Xserve G5s to Replace PowerMacs in 'Big Mac'?

[MacRumors]

A Slashdot comment (http://apple.slashdot.org/comments.pl?sid=91897&cid=7906298) offers up an interesting rumor floating around Blacksburg, VA.

The rumor here on campus [Virginia Tech] is that Apple is going to let them trade the G5 towers for G5 XServes, 1:1. So if you are wondering where the first 1100 XServes are going, look no further than Blacksburg.

Now, Slashdot comment postings do not make for reliable sources, but we've since received further word of the same... from those who claim to be familiar with the plans. The migration plan was reportedly part of the original PowerMac deal. The big question, of course, is what happens to the original 1100 PowerMac G5s?

[joeyboy76]

apple is going to refurb then resell to the public! :D

[LimeLite]

I bet they could sell them on the cheap to another university wanting to make an inexpensive top 5 super computer. Sell the whole lot for 10% off and that's a pretty sweet deal for a super computer.

[Knox]

And if true it does give the possibility that they could add more Xserves than the current 1100 PowerMacs, since it's going to be taking up a lot less space.

[punter]

do the xserves have as much computing power as the dual g5s?

edit: the apple site suggests the xserve can do 9.0Gigaflops.

[CheekyGit]

EBay! EBay! EBay! EBay! :D

[ITR 81]

Originally posted by punter
do the xserves have as much computing power as the dual g5s?

edit: the apple site suggests the xserve can do 9.0Gigaflops.

They should have the same processing power.

The old G5's will be either refurbished or taken to another university wanting a similar setup or give to charity.

[esc]

The Xserve and the PM G5 should be roughly similar in compute power. The big wins would be ECC memory in the new Xserve, much easier to manage spacewise and lower power consumption (0.09um chips in the Xserve).

On the other hand installing and rewiring would be a fairly massive task. If I were running the VT array it would probably hinge on how important having ECC would be to the current machine.

[Centris 650]

I agree that they will probably refurb them. I wonder if Apple will drop the price of the refurbs since they'll have so many? Doubtful but with, hopefully, new powermacs out soon the dual 2gig will drop in price. Could be a good time to get a sweet deal on a G5.

[backspinner]

I think VT won't mind getting the rumored 2.3GHz Xserves instead of the listed 2.0 ones ;)

[MattG]

Wow that would kick ass. They'd be getting faster machines, plus they'd be making room for MORE machines. #2 Supercomputer anybody?

[Mr. Anderson]

As the star poster child for Apple's high end computing, it doesn't surprise me that Apple is trying to make a deal here. If they can manage to get 30% more CPUs in the racks (which I don't doubt is physically possible) they could get the #2 Super Computer.

I think this might be the underlying goal here. VT has proven that G5 Clusters are the way to go for inexpensive super computing. Apple and VT might have reason to keep VT at #1 for Apple clusters for a while - so this benefits everyone. And if they can manage to dramatically increase the power with minor alterations/installation, that will be fantastic.

Now how do we convince them to get on Team MacRumors: Folding ;)

D

[edesignuk]

Originally posted by Mr. Anderson
Now how do we convince them to get on Team MacRumors: Folding ;)
hmmm...a small (LARGE) bribe may be called for :eek: ;) :D

[the_mole1314]

I could see this happening, but VT spent tons of money on ventalation and other things. Don't you think that they would try to get the most out of it and not just go for a cooler processor?

[jholzner]

Will they be switching to the Xserve G5 Cluster node option?

[DaveGee]

"Own a piece of the 3rd largest super computer in the world!"

Heck... Some smart person could even sell em for a premium... Stick a cool placard on it and who knows what some might pay...

D

[Rico17i]

Set them up at 1 Infinite loop to run the iTMS.
:p

[MongoTheGeek]

Originally posted by backspinner
I think VT won't mind getting the rumored 2.3GHz Xserves instead of the listed 2.0 ones ;)

Could be the reason that the 2.3 haven't been announced.

Imagine if they were 2.3s *and* VT got enough to fill the same rack space.

That might put #1 in striking distance.

[edesignuk]

Originally posted by MongoTheGeek
That might put #1 in striking distance.
Nice thought, don't think so. The Earth Simulator is in a different league.

[yoman]

1)More CPUs for VT
2)More refurb dual 2GHz G5s for the rest of us
3)VT becomes #1 supercomputer in world
4) we all celebrate madly and laugh at every PC users face. :)

[BenRoethig]

Originally posted by jholzner
Will they be switching to the Xserve G5 Cluster node option?

Probably, I don't think VT needs the storage of a regular Xserve. The cluster is also a grand cheaper. I wonder if it will be any faster with Xgrid?

[Chaszmyr]

I dont think Apple would refurb them and sell them to consumers, I think they would try to sell them as a cluster to someone else... Afterall, the racks that they sit on and such at VT were custom made

[rog]

Wow, they should be able to run iPhoto without all the lags and slowness!

[yamabushi]

Originally posted by DaveGee
"Own a piece of the 3rd largest super computer in the world!"

Heck... Some smart person could even sell em for a premium... Stick a cool placard on it and who knows what some might pay...

D
I had the same thought. Apple might even be able to recover 100% of the cost or more. Selling them in an online auction might be the easiest way.

It seems to me that this would be building an entirely new supercomputing cluster. Some of the earlier work would make things easier but this would still be a massive undertaking.

[unc32]

I'd bet that VT would keep them and distribute them around campus. Saves money for everyone.

[supatekmedia88]

Does anybody think that it would be a good idea to be
able to rent time on the Big Mac to do film/3D/compositing
renders? It would be great to have the shots and materials ready,
and bring your team to the site for a week or month.

[kenaustus]

I wouldn't be surprised if VT doesn't run them for a while and actually get some work done - which was their original goal. During that time VT (and Apple) would also continue to develop the "BigMac user manual" for others who want to join the party.

I think the ideal time to make the trade-in would be when the 3 Ghz chips are available, providing a significant increase in performance. That gives time for developing a rather impressive package for Apple to sell to other schools.

The "old" G5s will be an interesting option for many who would like a part of history or could be offered to VT students at a good price - an approach that other universities could also look at in the future.

[Rincewind42]

Originally posted by the_mole1314
I could see this happening, but VT spent tons of money on ventalation and other things. Don't you think that they would try to get the most out of it and not just go for a cooler processor?

Well, the issue is that they may have spent a ton of money on the ventilation system, but there is also the on going monthly cost of running it. If they have a cooler system then the costs of that system go down, so they would save money in the long run - money they could put towards expanding the cluster.

[nagromme]

I can believe this MIGHT be true. And if not, logically VT will SOMEDAY upgrade anyway. (Personally I'd wait for Xserve G6s, but that's probably not a deal Apple could promise!)

VT apparently payed full edu price for the G5s. That always surprised me. Why didn't Apple cut them a deal, like they would for a big laptop program or other volume buy? Did VT just not negotiate at all? Did Apple not see how valuable a win VT would be, well worth offering a discount to?

This may be the answer: they'd pay full price and get G5s for the initial ranking, but Apple would swallow the depreciation later and allow them to be traded in for XServes. (An Xserve Cluster Node costs about the same as a top Power Mac.)

