this just a marketing stunt?
Well, as anyone who's actually used Macs for a while knows-they hold their value much better than PCs. Given that VT bought the G5s at the educational rate to start with, they can probably sell them right now at only a very slight loss. OK, I'm sure that listing 1,100 G5s on eBay at the same time would depress the selling price, but I'm sure VT could find willing buyers. Or Apple can just take them back and sell them as "refurbished" units at a slight discount.
I'll pay one shiny penny, too.
I mod down pyramid schemes in sigs.
For instance, it would be processor abuse to bring it online and subject it to a /.ing.
Infuriate left and right
XServe has it, G5 doesn't.
With the release of the clustering software, when are all us Mac Zealots ;) going to get together and make the biggest, fastest (probably not fastest as bandwidth would be at issue) cluster ever?
That would be rather interesting, especially if it became self-aware and started manufacturing windows viruses. Or hot robot chicks.
Give the guy a break. Even on slashdot, not everyone knows that in clusters, size matters. Also, the Xserves are designed to be maintained in racks. Desktop boxes just take too much time to get out, take apart, reassemle, and replace. I guess they're more heat-conscious, too.
Besides, now they'll have room for a few thousand more.
To be honest, the setup was ghetto from the start. Who in their right mind would deploy full towers in racks? It is such a waste of space reserved normally for companies like Rackspace.com who used to use the metal gurillaracks and cheap ATX towers.
I asked it months ago why they didn't just get the boards + supplies and have cheap metal 3u (or maybe 2u) boxes made with proper front to back cooling (could be stamped out for $200 or less by a metal shop in qty). Similar to what hotmail uses.
If it is true that the upgrade is free, I guess it makes sense. But we all know the new Cray system (www.cray.com) will be #1 on top500, bringing the US to #1 again.
Southeastern Virginia REPRESENT!
Xserve G5 Cluster Node 2GHz DP/80GB/2xGigE/10Client
Dual 2GHz PowerPC G5
512MB DDR400 ECC SDRAM - 2x256
80GB ADM (1x80GB Serial ATA)
Mac OS X Server, 10-seat License
6-8 weeks
$2,499.00
Subtotal $2,748,900.00
Please note that your subtotal does not include sales tax or rebates.
$2,748,900.00
Apple Part Number M9215LL/A
Find out how to get your order for $91,235.99 per month*.
"Win treats sysadmins better than users. Mac treats users better than sysadmins. Linux treats everyone like sysadmins."
For those who keep wondering aloud if they'll be able to improve their ranking, due to ECC, 2.3 GHz G5s, etc...
#3: VA Tech: 10.28 TFlops
#2: Los Alamos: 13.88 TFlops
#1: EarthL 35.86 TFlops
So, even if they spent 3x as much and filled up the now 2/3 empty room *and* scaled linearly (which they won't, won't, and definitely won't), they *still* wouldn't reach #1. #2, however, might be within reach, if they go to 2.3GHz and the ECC is a huge help.
Now that they more or less know where they'll wind up and there's no point in being secretive, I'd love to see them show what one box does on its own, then 2, then 5, then 10, then 50, then 100, then 200, 500, and finally 1100.
Dear Slashdot: next time you want to mess with the site, add a rich-text editor for comments.
This is how Institutions spend money. It is based on grant money, either use it or lose it. Even though an upgrade isn't necessary, the money is there, so they have to use it.
This is a test. This is a test of the emergency sig system. This has been only a test.
"You honestly think that Virgina Tech had no idea about the nee G5's prior to Steve Job's and his keynote?" Actually, I believe they said they didn't, which is rather suprising...
why not have a few more Xserves, I mean they already have the infrastructure for that much heat/power/room
ok, so they are #3 now on the list and have space to put x3 Macs in the same rack space. I guess that they will "upsize Mac and fries" the future with more powerful Xservers and will be #3 again.
Can someone point to articles or sites or reports of scientific or engineering calculations done on the machine? From my view it is just lots of light and no heat.
Unbelievable waste of a nice computer.
Doesn't this mean that VT can no longer claim to be #3 in the rankings? The computer used to acheive it will no longer exist.
A Big Mac built with Xserves will probably be just as fast or faster, but it won't officially be #3 anymore. They'll have to wait for the next round to get ranked again.
These boxes don't need keyboard, video, and mouse hooked up, so the wiring of these boxes is significantly less complicated than their windows brethren.
