Einstein in the News
Anonymous
22 May 2008 12:45:15 UTC
Topic 13674
(moderation:
Quote:
Seriously it's easy to see there's some mean hardware there. One of my friend's computers, whom I've just enrolled for E@H, is wingman to one of the 2x4 = 8 core Xeon brutes.
BTW, ATLAS has only four cores per node. At the end that gives us more FLOPs per Euro and more network bandwidth per core. And having worked in the rooms of Merlin and Morgane what I especially like about ATLAS is that (apart from the Woven Systems core switch) it's surprisingly silent.
BM
Einstein in the News
)
Do you refer to c't 13/2008, S. 18: Supercomputer? There Andreas Stiller points out that the average Einstein@home host doesn't have the necessary "local capacity" ("lokale Speicherkapazität"). This is a bit misleading; actually the limiting factor for the searches we can run on Einstein@home is not so much the local power of the participant's machines, but the bandwidth with which we can transfer data between these and our servers. If we'd have a network to the Einstein@home participants as powerful as that we have on ATLAS - non-blocking "wire-speed" Gigabit-Ethernet between 1342 Nodes (4 cores per node, not 671 with 8 cores as Andreas wrote), we wouldn't need ATLAS.
ATLAS will run different types of searches for gravitational waves that are not suitable for Einstein@home, e.g. because you can't easily split the required data into such small pieces as we do for the continuous wave search. Probably similar to what Nemo does now (the "nice" CPU is usually Einstein@home).
Edit: ATLAS will probably contribute its idle time to Einstein@home from S5R4 on.
BM
BM
RE: Will ATLAS appear in
)
YES. The Linpack tests mentioned in the article have been done just for that.
BM
BM
RE: RE: Will ATLAS appear
)
Top 500 June 2008 Rank 58
BM
BM