So maybe Apple DID cut a deal, and we're now seeing what they deal is. (Who knows when the transition might be made--it need not happen right away--especially considering the labor needed.)

Meanwhile, cost comparisons of Big Mac will remain valid: the discount Apple swallowed applies to the "extra phase" off building the tower-based cluster to meet the deadline. But another organization looking to build the same cluster would just go right for Xserves--this getting just what VT ended up with, and for the same price.

If this truly was planned all along, then the racks those PowerMacs are sitting on ought to be standard width, Xserve-ready once the shelves are taken out. Can anyone tell if that's the case?

(Of course there would be a lot of racks left over!)

[Tommy Wasabi]

Originally posted by kenaustus


I think the ideal time to make the trade-in would be when the 3 Ghz chips are available, providing a significant increase in performance. That gives time for developing a rather impressive package for Apple to sell to other schools.

The "old" G5s will be an interesting option for many who would like a part of history or could be offered to VT students at a good price - an approach that other universities could also look at in the future.

Now that is great thinking. Think about that for a second. Breaking up a super computer and selling pieces of it to students that could continue to use it for their work. The university could roll over their inventory every 1-2 years and students would benefit - IOW, the university continues to get state of the art at no cost! (the students in essence pay for the computer).

[Viv]

1100 refurbs hitting the market place would mean losing 1100 new G5 sales?

Why not move them on to the campus desktops and add them back in to a XServe cluster via the new XGrid?

Viv

[Dreadnought]

The racks is what I am thinking of also. They are all custum built. And in every shelf and metal of the racks there is cooling. So, they can't just get a couple of shelfs out! I wonder if this is true! If it is, they probably will have to rebuilt everything, including the arco/cooling. Don't think they are going to do this! Especially if they have a couple of projects running on Big Mac (sounds a bit like we are talking about Mcdonalds, doesn't it?!)

[tsunake]

Originally posted by Tommy Wasabi
Now that is great thinking. Think about that for a second. Breaking up a super computer and selling pieces of it to students that could continue to use it for their work. The university could roll over their inventory every 1-2 years and students would benefit - IOW, the university continues to get state of the art at no cost! (the students in essence pay for the computer).

Not to burst your bubble, but the students already directly pay for the supercomputer.

[ghutchis]

Originally posted by BenRoethig
I wonder if it will be any faster with Xgrid?

I don't think you understand what Xgrid represents. It's a very nice, user-friendly interface to clustering protocols like MPI and allows Rendezvous discovery of new nodes.

This is great--it makes setting up a cluster easier and allows smaller groups to use desktop CPU resources for clustering horsepower.

It does not necessarily improve existing cluster technology. If I run an existing Xserve, G5, or other cluster I already have the equivalent to Xgrid. (Several such programs exist, including Sun GridEngine, OpenPBS, IBM LoadLeveler, etc.).

I think Xgrid is cool, but I don't see existing clusters switching.

[Mr. Anderson]

um, didn't they say Apple would swap one for one the G5 PM for G5 XServes?

That means the G5 PMs go back to Apple.....

.....and then maybe to Pixar? :D

D

[BenRoethig]

Originally posted by Mr. Anderson
um, didn't they say Apple would swap one for one the G5 PM for G5 XServes?

That means the G5 PMs go back to Apple.....

.....and then maybe to Pixar? :D

D

I'd rather one of them replace the Compaq 8000z in my basement.

[desdomg]

I think it would be bad PR for both parties if the G5s were now replaced with xServes.

At a later date have a cluster somewhere else with Xserves, but not negate all the positive publicity that Apple have with big Mac by replacing it only a few months after introduction.

[Grimace]

I think it would be a great marketing move if Apple sold those used G5s to the students of VA tech. They all know the buzz - let them try it out and convert people who may already be on the fence!

[danbirchall]

Originally posted by Mr. Anderson
If they can manage to get 30% more CPUs in the racks (which I don't doubt is physically possible) they could get the #2 Super Computer.

30%? Hahaha. Aim higher. One of VT's homebrew racks holds 12 G5's (4 shelves, 3 per shelf) for a total of 24 CPU's. A standard 42U rack (and VT's may very well be higher than 42U) holds 42 Xserves, for a total of 84 CPU's. That's 3.5 times the processor density, or 250% more CPUs in the racks.

Load those racks with Xserve G5's and you're looking at numbers very close to those posted by the Earth Simulator.

[Why Not]

As they have said in many places, the racks are indeed completely custom. They are not standard width from what I can see, as they sport three G5s side-by-side, which would equate to more than 24 inches, whereas standard racks are 19 inches wide. This is not to mention all of the specialized cooling within the racks that were specifically built for for that chassis. The only option I could see happening is finding a way to mount them vertically with specialized shelving that acts as a rack, but my guess is that the cost for that would be higher than replacing the racks themselves.

I could see Apple funding the project at this point, now that they know exactly what the possibilities are. Before this was brought to fruition, it would have been a very politically incorrect thing to just give a University that kind of month, but now that it has become fashionable to sport Apples as a high-end super-cluster, I could very easily see Apple paying for some, if not all of either a retrofit or a completely new structure. It would be essentially putting the money where their mouth is.

[ZildjianKX]

These seems like too much of a pain in the ass to do... after all the man hours and preperations that went into getting the G5 cluster up... and that means they did all that work for less than 6 months of use, just to tear it down for an Xserve cluster? Not to mention they had to open up ever G5 tower to install the PCI cards, and that they would have to move and repackage 1,100 towers... which weighs about 66 tons! And all the setup they had to do for each tower software-wise... and that they would have to redue all this for each Xserve...

[Shrike_Priest]

I could see this having several benefits that make it possible for Apple to do this.

A. The Xserves would use a lot less power, which would lower the maintenance costs for Va Tech.
B. The Xserves would take up a lot less space.
C. With rackmounts, it would be a lot easier to add a few cabinets at a later date. whereas the G5's are quite cumbersome to switch around. and adding just 9 of them would take up about as much room as adding 42 Xserves.
D. They have cooling/cable management etc ready, and could use that off the bat.

From a management point-of-view, I think Xserves are a hell of a lot easier to handle.

And with smaller spaces to fill, they get more room over. so if they feel like it, they can just add a few more nodes later on (when the 3Ghz PPC 980 Xserves arrive for instance), and in that way earn another spot on the Top 500 list.

If apple can get the Big Mac to number two, only bested by Earth Simulator, then they have a HUGE achievement on their hands, and it'd be great PR.

[macshark]

Yeah, VT is using a proprietary interconnect that requires PCI-X cards to be installed in each machine. It would be a pain to open up all those PowerMac boxes, take the interconnect cards out and re-install them into the G5 XServes.

As others have pointed out, the amount of time and effort that was spent to build the custom racks, put everything together and make sure it works is probably worth more than the value of the machines at this point.

Moreover, even though they may be able to fit more than 3X the number of G5 XServes in the same amount of space, even with the 90nm processors, the resulting complex would need much more power and generate much more heat...

G5 XServes would be a much better fit for building a new compute farm instead of one that is only 6 months old.

[york2600]

Anyone know the exact speed they're getting right now?

[FlamDrag]

VT did some serious customization of those machines as evidenced by the photos that were posted. It wouldn't be as easy as simply unplugging them and letting someone else buy it. They would have to get all the labor to make them at least semi-normal boxes again.