A power cable, one or two network cables. USB access on the front if physical access for KVM is ever needed. Makes for far cleaner wiring.
Is it just me, or has this reeked of bad engineering since the beginning? Most everything I've heard from them came from what sounded like Mac/G5 zealotry, rather than actual engineering. If you look at their slide show, they discounted a whole bunch of platforms because they would be 1.5 months later than Apple (to me the "Build and Benchmark" lines sounded like this hasn't been done. There was also no price/performance-type curve, etc.). Their justification for running OS-X also sounds like the sort of propaganda you hear from Mac bigots.
/proc/pci would overflow).
I'm not arguing the G5 is the wrong platform -- I'm just arguing the G5 wasn't chosen because it's the right platform, but because the guy in charge of the project is an Apple zealot. The G5 was quite possibly the right platform. MacOS-X is less clear; it's fundamentally an end-user OS. Linux has a huge amount of active development in high performance computing/clustering, and the kernel reflects that. The Linux kernel is also much more readable, and known by more people, so custom modifications (for low-latency communications with specific hardware, etc.) are a lot easier to do. This is not atypical -- when I was working on a little 18-way cluster of then-state-of-the-art dual PIIIs, I recall needing to tweak one or two things in the kernel (for reasons I won't go into, we had a large number of things on the PCI bus, and
It's not clear how well the G5 version of Linux is optimized (although it should be trivial to reoptimize it for the G5 if it's not -- probably a couple months kernel hacker time), but my impression is the VaTech guys didn't bother to investigate this, and instead said "We've got a full-fledged FreeBSD under MacOS!" and ran with it. Nothing I can find on their web site indicates any sort of competitive comparison.
The "minimal cost" of upgrading compared to the 5 million budget is bogus. That's over a thousand high-end Apple boxes. Those things cost considerably more than a grand each to produce (raw manufacturing costs), so that's easily over a fifth of their budget. They'll probably do some funky accounting to make it look minimal (give G5s to profs, who don't really want them, or place them in computer labs, or similar, and discount the cost it would have taken to do this in the first place). To be doing this change this late means they f-ed up big on some design parameter early on (didn't realize space/power/thermal limitations, didn't investigate other G5 platforms, or similar). I know university funding doesn't work like normal funding (money comes from sponsors, and some expensive things are "free", while some cheap things are unaffordable), but that doesn't make up for the huge cost.
I know people will be screaming that Apple or IBM probably footed much of the bill, but I think had VaTech talked to Intel or AMD, they could have gotten a similar offer (although the other chip houses probably couldn't have matched it).
Anyways, I won't rant and rave more. Everything I said is based on minimal information -- looking at the VaTech site, there just isn't that much. As a result, some of what I said might be wrong. Anyone from VaTech with actual facts care to provide some insight? By facts, I mean "The PIII-3.2GHz costs x dollars per box, while the G5 costs y dollars. We benchmarked them with z scientific computing test, and found the G5 has w advantage," rather than "we downloaded some benchmarks from a Macintosh web site." More importantly, provide insight on why you chose MacOS-X (got any benchmarks?)
I was going to sneak into the VT lab and steal a coupla those schweet G5s, but now that their getting rid of them, I won't need to. Hm... I wonder if I can return the tear gas.
How long till the 3 GHz G5 chips come out?
If he is going for the upgrade, he should go for speed as well as form factor.
Unloading 1100 G5 towers is easier than Unloading 1100 2GHz Xserve rack mounts...
But then again, there's always eBay!
I've criticized the whole idea before, especially the hype around it, when I can't see any big new ideas happening.
Statements like The price of the upgrade has not yet settled on, but Varadarajan said it would be minimal compared to the cost of building a new supercomputer from scratch are just ridiculous, since why would you need to build a new multi-million supercomputer from scratch if you have never even used your brand-new "old" one? That's like saying "tearing down our just-completed new villa and building the new house another way will save us so much more compared to building the same one again the same way!"
As Pike (of UNIX fame) and -- more recently -- Jobbs have noted, there's not a lot of innovation going on anymore.
Having said this the Xserves are very nice designs, so if anywhere, the cup for cool ideas goes to industry (Apple's engineers), not academia (Virginia Tech) -- also in the second round.
