Is VT Going For #1 Cluster?

schneb · 03-23-2004, 12:13 PM

There is a month delay in bringing out the dual G5 Xserve. No big problem. Some have speculated that the first run units are headed to Virginia Tech to replace the previous G5 tower cluster. It is my prognosis that Virginia Tech is going for the number one slot for the fastest cluster in the world. After all, this is free, positive publicity for both VT as well as Apple.

I am very curious to see just how small a space this cluster will occupy. The current leader, Yokohama's Earth Simulator, occupies a WAREHOUSE!

http://www.es.jamstec.go.jp/esc/eng/GC/index.html

...whereas the dual Xserves will most likely take up a single classroom.

This is going to be fun to watch--any thoughts?

Craig R. Arko · 03-23-2004, 01:08 PM

Given that they'd need to triple the speed to catch Earth Simulator, I kind of doubt they have the budget to do it.

IBM, on the other hand, should have little problem seizing the top spot in the next year or so. Maybe even holding it.

staypuft · 03-23-2004, 02:24 PM

I agree that VT is not looking for the number one spot. Though with the smaller size of the X Serve, they could fairly easily take the number two spot. This is assuming that they have the budget to increase the number of computers and not just replace them one for one. As Craig stated, they would need over 300% of their current power to reach number one, but they only need 135% to take number two.

schneb · 03-23-2004, 04:48 PM

Craig, I seem to remember that they were hoping to hit in the top ten and that they were flumoxed to find that they hit #3. It seems as if they would not have to triple the amount of processors, that the increase would be exponential and not a linear speed increase, given architecture of the G5 processor. Perhaps I am mistaken.
As for cost, given the success of the first experiment, could they not pool resources from other universities and research firms that could benefit from the cluster?
I'm probably on the losing side of this wager--it's fun nonetheless.

staypuft · 03-23-2004, 09:43 PM

You are correct in that the output is not a straight line, but your conclusion is actually the reverse of what happens. The theorectical output would be a straight line, but as you increase the number of the processors the actual output curve drops further below the theoretical output. So in actuallity, they would need more like 325% the processors to get 300% of the output.

As for getting money... my guess is that they are going to start with a 1 to 1 replacement of the current G5s with X-Serves. Then as they charge money to process jobs for companies they may add more X-Serves to the cluster since they will have extra space. Though we're all likely to find out in a month or two what they do.

View Poll Results: Is Virginia Tech Going For #1 Cluster?
Yes! Going for #1!	2	33.33%
No. They're satisfied with #3.	3	50.00%
Don't know--but interesting.	1	16.67%
What is a cluster?	0	0%
Voters: 6. This poll is closed

03-23-2004, 04:48 PM	#4
schneb MVP Join Date: Dec 2003 Location: Redlands, CA Posts: 2,300	Craig, I seem to remember that they were hoping to hit in the top ten and that they were flumoxed to find that they hit #3. It seems as if they would not have to triple the amount of processors, that the increase would be exponential and not a linear speed increase, given architecture of the G5 processor. Perhaps I am mistaken. As for cost, given the success of the first experiment, could they not pool resources from other universities and research firms that could benefit from the cluster? I'm probably on the losing side of this wager--it's fun nonetheless. __________________ Tour Israel and Jordan via CD-ROM

03-23-2004, 09:43 PM	#5
staypuft All Star Join Date: Sep 2003 Location: USA Posts: 628	You are correct in that the output is not a straight line, but your conclusion is actually the reverse of what happens. The theorectical output would be a straight line, but as you increase the number of the processors the actual output curve drops further below the theoretical output. So in actuallity, they would need more like 325% the processors to get 300% of the output. As for getting money... my guess is that they are going to start with a 1 to 1 replacement of the current G5s with X-Serves. Then as they charge money to process jobs for companies they may add more X-Serves to the cluster since they will have extra space. Though we're all likely to find out in a month or two what they do. Last edited by staypuft; 03-23-2004 at 09:48 PM.

Thread Tools
Show Printable Version Email this Page
Display Modes	Rate This Thread
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode	Rate This Thread:

03-23-2004, 12:13 PM	#1
schneb MVP Join Date: Dec 2003 Location: Redlands, CA Posts: 2,300	Is VT Going For #1 Cluster? There is a month delay in bringing out the dual G5 Xserve. No big problem. Some have speculated that the first run units are headed to Virginia Tech to replace the previous G5 tower cluster. It is my prognosis that Virginia Tech is going for the number one slot for the fastest cluster in the world. After all, this is free, positive publicity for both VT as well as Apple. I am very curious to see just how small a space this cluster will occupy. The current leader, Yokohama's Earth Simulator, occupies a WAREHOUSE! http://www.es.jamstec.go.jp/esc/eng/GC/index.html ...whereas the dual Xserves will most likely take up a single classroom. This is going to be fun to watch--any thoughts? __________________ Tour Israel and Jordan via CD-ROM

03-23-2004, 01:08 PM	#2
Craig R. Arko Site Admin Join Date: Dec 2001 Location: Minneapolis, MN Posts: 3,965	Given that they'd need to triple the speed to catch Earth Simulator, I kind of doubt they have the budget to do it. IBM, on the other hand, should have little problem seizing the top spot in the next year or so. Maybe even holding it.

03-23-2004, 02:24 PM	#3
staypuft All Star Join Date: Sep 2003 Location: USA Posts: 628	I agree that VT is not looking for the number one spot. Though with the smaller size of the X Serve, they could fairly easily take the number two spot. This is assuming that they have the budget to increase the number of computers and not just replace them one for one. As Craig stated, they would need over 300% of their current power to reach number one, but they only need 135% to take number two.