[MacRAND]

Originally posted by Macrumors
The migration plan was reportedly part of the original PowerMac deal. The big question, of course, is what happens to the original 1100 PowerMac G5s? While it is interesting to speculate on what will happen to the "original 1100 PowerMac G5" towers, what seems to be catching everyone's imagination is how is VT going to accomplish the "migration"?
1. Unlike the original mass installation, the "migration" can easily be accomplished in segments.
2. A look at the racks suggests that they are simply bolted together and that the veritical portions can easily be converted to standard racks for Xserve boxes by replacing or modifying the horizontal connectors. http://don.cc.vt.edu/
3. Before shutting down any portion of the existing Big Mac cluster, it would be an easy matter to perfect the design for the Xserves by ADDING 42 to 48 of them in NEWLY designed racks before disassembly of any of the old G5s. VT has the luxury of having plenty of time to test the Xserves to be added to their system (air conditioning, wiring, etc.), including any upgraded configurations that they have discovered since implementing the original design, before putting them online.
4. Once the first group of new G5 Xserves are on line, a like number of old G5s could be removed from service, PCI boards removed, racks reconstructed, and another group of Xserves installed, and so on.
5. After the first 2 installations of Xserve G5s, VT would have a very good idea what it is going to take in materials, manpower, and time to complete each segment.
6. Likewise, Apple need only supply 50 to 100 G5 Xserves as needed over a much more extended period of time (weeks instead of days) than was possible for the original deadline oriented massive install.

So, will another university step up to buy hundreds of G5 Macs from the VT experiment when G5 Xserves will now be readily available with less heating problems and substantially less space required? It wouldn't be worth it, a Buyer would be going backwards technologically and taking on a serious & expensive installment headache. The G5 Xserves have obsolesed the G5 towers (too big & too hot) for massive cluster purposes.

If Apple took the old G5s back, refurbished them, added an engraving "0007 out of 1100" and sold them at a decent discount including AppleCare for 3 years free, who wouldn't want a piece of the first Apple SuperComputer as a memorial to Apple's 20th Anniversary?

[Steven1621]

considering that they are equipped and configured to be a part of a supercomputer, i can see apple breaking it up into several small nodes or, as others suggested, just another Big Mac elsewhere.

[suzerain]

This was basically confirmed by Varadarajan himself, months ago. While giving a talk about Big Mac, he was asked about ECC RAM, since the G5 isn't capable of using it.

His response, which I found odd at the time, was [paraphrasing]: "I believe that our software will route around errors, but we will be migrating to ECC RAM in the future anyway."

Well, there was, at the time, no way they could migrate to ECC RAM (and stay on Apple hardware), so it was obvious -- to me, at least -- that something was up, and the XServe G5 seemed like the most logical solution.

[arn]

Originally posted by suzerain
This was basically confirmed by Varadarajan himself, months ago. While giving a talk about Big Mac, he was asked about ECC RAM, since the G5 isn't capable of using it.

His response, which I found odd at the time, was [paraphrasing]: "I believe that our software will route around errors, but we will be migrating to ECC RAM in the future anyway."



I remember this talk but don't remember him saying they were migrating to EEC ram.

link?

arn

[wizard]

Not to burst a bubble here but who would be willing to pay extra for a PC that has had heavy usage and will likely be outdated by the time they are available.

Not to burst anybodies bubble here, but if VT is truely thinking about migrating to rack mount units, it does appear that they made a mistake in purchasing G5's in the first place. Can you imaigne anybody in business building a super computer and then telling his boss 6 months later that the physical plant has to be rebuilt because they will migrate to STANDARD hardware. The standard of course being 19" EIA racks. Only in higher education can one get away with this.

Dave


Originally posted by MacRAND


If Apple took the old G5s back, refurbished them, added an engraving "0007 out of 1100" and sold them at a decent discount including AppleCare for 3 years free, who wouldn't want a piece of the first Apple SuperComputer as a memorial to Apple's 20th Anniversary?

[Rod Rod]

I'd buy a used PM G5 from the VT cluster.

I think $2699 would be a fair price, with 3 years of AppleCare optional on top of that.

Each of those machines has 4GB of RAM!

:)

[breakaway1029]

I think that if this "migration" was part of the original deal, Apple was just using BigMac1 as a publicity stunt, where VT gets the #3 SC in the world, and Apple becomes a low cost solution for supercomputing, a win-win situation for everyone. Now apple can resell the BigMac as a whole, or in smaller clusters that are more reasonable for schools who want a supercomputer but only have $1-$2million. Of course Apple would have to help with setup and/or management, but that becomes a good deal for them. Apple would be absorbing ~$5million in giving VT xserves and helping manage the miniMac SC's, but they would become the favorite SC "manufacturer" overnight, by having ~3-4 computers in the top 10, but only costing ~$15-$20 million total

[MacRAND]

Originally posted by wizard
Not to burst a bubble here but who would be willing to pay extra for a PC that has had heavy usage and will likely be outdated by the time they are available.

Not to burst anybodies bubble here, but if VT is truely thinking about migrating to rack mount units, it does appear that they made a mistake in purchasing G5's in the first place. Can you imaigne anybody in business building a super computer and then telling his boss 6 months later that the physical plant has to be rebuilt because they will migrate to STANDARD hardware. The standard of course being 19" EIA racks. Only in higher education can one get away with this.

Dave Pay "extra", no.
Discount with warranty, yes!
And, I don't have a G5 yet so I would love it.

Yes to migration. Remember their deadline to pull this off? NO MISTAKE, it was worth millions in income from government grants and business for WHAT they have accomplished and WHEN they did it.
Oh, yes. Very smart move.:D

[suzerain]

Originally posted by arn
I remember this talk but don't remember him saying they were migrating to EEC ram.

link?

arn

Sure, Arn...make me look stuff up.

It was the second to last question in Varadarajan's TenCon Keynote about the G5 cluster....right above the "how many pizzas" question.

Here is a relevant quote:


How do you deal with Error Correction in Memory?

There's a lot of traffic on Ars Technica and other places. We do failure recovery, memory doesn't report. One of the things we've noticed is that failures aren't an issue yet. The reason they can be competent is the LINPACK test, which is showing 16 digits of accuracy. We are planning on moving to ECC systems in the future. They may have to run things twice for a bit.


So, if they wqere planning to move to ECC systems...they had to come from somewhere...it meant eithger Power Mac changes, or new XServes. Obviously, the latter was the more likely option, the way I see it.

Here are two transcriptions of the keynote: Adventures in Troubleshooting (http://www.tombridge.com/rta/2003/10/tencon_keynote__1.html) and MacSlash (http://macslash.org/OReillyCon/03/10/28/2357235.shtml).

[Steamboatwillie]

Originally posted by DaveGee
"Own a piece of the 3rd largest super computer in the world!"

Heck... Some smart person could even sell em for a premium... Stick a cool placard on it and who knows what some might pay...

D

or, laser etched on the side, numbered editions:

"System 72 of 1100 that was part of the original Virgina Tech Supercomputer. Ranked 3rd fastest cluster in the world in 2003 at 10.x terraflops"

Better than an iPod engraving!

