I just have to say it: I really like your hostnames :-D
You mean clusternames? I rather doubt that the individual hosts of the Merlin and Morgane clusters have individual names (or someone at AEI has a lot of imagination and too much spare time). ;-)
1100 computers ok, but when was this cluster built? How many quadcore Intel 45nm processors does that equal to in FP performance? :) It doesnt have to be so many years old to equal just a few hundred QX9650 processors.
1100 computers ok, but when was this cluster built? How many quadcore Intel 45nm processors does that equal to in FP performance? :) It doesnt have to be so many years old to equal just a few hundred QX9650 processors.
That was my point, that kinda got missed. All you'd need are 2000 systems equivalent to Peanut's OSX system and S5R3 would only take about 7 months...
I think the combined horsepower that's out there could be that much, but the other platforms are at a disadvantage, hence the 395.7-day estimated completion...
I just have to say it: I really like your hostnames :-D
You mean clusternames? I rather doubt that the individual hosts of the Merlin and Morgane clusters have individual names (or someone at AEI has a lot of imagination and too much spare time). ;-)
CU
H-B
Okay, you are, of course, right. I was referring to the names of the clusters.
As for the question about how many really fast CPUs you'd need to equal an entire cluster of older CPUs... probably not that many, given that for example my laptop equals roughly six P3 Coppermines, and that's not a power-quadcore with uber overclocking, just a nice dualcore workhorse :-) But still, I think 1100 computers could make a significant difference even if they are a few years old; look at what for example Nemo can still do, and those are just older AMDs with two cores each. In this case, size matters ;-)
I just have to say it: I
)
I just have to say it: I really like your hostnames :-D
RE: I just have to say it:
)
You mean clusternames? I rather doubt that the individual hosts of the Merlin and Morgane clusters have individual names (or someone at AEI has a lot of imagination and too much spare time). ;-)
CU
H-B
1100 computers ok, but when
)
1100 computers ok, but when was this cluster built? How many quadcore Intel 45nm processors does that equal to in FP performance? :) It doesnt have to be so many years old to equal just a few hundred QX9650 processors.
Team Philippines
RE: 1100 computers ok, but
)
That was my point, that kinda got missed. All you'd need are 2000 systems equivalent to Peanut's OSX system and S5R3 would only take about 7 months...
I think the combined horsepower that's out there could be that much, but the other platforms are at a disadvantage, hence the 395.7-day estimated completion...
RE: RE: I just have to
)
Okay, you are, of course, right. I was referring to the names of the clusters.
As for the question about how many really fast CPUs you'd need to equal an entire cluster of older CPUs... probably not that many, given that for example my laptop equals roughly six P3 Coppermines, and that's not a power-quadcore with uber overclocking, just a nice dualcore workhorse :-) But still, I think 1100 computers could make a significant difference even if they are a few years old; look at what for example Nemo can still do, and those are just older AMDs with two cores each. In this case, size matters ;-)
In a Hartree-Fok computation
)
In a Hartree-Fok computation a cluster would be useless, You cannot parallelize all computations.
Tullio
RE: In a Hartree-Fok
)
Thanks! I will look after it.
I always wanted to find a computation that is not parallelizable.