Here's my proposal: Why don't the guy at VTech not build a new user interface that goes beyond the useful, but aged desktop metaphor that the Mac introduced to the masses twenty years ago? Or how about some serious study of automatic load balancing on the "old" supercomputer? They might "save even more" money by taking some time to learn from mistakes in the first round before diving blindly into the next generation of their Uber-Mac project.
(Sorry for the rant, but it seems such a waste of resources when not too far away people don't even have their jobs anymore.)
just check out the apple web-site.
:P
it's possible to add up to four network
cards per Xserver:
2 x Gigabit ethernet (standard)
add-on one: Fiber channel
add-on two: Fiber channel
theoretically one Xserver can so
transmit and receive (at the same time!)
4 gigabits per second.
anyone know if the installed VT G4
have one or two (or more...)
network cards?
i'll opt for all four networkcards
but how many switches are they going to
need to get TRUE BANDWIDTH(tm)
from every G5 to the other?
2(n-1) -> 2(1'100-1) = 2'198 ?
2'198 * 4 (networkcards) = 8'792 i/o ports!!!
1'100 * 4'000 Mbits/sec = 4'400'000 Mbits/sec (550'000 MB per second)!!!
1'100 * 500 GB = 550'000 GB (550 terrabyte!!)
maybe CERN should get one for their new
"watching TV on LHC" experiment?
One thing about this whole Virginia Tech cluster is that the project director is Srinidhi Varadarajan, assistant professor of computer science. In most cases, assistant prof means untenured prof.
I'm not saying that assistant profs are not capable people. However, it is somewhat amazing that such a great responsibility has been undertaken/placed on an assistant prof (especially if he/she does not have tenure).
(Yes, telepathy is possible, just very expensive - read last year's news if you don't believe it.)
While it can be very entertaining to have the whole world revolve around you (including politicians and hollywood stars) the ultimate goal of this kind of game is to kill the player.
My interest is not purly academical: I'm trapped in a street game right now and you might see why I very much want to break out of this one.
If you're wondering what a street game is like: It's like being trapped in The Matrix and The Game, with a bit of The Truman Show and Being John Malkovich thrown in.
The game is wasting a lot of money around me to annoy me and surprise me, but it's more like a cat playing with a mouse it intends to kill sooner or later.
If you want to help me break ouf of this game you can make 25.000 Euro, which is my offer for anybody who explains to me what this matrix is, I'm trapped in.
I'm not sure I can be reached by internet or by phone and even if you reach somebody it might be somebody pretending to be me (so don't waste your time on that). The only way I recommend (without really knowing what's going on, of course) is to get close enough to me that the game has to pay your for your cooperation. Since my offer is 25.000 Euro you should be able to get this money from the game if they need you to cooperate.
Please make them pay this amount as I really would like to see them run out of money!
You can find me in the following places. (please have a look at my homepage and/or my journal for updates to this travel plan):
Addis Ababa: just getting thrown out by the police.
Berlin, Potsdamer Platz, 28-29 January 2004.
New York, Central Park, 01-03 February 2004.
Chicao, ?, 04-06 February 2004.
Los Angeles, 07-12 February 2004.
Sydney, 13-15 February 2004.
Tokyo, 16-18 Feburary 2004.
Bombay, 19-21 Feburary 2004.
THIS ACCOUNT IS NO LONGER IN USE, PLEASE DELETE.
...is my home
CAn'T CompreHend SARcaSm?
There is no evidence, aside from the fact that SCO has pissed a lot of people off, that it's a GNU/Linux user.
In fact, it's arguable since this person obviously has some knowledge about Windows APIs, it may be someone outside of the community trying to stir up trouble.
Whoever it is, however, what you're doing is wrong and should be stopped.
One other question... when a Windows user acts stupidly do people judge all windows users? If not, then why when one Linux user acts stupidly, if it is indeed one, which is doubtful, do people rush to judgement against the entire community??
My point is that they shouldn't.
GJC
Gregory Casamento
## Chief Maintainer for GNUstep
If it will be like any other upgrade on campus, a likely candidate for the G5 may be housed in The Math Emporium. Any other freshmen hokies on here going there, or perhaps reading this thread from there right now? :-)
oops ...