[3.1416]

Originally posted by edesignuk
Nice thought, don't think so. The Earth Simulator is in a different league.
The Earth Simulator is 35 teraflops. Big Mac is at 10, and Big Mac II could easily fit 4 times as many CPUs in the same space. Could be interesting, although there's the question of whether VT wants to buy 4 times as many Xserves.

[desdomg]

I think this rumor is fantasy.

If true Apple would be sending out the following message to all those G5 buyers:

"G5's were the fastest but now they are obsolete and are being replaced by xserves"

And VTech would be spending more time building supercomputers than using them.

No, I think Apple will want to milk this success for all they can and V Tech will want to get some work done.

[ffakr]

This really is a stupid rumor.

- Apple and VT stated over and over that they did a retail purchase for these machines. Retail purchases do not include free upgrades from desktops to rack units
- As mentioned over and over, room/racks were custom fit for the towers, migrating to 1U racks would not be trivial.
- VT nodes have a lot of memory in them, if Apple swapped them out, that memory would be useless in the xServes and they'd take a bath trying to resell refurbs with that much memory in them. If they offered xServes with same memory config, they'd take an even bigger bath since ECC is so expensive.
- VT has already stated that one benefit of using G5 towers was that as they upgrade, they could re-use the towers around campus. They aren't giving them back to Apple.

This is a really silly rumor. With the memory in those, Apple'd probably take a hit of a couple thousand per node if they offered free upgrades. Apple execs have told me more than once 'Apple isn't in the business of loosing money anymore'. I've long said that Apple should dump cash into the Game development scene.. provide free coding help, discounted equip.. Apple's response has basically been that they don't spend money on things like this... nor would they take a $2Million hit trying to give VT free machines.
VT Isn't going to rip out their custom racks either.

This is what is going to happen. VT has finite space. They'll add new machines as feasible and when they need space, they'll de-comission G5 towers. Maybe they'll add a couple racks of G5 xServes further down the road and pull a rack of towers. Rinse and Repeat.
They aren't going to completely dis-assemble Big Mac, redesign the entire room around 19" racks, then rebuild it from scratch. A supercomputer isn't worth jack if it's in pieces.

[HumanJHawkins]

Originally posted by Macrumors
The big question, of course, is what happens to the original 1100 PowerMac G5s?

Is 1100 Dual 2.0 G5s really a significant figure? I mean, it would be a lot of Macs if they were in my living room, but I doubt they would be enough to effect worldwide (or even US) prices...

Anyone know how many dual 2.0s Apple sells per week?

[MacRAND]

Originally posted by desdomg
I think this rumor is fantasy.
If true Apple would be sending out the following message to all those G5 buyers:
"G5's were the fastest but now they are obsolete and are being replaced by Xserves"
And VTech would be spending more time building supercomputers than using them.
No, I think Apple will want to milk this success for all they can and V Tech will want to get some work done. Here's the reality:
1. G5 dual 2GHz PowerMacs were the fastest at the time; and the smallest.
2. Other G5's will come along and will be faster and cooler to run
3. VT should have bought Xserves in August 2003, but none were available
4. VT considered Dell, IBM and other off-the-shelf computers, but selected the new Apple G5 with dual IBM server chips.
5. Everyone knew that the Xserve would eventually go G5 and both Apple and VT anticipated this.
6. VT successfully clustered off-the-shelf G5 towers into a supercomputer, even though Xserves would have been a far better choice, but at the time there was no choice.
7. Now there is a choice, which will reduce the space needed to house their certified supercomputer, and requiring far less power to run and cool the heat.
8. Had a dual chip G5 Xserve been available last August, what would VT have purchased? At least they chose PPC G5s over IBM, Dell and others.

The migration makes a lot of sense, and it should work wonderfully.
Do substituted 2200 CPUs invalidate their claim as a # whatever supercomputer?

Change and progress are always part of the mix.
How about 500 2.0's and 600 2.6's?

[Didge]

http://www.smalldog.com/product/46388/atkibbles

[Cogz]

Back when Big Mac was built they were talking about how they did all this custom work to build the cooling systems etc. They ALSO mentioned somewhere that they were also in the process of planning and building an entirely NEW facility that would be built from the ground up to house the successor to Big Mac.

I don't see the dismantling the previous supercomputer just yet, but there IS an upgrade path. This was mentioned awhile back when they were talking about adding non 2ghz powermacs into the cluster, and it was noted that it couldn't be done. All the nodes have to be of the same speed supposedly. So rather than add nodes that are slower than are available, they will just wait till the new facility is ready (and from what I understand, its even BIGGER than the one in use now) and build an entirely new supercluster. The old machines? I dunno, that is up in the air.

Btw, I have no inside info, I am just going off of stuff I remember reading on this and other websites.

Cogz

[lewdvig]

It's part of the deal cause they are going to buy another 3300 or 4400 units probably. You would do this too if meant getting another 15-20 million deal.

50 tera flops...

[Zech Marquis]

I would like to see VT hang on to the G5s and let the engineering school or CS school use them. OR, gradually add in the G5 Xserves, or even--add some G5 Xserves and make the Big Mac even FASTER...

[AidenShaw]

The VA Tech racks are standard Liebert 46U 25" racks. If you go to the Liebert website, you can get the exact dimensions and everything, including a white paper on the VA Tech cluster (http://www.liebert.com/support/whitepapers/documents/virgtech.asp)

While it's true that the Xserve is based on a 19" rack - any carpenter with a 7" wide piece of metal and a drill could make a filler panel that would let them replace the Powermacs with 1U systems.

Cooling might be a big problem - the rack currently cools 12 dual systems, putting 46 systems in the same space might provoke the China Syndrome.

[Sir_Giggles]

I wouldn't rule out a major announcement within the next 12 months about G5 XServe-clusters overtaking the Earth Simulator as the top supercomputer.

The ECC RAM coupled with Xserve architecture, smaller size, lower power consumption, increased reliability, and cluster nodes bodes well for 4000+ units in the same space as occupied by Big Mac. I think Apple and it's VT partners are preparing to unleash the Number 1 Supercomputer using Apple computers.

The cost to build the #1 supercomputer would cost under $20 million, with lower operating costs over it's lifetime with the increased reliability of Xserve-grade Apple architecture, and you'll see a paradigm shift in supercomputing technology.

I don't imagine Apple and VT are aiming for #2 spot. We'll see really great news in the upcoming year. I bet my left nut.

:D :D :D :D :D :)

[MOM]

To all the posters happily adding extra Xserves and increasing the power of Big Mac: Remember there are latency issues. These things don't scale in a linear fashion-diminishing returns. I don't know what the curve looks like, but I bet they can't just keep adding units and get more out of it. They'll probably never catch the Earth simulator until the G5 is at least 3.5 X faster than now. Also, funding issues in Academia ususally preclude any fancy sell the old buy the new scheme. I'm sure they had a budget, spent it on the G5 and unless the trade-in was in the original deal, there won't be anything else. You don't get funding to build a supercomputer and then have significant amounts of cash left a few months down the road. Everything is on paper and planned out.

[ffakr]

It was my understanding that the money for Big Mac came in the form of a Grant (mostly at least). VT got a grant to build a supercomputer.

