*DAMN R6 Forum

Virginia tech has decided to replace the tower G5's with Xserve G5's.
The storage space for the cluster is now 280m2 but will be decreased to 92m2. This means more space to put more processors >>> more power!
VT will prolly aim for the second spot on the supercomputer list!

Check this link (http://www.techweb.com/wire/story/TWB20040127S0003).

Wasn't this news released like a week ago?

Sort off... its just more official now...

Apparently there is quite a substantial interest in the VT Super Comp set up. A few agencies including Nasa and FBI are apparently interested in buying the VT 'Kit' used to build the comp....

It can't be anything less than more great damn news for apple though :D  top spot No.2 here we come

where's Aramath?

its the third biggest in thee world and they only spent 5mill shit. :o

Unbelievable isn't it.. i belive the exact(ish) cost was 5.2 Million US Dollars... not bad at all when u consider the 2 comps that beat it... not bad at all

This is actually almost month old news.  They talked about this at Macworld.  Don't know why some news services are just picking it up now.

Holy hell, just 5.2 million (7 according to the article, but who's counting!?) compared to the 200 million of the two leading computers?! Thats insane!! Anybody know the comparison between the Virginia Tech computer (to be) and the current top two? In terms of computing power.  
EX: If the Virginia Tech one is just a bit behind those two in terms of computing power, but over 100 times cheaper...i think Apple might be getting alot of orders in the future.  Or, they hopefully will.

Yep these are the figures as im aware ;D you might want to sit down to read these though if your currently stood up!  :o

the Top spot is held by Earth simulator in Japan

Earth Simulator 35.86 Terraflops     Cost: 350 Million US Dollars
Los Alamos Lab  13.88 Terraflops     Cost: 215 Million US Dollars
Virgina Tech       10.28 Terraflops     Cost: 5.2  Million US Dollars

so by a mile and a freaking half you get shit loads more for you buck. Its so much cheaper its almost stupid. just doubling the size of VT's SC would cost a total 10.4mill... and they could be hitting near on 20Terraflops - and thats a very very firm 2nd place!  

imagine using it for gaming...aaahhh *drools*

The installation has been up and running for a few months, but the swapping of G5s for Xserve servers will shrink the size of the installation. ?We'll cut the space used by a factor of three,? said Lynn Nystrom, university spokesperson. ?We'll go from 3,000 square feet to 1,000 square feet.?

What a geeky way to say that.

lol... well come on you'd feel a bit geeky if you had that as your desktop ;)  
A third the size.. Surely it only makes sense to fill all that space up with extra Xserves.. that would defianatly put them close to top spot.. hell if they really did do that i think they would hit over 30Terrabytes!! omg

If they filled up the extra 2000 square feet of space they now have open (once the new Xserve's are in place) that would put Virginia Tech behind Japan's Earth Simulator by about .3 terraflops, for 15.6 million dollars total, which is 16 times cheaper than the Earth Sim.  

So, imagine what you would get, if the people who own the Japan Earth Simulator, had all that cash now, and instead of buying their existing system, bought 16 of these fully loaded Virginia Tech Xserve style super computers....thats like...576 Terraflops, for almost the exact same price.  I feel faint. (http://dynamic.gamespy.com/~damnr6/yabbse/YaBBImages/shocked2.gif)

And Virgina tech apparently want to build a second on and have it up and running by 2006! Its gonna be the daddy!

The sad thing about these machines falling into Hokie hands is they dont have the mental capacity to use them. Anybody going to Tech from the region must have been passed up by Georgetown, VIRIGINIA, or William and Mary.

Exept its usage probably will be not predominatly by VT students... I can imagine the Uni making a lot of cash 'letting out' the supercomputer to companies and research faculties needign that bit of technical ooomph ;)

Surely there is some roof how powerful clustered powermacs can get, a bottleneck somewhere which the earth simulator dosen't have...

I think it should be fairly easy to get past a problem like that, you just have to be a bit inventive.

To me...that sounds impossible, although i might be wrong.

So if they added lets say... what was that number... 16 x 1100 G5's...hmmm i need a goddamn supercomputer to count this.... If they put together 17600 G5's it would be 16 times as powerful as it is now...? Or lets say 100.000 G5's what about then?

I havent really got into the whole VT cluster thing, and im no expert at LAN either but...isnt gigabit lan the fastest cable to connect computers together?...sounds like a bottleneck to me

I suppose it all depends how it works... Im no expert on this an i know little about the VT setup but im guessing that really there shouln't be bottleneck issues over the switches as the nodes are only using these to move the results of information... all the work is still sitting on the processors on each node, only the 'question and answer' is traveling over the network...

Im not sure ive just said what im thinking... its allworked out up top but its been a bloody long day and i can't think.. someone please explain what i think i mean?

Heh, the GBit network cards in modern computers are far from the most advanced, what they use here is called something like 'HyperBand'? Anyways, it's much more powerful and it's what any supercomputer-maker with respect for his work uses.

So basically you could put together 1million G5's and make a computer smarter than a human? create AI?

hehe, my examples keep getting bigger and bigger  :D

Yes, but it wouldn't be smarter than just "any" human....only smarter than YOU, Joka :)

If you can just add G5's indefinately and each time you add a new G5 the computing power just gets bigger, and there no roof.

How come you can't do this even cheaper with PC's? If you can add computers indefinately and divide the work load amongst the connected computers then you should be able to do this with any computer right?

How about G3's? Does a 100.000 G3 powermac cluster equal a...100 G5 cluster?

Just curious...

Theoretically, they could scale this XServe approach indefinitely. However, realistically, the more nodes you have, the less efficient it will be per node. While I'm no supercomputer expert, I'm pretty sure they are quite far from any serious diminishing gains.


It's called InfiniBand, and they are more interested in the low latency it provides. The 10 gigabit bandwidth is just an added bonus over gigabit ethernet.

lol i bet.


I doubt that very much. The G5 is built in a way which is far more friendly to a potential supercomputer builder... Its I O Archetecture (omg that is spelt shockingly badly)  means that it can effectively interconnected.
I suppose it all depends what you mean when u say... "any computer" ;)

Theoretically just adding shitloads of computer power doesn't do shit to simulate a human brain if it doesn't have any kind of software emulation, raw electronics can't gain sentience, ever.

Thanks Ace, forgot the name of it for a sec.

Just speculating about the setup of the system, to possibly avoid the bottleneck problem.

So, first hurdle in my plan, for this all to be a supercomputer, does it all need to be connected together, so each computer(node) can talk directly to every other computer in the system?  (thats alotta cable)
I would think no, and this is why.

My little crack-pot idea, is to break the computers up into individual clusters, of perhaps 10? 15? who knows, just break them up, and make a setup kind of like a pyramid (if you drew it on paper)
Like, having the clusters being the grunts of the operation, doing the raw computing, then sending it back up a level, to a cluster or individual computer, which receives information from mebe 30 (or more, depending) clusters, and combines all that data together to form a piece of info, then forwards it on to another level up, which does the same thing (receiving info from again, 30 or so computers) and so on, until the information is finally compiled completely, so it can be displayed on one screen.

Or maybe im just retarded and should delete this right away because i have no idea how computers divy up the workload.  That always works too =)

     That's actually more or less how some parallel supercomputers work. The task at hand is divided between clusters of processors, where each processor has its own RAM for its computations, and each cluster has communal RAM to allow the processors to communicate within the cluster.

in fact i guess computers work like that just on the base level... so i gues u could say that its lots of pyramids x2(duel processors ;) ) x freaking thousands = Hello i own the 3rd fastest supercomputer in the world