(n*(n-1))/2 -> (1'100*1'099)/2 = 604'450
sorry %^)
604'450 * 4 = 2'417'800 unique connections
When I first heard about this cluster I remember thinking that it didn't make much sense at first for them to use big PowerMac G5s when the Xserves seems to be a much more logical solution for such purpose(power consumtion, takes much less space, etc). So my thought is that they actually planned this already, they just wanted to make a big marketing hit with the PowerMac G5s to sell more of them, and then now they just replace them for something that makes more sense.
:)
Might be wrong, but that's just an idea.(maybe the fact that I just watched "pirates of silicon valley" makes me do a bit more thinking about Apple's marketing strategies now
Will they even be selling them?
I don't know anything about VT, but how many computer labs could benefit from new G5's?
How about other departments? Do they have a need/use for them? If nothing else, put them on faculty desktops.
Then there's always the possibility of reselling them to the current students.
I graduated from Virginia Tech in 1999. At that time, there were quite a few labs that used Macintosh systems, most notably the Math Emporium, a 217-computer lab for math classes. I also know that the English department labs used Mac systems when I was there, and I don't expect that has changed, either.
Personally, I don't expect any of the 1,100 G5 desktops will make their way back to Apple or be sold outside of the University. There are plenty of computer labs that could use the upgrades, and if there are any systems left over I'm sure a mini-cluster for testing out new software releases wouldn't be unreasonable use for a few.
Although I do like the idea of engraving the systems, like "This computer was node 243 of 1,100 in Big Mac v1." It would be something that you can show to alumni, because showing stuff like that to alumni results in additional contributions to the University.
And as others have pointed out, the graduate student labor is "free".
CT
Because the next list in June 2004, and the one after it, and the one after that, will include many new very, very expensive clusters (some $100M+) with performance far, far beyond 10Tflops.
l ist_for_CPU.gif
So, yes, someone can build a 1101-G5 cluster right now, and be faster than VT's cluster. But they won't be on any list, and they definitely won't be anywhere near #3 on the next Top 500 list. And neither will VT.
That's why the whole VT #3 thing is the coup that it is: the timing was *perfect* for them to take the #3 spot for a mere $5.2M. The PR and grants they'll get *because of* that are more than worth it. That will never happen again for anywhere near that small an amount of money anytime soon.
See some of the new clusters that will be in the Top 10 on the next list:
http://www.bayarea.net/~kins/AboutMe/GIFs/TOP500_
this, the entire supercomputer is going to be replaced, propably faster than the first one took time to built, now try upgrading the Earth Simulator! And then, try selling the parts to other people!
Mac are so easy to use!
I can see the incoming computer science students next fall, being required to purchase a G5 on entry. I can see them recieving "gently used" refurbished machines. I can see it...
Microsoft was right all along...
'mmmmmmmmm.... forbidden donut'
I don't know if this scales up properly, but 2.3 Ghz is 15% faster than the 2Ghz machines--x 1100, that increases their 10.3 teraflop system up to 15.45 teras, which should handily beat number 2's HP system at 13.9 teraflops. Even if it doesn't scale, it should get pretty darn close to no. 2.
Doh! That should be 11.85 T's. Still pretty fast anyways. That's what you get for being an arts major :-)
I wish I could afford to upgrade 1100 of my g5s. They are taking up so much space and power, I'd save a bundle dropping another 7.2 million on hardware. Really, how can I get that kind of funding? Seems like just yesterday they were having trouble getting peak performance from the cluster.
TallGreen CMS hosting
IBM would love Apple to try that sales pitch.
Apple: We built a really big supercomputer!
IBM: We build a few supercomputers a day. Of all sizes.
As has been pointed out countless times again, the cost was NOT a "mere" $5M. Their total hardware cost was ~$7M ($5.3M for the computers and memory, another $1.7M for the infiniband hardware), there was another $1M to upgrade an existing building.
"The total cost of the asset, including systems, memory, storage, primary and secondary communications fabrics, and cables is $5.2M." (Source: http://don.cc.vt.edu/tcfslides.pdf)
That $5.2M INCLUDED the Infiniband cards, switches, and cabling.
"The total cost of the asset, including systems, memory, storage, primary and secondary communications fabrics and cables is $5.2mil. Facilities upgrade was $2mil. 1mil for the upgrades, 1mil for the UPS and generators." (Source: Interview with Dr. Varadarajan)
There was then an additional $1M for "facilities upgrades", and $1M for power infrastructure.