If I'm not mistaken about this, it means..

1) They didn't deplete their budget on this, they were given money to create this.
2) Money follows success. When you get a grant for a supercomputer and you do this well for that little... you won't have too much trouble finding more money down the road.

I think it's entirely possible that more funding will follow and that some G5 xServes will show up before Big Mac II is laid out. It wouldn't take that many new racks to threaten the current #2 machine, but more importantly adding more nodes would fend off all the upcomming clusters that will be nipping at Big Mac's heels.

Further more, If they did get a large amount of grant funding, then the Students did not pay for the cluster. (in fact, tuition typically doesn't pay for that much around the school.. even were I work where tuition is $37,000/year).
This doesn't mean that the University would be likely to sell the machines to students, however.
Virginia Tech is a state school and as such, equipment is very difficult to remove from the state inventory, especially when it isn't obviously out of date. My last employer was a state University and we'd NEVER be able to sell recent and useful machines to students. In fact, the easiest way to get old machines to students would be to de-inventory them (if they were old enough), toss them in the trash, and tell the students where the dumpster was.

[Corpus_Callosum]

Originally posted by unc32
I'd bet that VT would keep them and distribute them around campus. Saves money for everyone.

This was my thought. Should be easy to find 1100 uses for a G5 on campus.

[Gee4orce]

Maybe they are going to distribute the original 1100 G5s around the University for use as dektop machines ?. With the advent of XGrid these machines could be act as supercomputer nodes in their down time

[vitaboy]

By the way, why does everyone keep on referring to the VT cluster as "Big Mac?" The thing has an official name now, and it's called "System X."

I just find it curious no one EVER mentions it by its real name. :p

[MacRAND]

Originally posted by vitaboy
By the way, why does everyone keep on referring to the VT cluster as "Big Mac?"
The thing has an official name now, and it's called "System X."
I just find it curious no one EVER mentions it by its real name. :p Really? "System X"?
That's interesting!
Isn't that the name of a $600 Paintball Gun? and Apple's current OS?
Problem is, nobody I know has ever heard of System X when referring to the Virginia Tech (VT) 3rd ranked SuperComputer in the academic world (shhhh, all the big U.S. government supercomputers are SECRET; can't be verified because like "Area 51" in Nevada, they don't really exist, essentially don't count or can't be counted - take your pick). ;)
While the VT Supercomputer may have been given an "Official Name" of System X at Virginia Tech apparently in late November 2003, the VT cluster of 1100 G5 Macs had already been nicknamed BIG MAC by the public and the media since at least early September 2003 and repeatedly referred to by that name alone in articles and reviews thoughout October and November 2003, AND essentially capturing everyones imagination by that moniker ever since. "System X" cannot compete.
VT or whomever can call it "System Blue" or "Number 3" for all anyone else cares, because the name Big Mac is what has stuck as its identity in the public mind (outside of being the premiere hamburger of McDonald's fame, with "special sauce, pickles, onions, lettuce, cheese, on a sesame seed bun...").

'Big Mac' Supercomputer one of World's Fastest
by Matthew Broersma
ZDNet (UK)
October 23, 2003, 6:21 PM PT
A supercomputer built by Virginia Polytechnic Institute and State University from 1,100 dual-processor Macintosh G5 PCs looks likely to rank with the five fastest machines in the world, despite costing a relative pittance.
In preliminary performance tests carried out on 2,112 of the system's 2,200 processors, the so-called "Big Mac" cluster achieved 8.1 teraflops, or trillions of operations per second, according to figures published on Wednesday. The system is still being tuned, and final results won't be announced until next month, but the performance figure would place the Big Mac at No. 4 on the list of the world's fastest 500 supercomputers.
The figures are remarkable partly because Macintosh hardware has long been absent from the top 500 list, but also because of the Big Mac's cost. In a world where the top machines traditionally cost $100 million to $250 million, and take several years to build, the Mac-based system cost just over $5 million, and was put together in about a month. ...
The third-ranked system on the official list is, like the Big Mac, a cluster: it was built by Linux Networx for Lawrence Livermore National Laboratory from 2,304 2.4GHz Xeon chips, and runs at 7.6 teraflops. Another HP-built machine powered by Intel's Itanium 2 processors has not yet officially entered the list, but it would rank above the Big Mac, at 8.6 teraflops, according to Dongarra's figures.
http://zdnet.com.com/2100-1103_2-5095026.html

Big Mac Terascale Computer Principal to Address NCTC Technology
ROANOKE Va., Nov. 14 /PRNewswire/ -- Dr. Hassan Aref, dean of Virginia Tech's College of Engineering and a former chief scientist at the San Diego Supercomputer Center, will present the evolution of the Terascale Cluster computer built by Virginia Tech, nicknamed "Big Mac" by some.
Big Mac utilizes 1100 G5 Apple computers and technology devised at Virginia Tech to cluster the machines together. "Virginia Tech's idea was to develop a supercomputer of national prominence based upon a homegrown cluster," said Dr. Aref.
Big Mac now ranks third among the world's 500 fastest supercomputers...
http://ask.elibrary.com/login.asp?c=&host=ask%2Eelibrary%2Ecom&script=%2Fgetdoc%2Easp&query=refid%3Dovft%5Fkey%26querydocid%3D1G1%3A110175439%26dtype%3D0%7E0%26dinst%3D0%26pubname%3DPR%2 BNewswire%26title%3DBig%2BMac%2BTerascale%2BComputer%2BPrincipal%2Bto%2BAddress%2BNCTC%2BTechnology% 2B%26%2BToast%2E%26date%3D20031114%26author%3D&title=Big+Mac+Terascale+Computer+Principal+to+Address+NCTC+Technology+&pubname=PR+Newswire&author=&date=&ctrlInfo=&refid=ovft_key

SYSTEM X (not unique; no one cares)
A Sherlock search of the name "System X" turns up a line of PaintBall guns, a Lotto System-X, X-system music, and other things, including Apple's OS X, but only one small reference to the VT supercomputer on VT's own website. Most significantly, I cannot find any use of the moniker "System X" in relation to Virginia Tech's supercomputer project anywhere in the news media.

Besides vitaboy, who else noticed?
:confused:

VT reference http://macsupport.about.com/gi/dynamic/offsite.htm?site=http://www.eng.vt.edu/tcf/faq.html
paintball gun "System X Vengeance Pro 2.0" http://www.xtremez.com/paintball/product_list.asp?dept=91&last=91
music http://www.songsearch.com/catalog/x/x_system.html

Bottom line, "Big Mac" is the nom de plume that has stuck in the public consciousness for VT's BIG Mac G5 supercomputer, which incidentially runs on Apple's Operating System X (OS X).

[simply258]

Originally posted by ITR 81
The old G5's will be either refurbished or taken to another university wanting a similar setup or give to charity.
Yeah, OK. $5 mil to charity.

[Rod Rod]

Originally posted by MacRAND
Really? "System X"?
That's interesting!


MacRAND, you have the Kleenex is facial tissue, Coke is soft drink, Xerox is photocopy (verb and noun) argument nailed down securely. Thanks.

