Was at the computer shoppe the other day, drooling over the fancy new motherboards. Saw one that supported four (that's right - 4!!) of the new AMD 64s. Would have taken my entire pension for the next year to buy and populate that monster!!
Just for grins, please give us a shock on how much you’d estimate the price to be??
Thanks
TA
My recollection says the motherboard was over $1100 itself (totally bare)
My recollection says the motherboard was over $1100 itself (totally bare)
How do you say nicely – “Thanks for the Bad News” – That would be one powerful system if configured properly, but Wow, what an expensive piece of power.
Okay, now that I have gotten some work units uploaded from my AMD 64 4200+ dual core I can compare it against my Pentium D 840 (No HT). Here are the results.
SETI
AMD 64 4200+ 4089.55 seconds per work unit average per core.
68.16 minutes or 42.25 work units per day with both cores running.
Pentium D 840 3378.40 seconds per work unit average per core.
56.30 minutes or 51.15 work units per day with both cores running.
EINSTEIN
AMD 64 4200+ 20498 seconds or 8.44 work units per day.
Pentium D 840 26513 seconds or 6.52 work units per day.
As you can see, Einstein loves AMD, and SETI loves Intel. Both comparisions are using optimized clients and apps where available.
Yea, I have a dual socket Tyan mobo K8WE with a pair of Opteron 275's. (2.2Ghz) for a total of 4 cpu's. Each cpu does a Einstein wu in 5.5 hrs give or take a few minutes. http://einsteinathome.org/host/336903/tasks
Any one here up to designing and building there owu MB to hold all there CPU's? Even with slower P4's or AMD XP's having ten on one board would save a lot on other hardware. Having a whole farm in one box would be good.
Would still need a 2nd computer for a backup though.
Any one here up to designing and building there owu MB to hold all there CPU's? Even with slower P4's or AMD XP's having ten on one board would save a lot on other hardware. Having a whole farm in one box would be good.
This is the basic idea behind the "blade servers" and rack servers - rather than one huge MB, there's a backplane that has all the "support" stuff, and "lower-cost" cards holding one or two CPUs each that slide into the backplane. If they weren't so &*%^(* expensive... but for whatever bizarre reason, most brands have hard drives and such on each "blade" and are priced over USD$2K each. Apple at least has a "cluster node" that leaves out everything but what is needed for raw CPU power, but with dual 2.3GHz G5's, it's still $3K - lots less than the $4K for the same rack-mount CPUs _with_ drives, video, etc., but still...
Bottom line is that mass production savings means the cheapest way to go is one or two CPUs per MB, one MB per case... sigh. Whoever it was that said S-100 should have won (Paul?) was right.
What we REALLY need for parallel computing to take off is for some company to build a "blade server" type box where each "blade" has _only_ a CPU, _OR_ RAM, _OR_ hard drives. The chassis would have the fans and the PCI slots and so forth. The customer could either put one CPU and one RAM and 14 hard drives in (for a data server) or one RAM and one hard drive and 14 CPUs in (for a compute server). Any hardware manufacturers listening?
RE: RE: Was at the
)
My recollection says the motherboard was over $1100 itself (totally bare)
If I've lived this long - I gotta be that old!
RE: My recollection says
)
How do you say nicely – “Thanks for the Bad News” – That would be one powerful system if configured properly, but Wow, what an expensive piece of power.
Thanks,
TA
Theory of Gravitational Waves & LIGO
Laser Interferometer Space Antenna - LISA
JPL-Caltech
Okay, now that I have gotten
)
Okay, now that I have gotten some work units uploaded from my AMD 64 4200+ dual core I can compare it against my Pentium D 840 (No HT). Here are the results.
SETI
AMD 64 4200+ 4089.55 seconds per work unit average per core.
68.16 minutes or 42.25 work units per day with both cores running.
Pentium D 840 3378.40 seconds per work unit average per core.
56.30 minutes or 51.15 work units per day with both cores running.
EINSTEIN
AMD 64 4200+ 20498 seconds or 8.44 work units per day.
Pentium D 840 26513 seconds or 6.52 work units per day.
As you can see, Einstein loves AMD, and SETI loves Intel. Both comparisions are using optimized clients and apps where available.
Team MacNN - The best Macintosh team ever.
I just started crunching with
)
I just started crunching with my new AMD 64 X2 Dual Core Processor 4800+ and it seems to turn out results at a steady (+/- 1 minute) 18,450 seconds.
RE: I just started
)
Richard. Have you tried it on any other projects yet? I would be interested in hearing of your results.
Team MacNN - The best Macintosh team ever.
RE: Just fine I Want
)
I Want One? Droooool
RE: Dual Xeons work fine
)
Quad Xeons work better ! ;]
Now where did put that 8way Xeon box.....hrmmm 8)
RE: Yea, I have a dual
)
How much did that set you back?
Any one here up to designing
)
Any one here up to designing and building there owu MB to hold all there CPU's? Even with slower P4's or AMD XP's having ten on one board would save a lot on other hardware. Having a whole farm in one box would be good.
Would still need a 2nd computer for a backup though.
Try the Pizza@Home project, good crunching.
RE: Any one here up to
)
This is the basic idea behind the "blade servers" and rack servers - rather than one huge MB, there's a backplane that has all the "support" stuff, and "lower-cost" cards holding one or two CPUs each that slide into the backplane. If they weren't so &*%^(* expensive... but for whatever bizarre reason, most brands have hard drives and such on each "blade" and are priced over USD$2K each. Apple at least has a "cluster node" that leaves out everything but what is needed for raw CPU power, but with dual 2.3GHz G5's, it's still $3K - lots less than the $4K for the same rack-mount CPUs _with_ drives, video, etc., but still...
Bottom line is that mass production savings means the cheapest way to go is one or two CPUs per MB, one MB per case... sigh. Whoever it was that said S-100 should have won (Paul?) was right.
What we REALLY need for parallel computing to take off is for some company to build a "blade server" type box where each "blade" has _only_ a CPU, _OR_ RAM, _OR_ hard drives. The chassis would have the fans and the PCI slots and so forth. The customer could either put one CPU and one RAM and 14 hard drives in (for a data server) or one RAM and one hard drive and 14 CPUs in (for a compute server). Any hardware manufacturers listening?