They also had the benefit of free labour (millions of Mac zealots)
Huh? Millions of "zealots", eh?
Even if we GROSSLY overestimate labor, let's say a MILLION DOLLARS, the total cost is still $8M. So screw the free labor argument: even if they paid a MILLION DOLLARS to put it together (which is a huge, gross exaggeration), they're still much, much cheaper than anything close. Also, ANY academic institution has this same benefit.
and have not factored in the cost of power and cooling (at 2MW total power and cooling, this is a pretty significant expense, about $5,000 a day) or the support costs.
Sorry. Other clusters don't include power in their capital costs. And cooling *is* equipment included in the VT cluster. Ongoing support costs are NOT included in the costs of any of the other Top 500 clusters. The only thing different about the VT cluster was that the $5.2M figure didn't include some of the infrastructure costs other clusters have. But even the ASCI clusters are asset + infrastructure only, and do NOT include buildings, energy, or support costs. So, sorry again. And even at $7M + our imaginary labor, it's still ridiculously cheaper.
Even NCSA's new Tungsten cluster is $12M for the ASSET ALONE. That does not include building, support, infrastructure upgrades, or anything. Just the computer. And Dell installed it for free. So are they "PC zealots", since it was free labor?
They also had the benefit of free labour (millions of Mac zealots)
Actually, my understanding from the initial announcements was that most of the work (coding, LAN wiring) was done by undergraduate Hokies, in exchange for pizza and caffine while working, and of course bragging rights afterwards. The cost of a few hundred large cheese pizzas and a few hundred two-liters of Dew/Pepsi/Coke is low, compared to normal pay scale-- probably cost less than minimum wage per man-hour for VaTech.
//Information does not want to be free; it wants to breed.
Dreaded Math Emporium
Is my home good enough?
No TiVo and no caffeine make me something something...
I think you may have your facts wrong...
Apple announced the G5 base Xserve systems only a couple of weeks ago and they top out at 2GHz currently using a 90nm version of the PPC 970. They have not announced any systems based on G5s with higher clock speeds.
Can you point to documentation about systems using 2.3+ GHz G5s (PPC970/fx etc.) and when those will be available and when VTech / public will be getting them?
The main reason, so far stated, that they are swapping the systems is for space saving and power savings (electrical and cooling) thanks to the 90nm G5s being used.
Since the G5 towers were all part of a supercomputer, you can rest assured that all the wires and circuit pathways have been sufficiently burned-in. Nothing sucks more than getting a new computer and having to leave it crunching numbers for a week or more until it reaches peak performance....
"Orders of magnitude better"? Show me a benchmark. Please. You don't get "orders of magnitude" performance differences between the two kernels. Maybe a factor of 2x between the world's worst and the world's best kernel. Please. Altivec optimizations? Where in the kernel do you do vector or numerical procesing. That's what Altivec does. You Altivec-enable applications, not kernels (unless you believe MarketSpeak). You have one kernel optimized for the hardware, and another optimized for the task at hand. Which wins? Only benchmarks can tell. Not clueless morons yelling "Troll!!!!"
"Third fastest" by a fairly useless metric. In reality, supercomputers are mostly constrained by (a) individual processor speed and (b) communications. The G5 wins on (a), but almost certainly loses to everything on (b). The reason those supercomputers cost so much is because of very, very high-speed, low-latency custom interconnects. I can build a cheaper computer with more FLOPS out of ten thousand TI DSP chips, but it'll be almost useless for Real World problems.
But if we compare to supercomputers in the same calibre, you'll find x86 has a huge lead on interconnect. They have server-grade machines with very high throughput low latency chipsets. This doesn't matter for consumer applications, so you don't see this on Mac chipsets (also, Mac chipsets are designed by Apple. PC chipsets are designed by big-ass companies with hundreds of millions of dollars).
Again, if you wanna convince someone, post some numbers. Price vs. real-world scientific computing speed. The analysis that should have been done. Not some ravings about how much faster your beloved Mac is.
They should consider reselling them to the students who volunteered their time to help set up the Supercluster. I know they already got free pizza, but a discounted G5 would probably be extremely appreciated by most of them.
"Reality is merely an illusion, albeit a very persistent one " -Albert Einstein
Was in the days of the 2.0.x kernels. Nowadays, Linux blows away *BSD. Of course, Darwin is optimized for Mac hardware, so it's not clear who wins there. Only benchmarks will tell.