Bottom line, "Big Mac" is the nom de plume that has stuck in the public consciousness for VT's BIG Mac G5 supercomputer, which incidentially runs onApple's Operating System X (OS X).

also, you misused the term "nom de plume." you probably meant "nom de guerre." the VT cluster isn't an author going under an assumed name; it's an object which had been given a temporary alternative name. Thank you and have a great day. :)

[jcgerm]

Originally posted by Gee4orce
Maybe they are going to distribute the original 1100 G5s around the University for use as dektop machines ?. With the advent of XGrid these machines could be act as supercomputer nodes in their down time

I find this whole rumor pretty funny. It took quite a while to assemble the thing and get it running the first time. If they're all replaced, it'll be a lot of wasted time.

And even if the 1100 Power Macs were replaced, Tech wouldn't use the computers. We aren't a big pro Mac University really. The whole engineering department (computer science included) only uses PC's. They REQUIRE that students own PC's because of the software we use. Hell, the CS department is migrating the undergraduate curriculum to C# currently.

The only significant number of Macs that the university uses other than the supercomputer is in the math emporium (Few hundred iMacs for general student usage, i.e. anyone who goes to the university). I don't really know why they use Macs because all the math software on them runs on PCs. Granted, there are a few CS professors that love Macs, but they're not in the majority. At the moment, there are probably only a few hundred students on campus that use Macs due to their majors.

[thogs_cave]

Originally posted by MacRAND
4. VT considered Dell, IBM and other off-the-shelf computers, but selected the new Apple G5 with dual IBM server chips.


Just one small anal-retentive nit to pick - the G5 is not a server chip, it's a workstation/deskop chip. There are some decided differences between it and the POWER CPU's that it is based on.

[kingicon1]

I go to a state school myself.. use to work for the IT department as well..

I know here, they cant sell old equipment.. not quite sure why, but they just cant.. I think that might have something more to do with there dealings with IBM, but they do have alot of old stuff just laying around(including a older mainframe).

Also, Within the past few years they started the laptop program and every student gets a IBM Thinkpad. The only computer labs on campus were/are the mac labs for the art and design students. Starting last year they started giving iBooks to the art students which makes me wonder if the labs will even stick around.

I realize that VT is a larger school, but if they do replace the G5's with xserver's, it might be alittle difficult getting rid of the old towers.. unless they were going back to apple.

Also, I dont quite remember where the RAM came from, but it was installed by workers when they were setting up the cluster. If the RAM came from Apple, you would think it was pre-installed. If the towers did go back to Apple, I would think that VT would just keep the RAM around.

[spankalee]

Originally posted by Why Not
As they have said in many places, the racks are indeed completely custom. They are not standard width from what I can see, as they sport three G5s side-by-side, which would equate to more than 24 inches, whereas standard racks are 19 inches wide.

Are you sure they're not 19" racks? From all the photos I've seen it looks like 3 G5s only fit by sitting behind the rails. Three G5s would be 24.3 inches wide, and the racks sure look like 19" based on how far the machines go past the rack rails.

In fact I've always wondered why they used standard racks if the machines were too wide; it seemed like a waste of money. But if they're going to replace the PowerMacs with XServes it makes sense.

[Yawray]

from what i read ... [i guess you wont beleive me without the URL but here it goes]

1. VT saved all the boxes with keyboards and mice for the g5's used in the cluster, for the eventual sales of said g5's to the student population at a discount

2. VT is set to update the cluster with totally new hardware/architecture in 2006.

3. VT could possibly phase in xServe dualies like 100 a week or so if they wanted too ... and save space and start getting ready for phase 2 of the supercomputer ...

4. and keep the speeds essentially the same, or for a little spike there phase out the g5 towers over time as students buy them. or with Xgrid possibly keep them connected as remote nodes across the campus.

5. by that time the rumor mill will be rife with new talk of solidstate hard drives and 3ghz system busses and dual-core 6ghz processors with 8mb of level1 cache, quad-interger units and 8altivec registers per core ... and if apple doesnt come out with that they are doomed as usual. ;)

[macshark]

Originally posted by thogs_cave
Just one small anal-retentive nit to pick - the G5 is not a server chip, it's a workstation/deskop chip. There are some decided differences between it and the POWER CPU's that it is based on.

G5 is a server chip. PPC970 was originally exclusively targeted at IBM's low-end servers and very high end embedded applications until about a few years back when Apple made a deal IBM to secure the supply of this chip for Macs add AltiVec.

Yes, the G5 does not have multiple processor cores like the Power4, and the L2 cache is not nearly as big. Power4 was a "cost is no object" product targeted at IBM's very high end servers. PPC970 can be regarded as a integrated, cost reduced re-incarnation of Power4 targeted at high volume applications.

G5 has a very high bandwidth, scalable bus and very good support for multiprocessing. I don't know how one can classify this processor as "not a server chip".

[macshark]

Originally posted by HumanJHawkins
Is 1100 Dual 2.0 G5s really a significant figure? I mean, it would be a lot of Macs if they were in my living room, but I doubt they would be enough to effect worldwide (or even US) prices...

Anyone know how many dual 2.0s Apple sells per week?

The rumor is that Apple sold more than 150,000 dual 2G G5 PowerMacs so far. Maybe tomorrow, at the quarter-end conference call, someone from Apple can give us some actual numbers. Historically, they have given total PowerMac sales per quarter but they have refused to break down the sales data in terms of frequency or ASP.

[crkfc]

i doubt they would rip apart their current cluster just to replace it with xserves... i do believe that they will use xserves in their next super computer, which they have already announced is in the planning stages. they are building an even larger building for it already, i hear. this rumor really doesnt sound true at all... but if it is, that's cool.

[AidenShaw]

Originally posted by spankalee
Are you sure they're not 19" racks? From all the photos I've seen it looks like 3 G5s only fit by sitting behind the rails. Three G5s would be 24.3 inches wide, and the racks sure look like 19" based on how far the machines go past the rack rails.

In fact I've always wondered why they used standard racks if the machines were too wide; it seemed like a waste of money. But if they're going to replace the PowerMacs with XServes it makes sense.

The racks are standard 25" wide racks, with outside dimensions of 29" by 38.2". Look at the following picture, and notice that the racks are quite a bit wider than the 24" floor tiles:

http://don.cc.vt.edu/rackpilottest/slides/IMG_0819.JPG

If you look at other photos at http://don.cc.vt.edu/, you'll notice that the left edge of the first of these racks is lined up with a tile joint.

(BTW, 19" racks are usually 24" wide on the outside, and line up nicely with the floor tiles.)

A filler panel could be used to narrow the 24" rack to hold 19" equipment.

[thogs_cave]

Originally posted by macshark
G5 has a very high bandwidth, scalable bus and very good support for multiprocessing. I don't know how one can classify this processor as "not a server chip".

Look, don't get me wrong - it's a fine chip, and I love my Dual 2.0. (Well, in a platonic way.) But, just like Sun's UltraSPARC-IIIi .vs. the UltraSPARC-III, there is a target market intended. Features like sleep and bus slewing are good for desktops, but not necessary (or desired) in a hard-working server. I'm probably getting old and crusty, but to me a RISC server needs whopping amounts of cache fer starters...

Of course, I wouldn't turn down (and will probably soon be ordering) a G5 XServe. But to me it's a different beast then a large IBM or Sun system.

[ffakr]

Originally posted by thogs_cave
Look, don't get me wrong - it's a fine chip, and I love my Dual 2.0. (Well, in a platonic way.) But, just like Sun's UltraSPARC-IIIi .vs. the UltraSPARC-III, there is a target market intended. Features like sleep and bus slewing are good for desktops, but not necessary (or desired) in a hard-working server. I'm probably getting old and crusty, but to me a RISC server needs whopping amounts of cache fer starters...


MacShark is closer to the truth than you think. IBM noted in their official press releases that the 970 was intended to be a low end server and workstation processor.
IBM wanted a Power[PC] that they could sell in the low-mid range server market. That market is eating away at the high margin servers in a big way.
There are far more people looking to buy a $5000 server than looking to buy a server with $5000 cpus in it.

[TomSmithMacEd]

Why would they want to trade in for xserves? What is the purpose?

[fabsgwu]

c'mon, obviously they should wait at least a little while untill the 90nm chips and higher clock speeds etc. The computer cluster can't taken down and set up over night.

[MacRAND]

Originally posted by TomSmithMacEd
Why would they want to trade in for xserves? What is the purpose? Isn't it obvious? Runs cooler, less power demand and saves space.
42 Xserve G5s @ 90nm to a 19" rack vs.
12 G5 towers @ 130nm to a 25" rack

and, although the Xserve chips are still clocked at 2.0, their newer design with help to move the whole cluster to ECC systems efficiency.

And instead of having to replace the old G5s "overnight", VT can take their time with incremental replacement. Plus, their G5 installation labor (except for Pizza & Cokes) has been "voluntary" and essentially free up until now.
Plus, the Xserves are reportedly already part of the original contract with Apple and come at no additional cost to VT.

ANSWER: Why not trade?
:cool:

[MacRAND]

Originally posted by thogs_cave
Just one small anal-retentive nit to pick - the G5 is not a server chip, it's a workstation/deskop chip. There are some decided differences between it and the POWER CPU's that it is based on. Exactly, but according to IBM's own descriptions, it is a "server chip" designed by IBM to be suitable for desktop computers like the G5 and for medium to low-end servers. Don't forget that it is 64-bit, which Apple has yet to take full advantage of.

In fact, the PPC 970 is having an unexpected economic impact on IBM's higher-end servers and much more expensive server chips, partly because of the amazing success of the Virginia Tech "Big Mac" cluster built from off-the-shelf parts (G5 computers) at a fraction of the cost of other less competitive supercomputers which are not rated as Number 3 in the world.

IBM's new baby "Blue Gene/L" computer reportedly uses Power4 processors that run at a clock speed between 1.3GHz to 1.7GHz, much lower than PPC 970 chips, but with superior marks for latiency and other hallmark "non-failure" factors important to servers as opposed to just PCs.
http://zdnet.com.com/2100-1103-999873.html?tag=nl
http://archive.infoworld.com/articles/hn/xml/01/11/09/011109hnbluegene.xml
http://www.research.ibm.com/resources/news/20031114_bluegene.shtml

VT computer engineers say they selected the G5 not only because of price, but because the IBM PPC 970 CPU chip had more acceptable "server" design qualities (than the PC chips that they had been looking at) and they believed G5 chips would run effectively and efficiently in a cluster. Apparently, they were right.
:D

[macshark]

Originally posted by thogs_cave
Look, don't get me wrong - it's a fine chip, and I love my Dual 2.0. (Well, in a platonic way.) But, just like Sun's UltraSPARC-IIIi .vs. the UltraSPARC-III, there is a target market intended. Features like sleep and bus slewing are good for desktops, but not necessary (or desired) in a hard-working server. I'm probably getting old and crusty, but to me a RISC server needs whopping amounts of cache fer starters...

Of course, I wouldn't turn down (and will probably soon be ordering) a G5 XServe. But to me it's a different beast then a large IBM or Sun system.

I guess one of the more interesting trends is that the concept of a "server" has changed quite a bit in the last 5-8 years. Until about 1997-1998, a "server" was typically a $20K+ computer that came in a 1/4 rack or larger enclosure. These "servers" had significantly different I/O systems than the average desktop workstation (wider, faster SCSI interfaces for disks, support for larger memory configurations, ECC, etc.), so even when they used processors similar to the desktop workstations (e.g. Sun Servers), the company that built the system could justify the increased price.

The computing model was based on SMP. Except for a very few high end machines from Cray & SGI that used scalable interconnects, the scale of the system was determined by how many processors you could put on a bus, bandwidth and electrical issues being the major constraints.

Also, most computer companies that built servers made sure they segmented the market and did not let their lower priced (lower margin) desktop workstation products eat into the sales of their higher priced (much higher margin) server products.

What happened starting from 1998 caused major changes to this situation:

1. Linux became a credible server operating system and managed to push basic x86 architectures into low-end and mid-range UNIX server applications at the expense of Sun, HP, etc.

2. Intel processors (PIII and P4) made tremendous performance gains and exceeded the previously top of the line solutions from Sun/HP/Compaq/IBM for small server performance.

3. During the Internet/Web craze, the need to build very large server farms very fast at a low cost lead to a number of companies that manufactured cheap 1U/2U servers that can be very efficiently stacked in racks and deliver exceptional price/performance for applications that can be mapped to a distributed compute environment.

As a result of these changes, there is a big shift in the industry to replace older, large, expensive SMP architectures with racks of inexpensive 1U/2U servers for many applications.

This forces even traditional server vendors like IBM to deal with the problem of delivering low cost, efficient, low priced servers to deal with the competition. Looks like Sun is pretty much going to go ahead with Opteron and IBM has developed the PPC970 for this application.

[backdraft]

They'll probably sell some to Pixar and use the rest internally. :rolleyes:

-backdraft

[jdang]

just by addings more Xserves doesnt mean they are going to take over number 2 and definitely not number 1 spots on the top 500 list. the more nodes you add, the smaller the performance boost. its hardly linear at all. more xserves would indeed help, but it think it would be foolish for them to get xserves right now. they should wait till the technology peaks right before the next top500 list comes out. this will give Vtech and apple the publicity that they need. Vtech has been dying to get into the top 30 research universities in the nation, and apple has been itching to get into the new enterprise market. it would be nice to see both of these happen.

[MacRAND]

Originally posted by jdang
just by adding more Xserves doesnt mean they are going to take over number 2 and definitely not number 1 spots on the top 500 list. The apparent intent is not to ADD the G5 Xserves to the 1100 existing G5 Macs, or to challenge #2, but simply to incrementally replace the original G5 towers with G5 Xserves.
However, before taking 48 off-line, 48 Xserves could be installed in new racks and temporarily added to the system (1148) to make sure they run well before removing 48 old G5 towers (back to 1100).
:cool:

[thogs_cave]

Originally posted by macshark
As a result of these changes, there is a big shift in the industry to replace older, large, expensive SMP architectures with racks of inexpensive 1U/2U servers for many applications.

This forces even traditional server vendors like IBM to deal with the problem of delivering low cost, efficient, low priced servers to deal with the competition. Looks like Sun is pretty much going to go ahead with Opteron and IBM has developed the PPC970 for this application.