That's what the original post was asking. Why pick G5, 'cause the info wasn't on the web site. If it was picked for political/funding reasons (meeting a Top 500 deadline, rather than engineering the best possible cluster), that makes sense.
The original post said the G5 was probably the right thing, but was unconvinced by MacOS. There's a big question as to whether it makes sense to go with MacOS, and that can only be answered by benchmarks. The benchmarks should focus on interconnect speed, rather than just raw CPU performance. It's entirely unclear whether a tightly-optimized crossplatform OS (Linux) or a poorly-optimized, but platform-specific OS (Darwin) will win. Again, VaTech should benchmark the two, and see what's best. Not go with gut feeling or Apple's promises.
But there's also a difference between G5 and Apple architecture. G5 is a good chip. But the rest of the Apple architecture cannot compete with the interconnect technology on most other supercomputers, and has had trouble keeping up with chipsets in Wintel boxes. The problem isn't Apple's competence -- it's Apple's size. It's hard to come up with cash to create something to take over high-end high-speed crossbar interconnects, like you find on high-end x86 boxes. Performance, at 1100 CPUs, scales with interconnect performance rather than with number of CPUs. Even on 64/128 CPU SGI boxes, interconnect is a hard constraint. When it's talking through a PCI card, how fast you get to/from PCI makes a big difference, and that's where Apple may be losing.
Give me a break. As a native Vehjenyan I'll say it, if it ain't UVa, W&M or one of the private schools, it is rapidly approaching Radford U. C'mon, it's a straight up engineering drunken jockocracy. Besides, everyone knows a Hokie is a queer chicken. ;-)
Comparing it to Windows will be a moot point, since El Dorado is going to have a 40% larger code base than XP.
And really, all Apple was doing was rebadging IBM Silicon in the first place.
'1984' evil empire indeed!
---
Can I point to irrefutable proof that Apples next speed bump for their Xserve is going to 2.3 Ghz? No. I can't say that anymore than I could have pointed to proof the the first G5 Desktop was going to come in a dual 2 Ghz model.
What I can point to however is common sense. The common sense being that apple, with the advent of it's new 90nm fab process, will most likely i increase the speed of it's xserves. It's line of G5 desktops has not increased the speed of it's fastest model since it was released.. I wager that a new speed bump will occur in the next 3-6 months.
I mean, it comes down to a simple question..Do you think that in the next 3-4 months, apple will have started production of new xserves with faster G5 cpu's? Do you think that a new xserve in 3-4 months is reasonable (6 months after when the xserves were first announced) No only is that reasonable, it makes sense, and is totally consistent with apples past marketing/production strategy of updates/speed bumps 2 times a year.
Why would VTech just spend all that money to get on the top500 list as quick as possible, and then 4 months later say that they are going to switch the whole thing over just to save electricity? It makes much more sense for them to upgrade everything to rackmount servers that save space and electricity AND are Faster.
Do you really think apple, (assuming that they plan on updating their new xserve with a speed bump half a year later) would let one of their biggest purchasers, buy over a thousand new computers to place in the first mac supercomputer
Or.. does it make sense that (assuming that they plan on updating their new xserve with a speed bump half a year later) Vtech will look at this as an opportunity to speed up their supercomputer, save electricity, and space. It is in apples best interest to discuss new product releases with Vtech, the same way they were in talks with Vtech about building a supercomputer before they ever announced the desktop G5. This is a chance for them to take on the the spot of #2 (And if everything scaled nicely, they could cociveably beat the earth simulator by packing the room with xserves at 1/3 the size of the desktop G5's)
Apple would probably tell them, "Hey, if you wait 6 months we will have faster xserves, and if you replace all the powermacs with the xserves and maybe add a couple hundred more, you will be in 2nd place on the top 500.
Or, maybe they are INCREDIBLY impatient, and need to save space, and money on cooling RIGHT NOW! And as such, can't wait a few months for faster, cooler, smaller rackmounts.
Ok then... please don't state speculation as fact like you did in your prior post.
Much better IPC than x86, but the MHz difference is adequate that the x86 is still faster than the G5 on a lot of benchmarks. This is not so true of unoptimized applications (Office computing, etc.), but for vector ops, the deep pipeline of the P4 really helps. The P4 is much, much better for well-written scientific computing code. Then again a lot of scientific code isn't well-written, so the short pipeline of the G5 shines.