I think it's a pendulum shift. In other words, it'll swing back again, much like the "empower the user" psychosis in the mid 90's. PC's on every desktop, and decentralize *all * the computing resources. That is, until folks found out the support costs were a nightmare...

The 1U systems are neat, but they have their place and are not a panacea for all of computing - they are actually driving *up* the load on datacenter cooling and power consumption, and there are just some things that a whackin' big SMP does better.

As for Linux solutions - I've been noodling with it since Linus released the first kernel - I actually pooled all the money my wife and I had to get a 386SX/25 motherboard to build a system. Since then it's evolved into a pig. I'm working on implementing a SuSE-based IBM Opteron setup right now, and it's taking forever. Even the IBM guy I'm dealing with admits it's a step backwards.

In the end, you have to pick what works best for your application. For much of the stuff I work on, big iron is the *only* cost- and time-effective solution. The pendulum will continue to swing... :-)

[3G4N]

think secret has picked up sources on this too...

http://thinksecret.com/news/virginiatech2.html


\

[ffakr]

Originally posted by 3G4N
think secret has picked up sources on this too...

http://thinksecret.com/news/virginiatech2.html


\

I noticed this too.. they claim it's pretty much a done deal.
Frankly I'm completely shocked if it turns out to be true. I'd have to guess that Apple's going to make them cover any expenses that they don't make up after selling refubs.

I'm still taking a 'we'll see' stance on this one.

[Mord]

Originally posted by thogs_cave
Just one small anal-retentive nit to pick - the G5 is not a server chip, it's a workstation/deskop chip. There are some decided differences between it and the POWER CPU's that it is based on.

IBM seems to think it is a server chip seeing as they use 1.6ghz 970's in there eserver's

[MacRAND]

Originally posted by 3G4N
think secret has picked up sources on this too...
http://thinksecret.com/news/virginiatech2.html
\ Well, that's about as official without being official as it can get.
I wonder if the matching of clock speed at 2GHz between the original G5 PPC and the new G5 Xserves is a coincidence? More than one person has said that it is important in a cluster of this design that ALL the nodes be the same speed. How about "dual" configuration?

[denjeff]

would the shorter interconnections between the servers boost up the power? I am not into supercomputing, but I remember from class that if the wire is shorter, the communication is more efficient. the course was not about fiber networks, so I don't know if it's applicable to this... but there is a big difference between 1100 powermacs and 1100 xserves...

[Mac89]

Virginia Tech usually auctions its surplus property through the Commonwealth of Virginia's Dept. of General Serivces, Division Purchase and Supply. They have auctions about 10-11 times/yr. Anyone present may bid. State law may force VT to auction the machines. Only time will tell.

http://vbo.dgs.state.va.us/VBO/Docs/Auction_Bid_Schedule.asp

[MacRAND]

Originally posted by Mac89
Virginia Tech usually auctions its surplus property through the Commonwealth of Virginia's Dept. of General Serivces, Division Purchase and Supply. They have auctions about 10-11 times/yr. Anyone present may bid. State law may force VT to auction the machines. Only time will tell.

http://vbo.dgs.state.va.us/VBO/Docs/Auction_Bid_Schedule.asp Why would VT get to keep the old G5 Macs if it is a 1:1 exchange with Apple?
VT gets the new G5 Xserves
Apple gets the old G5 towers in exchange
:cool:

[sambo.]

at least they won't be gobbling up the entire production run of the G5 revision when it gets announced.
i think Apple would be keen to see the VT cluster take outright honours in the "Fastest Computer EVVARRRRR" stakes. :D

[homerjward]

why's everyone saying system-x is apple's first supercomputer? apple's had one since august of 99! want a supercomputer? buy a g4. not that a gflop is much anymore lol, but it was once. it even had freakin export restrictions. an apple supercomputer is nothing new. it's not a supercomputer anymore is it? :confused: if i were vt i wouldnt care about catching the earth simulator, because by the time they did, japan could have 1100 earth simulators in a cluster! :eek: :eek: :eek: :eek: :eek: what i'm waiting for is an ipod raid cluster! world's largest mp3 player, baby!

[BeoVir]

why's everyone saying system-x is apple's first supercomputer? apple's had one since august of 99! want a supercomputer? buy a g4. not that a gflop is much anymore lol, but it was once. it even had freakin export restrictions. an apple supercomputer is nothing new. it's not a supercomputer anymore is it? :confused: if i were vt i wouldnt care about catching the earth simulator, because by the time they did, japan could have 1100 earth simulators in a cluster! :eek: :eek: :eek: :eek: :eek: what i'm waiting for is an ipod raid cluster! world's largest mp3 player, baby!

I am a small bit confused, what super computer from 99? :confused:

[chaos86]

I am a small bit confused, what super computer from 99? :confused:

the original powermac g4s and cubes were advertised as the worlds first consumer supercomputer or something like that. the g4 that youre probably reading this on is, in fact, a supercomputer by the standard of 1 gigaflop = supercomputer. (that standard was made up like a decade ago)

[TednDi]

If apple is using Virginia Tech's sucess to test the supercomputer concept. Why wouldn't they give som R&D money to the university. As long as the code that gets written and tested gets back to apple eventually and the results of the efforts get published. Apple could get some tax breaks and publicity and R&D on the cheap. So, why wouldn't they get the first crack at the new hardware. Perhaps they have the 3ghz dualies who would tell? VT no, not if they wanted future funding and top secret technology. Apple, no not until they could boast of a big gain.

Makes sense to me. But, then again, I am sitting on the outside looking in?

http://www.cfr.vt.edu/

http://www.cfr.vt.edu/publicity/index.html

http://www.cfr.vt.edu/publicity/vtmagwinter03.html

[macsrus]

The Title says it all

[ffakr]

the original powermac g4s and cubes were advertised as the worlds first consumer supercomputer or something like that. the g4 that youre probably reading this on is, in fact, a supercomputer by the standard of 1 gigaflop = supercomputer. (that standard was made up like a decade ago)

Yea, the deal was/is that computers fall under the designation of military ordinance (or something related) if they are fast enough. The idea is, we don't want to be selling F-16 or computers fast enough to help develop F-16s (or Nukes) to our enemies.
Prior to the release of the G4, the Government hadn't revised the definition of a "Supercomputer" for a long time. Supercomputers were regulated so they couldn't sold to any country that had arms restrictions in place. The old limit was 1GFlop. Since a G4 could sustain 1GFlop, it was a supercomputer according to the government. Not only could apple legally call it a supercomputer when it obviously wasn't, Apple was also not able to export the G4 to a number of countries.. I believe that included France for a period.
Congress revised the limit to, I believe, 50 GFlops.. which was fast for a single computer node back in 1999 or 2000 but it was still far from supercomputer power. It basically took all PCs out of the range of Supercomputer for a while.
Flash forward to today...
2 x 2.5 GHz G5s with two double precision floating point units per cpu. that is a theoretical max of 10 GFlops if you could push a FLOP through each cycle. Each chip also has Altivec which can process 4 32bit FLOPs at once. That's an additional theoretical max of 20 GFlops. You'd never see sustained rates this high.. but we could see a desktop machine within a couple years that was pushing 50 GFlops... maybe 2 dual core chips with altivec running at 4 GHz... with Altivec.