Note that the recently announced Xserve systems are already using PPC 970s built in the 90nm process (IBM's process by the way not Apple's).
Also note that the press release states that they will be switching to the "current" Xserve systems not waiting months for a potential speed bump.
Your speculations have merit but I think the current fact go against them at this time.
In theory the now have room for another 2,000 or so Xserve systems thanks to space savings... they may speed it up that way.
The Prof is a smart guy, and I'm sure he knew before the damn thing was even built that the X serves were on the way, probably becuase Apple TOLD HIM IT WAS, and cut him a sweetheart deal for the upgrades.
This will cost me some karma, but fuck it. I have no idea how come all you naysaying negative nabobs of negitivism on this post feel the continued NEED to badmouth everything possible about this project whenever possible. THEY DID IT. It deserves respect and maybe observation to see if you can learn anything from that success.
Here's the karma killer....get ready...it's coming....
I bet if was a goddamn Linux cluster we would see 1/5th the criticism. But since Apple hardware is involved, lets fuck it up so we can feel better! Yeah! :/
As a comparison to what they are replacing (not counting extra RAM)...
....$97,588.56 per month*.
(Higher Ed. pricing)
Power Mac G5 Dual 2GHz
Unit Price: 2,673.00
Subtotal: $2,940,300.00
So they are saving some money in the deal unless Apple discounted them by not shipping with video adapters, keyboards, etc. (likely)
The original post never claimed Linux was better optimized on the G5. It said Linux tends to be better optimized in general, but doesn't have G5-specific optimizations. This is true. Linux is being actively optimized by IBM, SGI, etc. specifically for scientific computing. Darwin is optimized like ass, but has G5-specific optimizations. Which of those is better? Only benchmarks will tell. Got some? Show 'em. Otherwise, shut up.
Also, what's up with Mac fanatics and Altivec? Altivec support is in the compiler, not the kernel. Kernels don't do vector math. Kernels do integer operations. SSE/SSE2/Altivec and others don't go in a kernel. They go in applications.
Apple-crazed Morons. Knock down any posts that don't fit their world view with "Troll", and don't even bother to respond with facts. Almost as bad as SCO. If you get facts, show 'em. If you've got religion that says Apple is better by virtue of the sticker on the case, go crawl back into your hole.
Maybe you ought to stop berating the parent. Everyone on this post is speculating. Christ man, relax about it.
I never could understand why some people on slashdot insist on nitpicking the stupidest things. ("IBM's process by the way not apples") How you interpreted that the parent was "stating fact" is beyond me - it's called an OPINION.
He was just giving his ideas. I thought that is what slashdot was about?
by the way..
PPC 970fx
and
PPC 970fx
Now get off your high horse
I am but a sheepish little girl.
English is nowhere near the topic of this discussion nor this thread.
I went to UB right after they got the Cray Origin2000. All I ever saw anyone do with it was play a cheesy 3-D video game on it.
Even in mediocre translation, that book changed my life.
It is cowardly, and a betrayal of whatever it means to be a Jew, to act as a white man
-James Baldwin
I find it strange that Apple didn't warn them that an Xserve version of the G5 system was being designed. This solution was certainly much better than the dual workstation case. Apple just never ceases to screw over their customers.
with all those G5s without a home, and me in need of a new computer desk, I could put the new G5 on top of my new desk, composed of 20 or so G5s stacked on top one another.
-Kids in the back seat causes accidents.- -Accidents in the back seat causes kids.-
They should engrave each one with a number and put on a little plaque that says "once part of the world's 3rd fastest supercomputer."
Come on, there's collectors' item potential written all over this! With a signed letter of authencity by Srinidhi Varadarajan.
I used to work at a university... games like this were played all the time, even at private schools. At public schools the politics and funding issues are much worse since you don't really even have control of your own budget, and everything has to be approved by the state government.
-- I speak only for myself
Since the Xserves take up only a 1/3 of the space. VT could pull out just a 1/3 of the PowerMacs install the Xserves and run 1100 Xserves and the 733 Power Macs that haven't been pulled out yet. Run it for a while in that configuration just to see what the score would be and then pull out the remaining Power Macs.
Now what we need is for all of /. to offer their CPU cycles to an open source cluster to create the #1 virtual super computer in the world for all time.