I have built a share of Beowulf clusters, exclusively with AMD MP CPUs (in 2001-2002).
Looking back, the main liniting factor for "growing" these clusters was power consumption and heat production of the completed product. Transmeta breaks through these two limitations.
I worked for a company that put together a 48 node cluster, including the head node. These were dual AMD MP processor systems so we had 96 processors total. In order to build this cluster and test it we had to run it between 3 power sources, Two separated services in the building, and a generator "out back." Our service panel was extremely hot to the touch, to the point that there was much concern that we were going to burn the workplace down.
Moekandu has done the math and I concur. Power is the most limiting factor for larger clusters. *Anyone*, hobbyists, small businesses, and even corporate IT departments will benefit greatly from being able to produce more CPU cycles from far less power!
Heat dispensed by individual CPUs is another factor that seems to go hand-in-hand with power consumption.
Space is another factor that can be considered, I guess... I had 96 processors in 48 2U rack mounted pizza boxes. This neat little project fit into 3 packed racks including 3 KVM switches and two 48 port switches. Now I see that in that space with the Orion MultiSystems option, I can pack in 96 nodes in a fraction of one cabinet (1/3) and I have more room for much more computing fun. Also, no more need to keep a fire extinguisher on hand...
The Transmeta products show themselves to be an option that is sure to be considered more and more in the next year. It is truly in a different class and opens up many opportunities.
Slashdot Mirror
Trash your TV.
OK... I'm a long time listener first time caller.
I have built a share of Beowulf clusters, exclusively with AMD MP CPUs (in 2001-2002).
Looking back, the main liniting factor for "growing" these clusters was power consumption and heat production of the completed product. Transmeta breaks through these two limitations.
I worked for a company that put together a 48 node cluster, including the head node. These were dual AMD MP processor systems so we had 96 processors total. In order to build this cluster and test it we had to run it between 3 power sources, Two separated services in the building, and a generator "out back." Our service panel was extremely hot to the touch, to the point that there was much concern that we were going to burn the workplace down.
Moekandu has done the math and I concur. Power is the most limiting factor for larger clusters. *Anyone*, hobbyists, small businesses, and even corporate IT departments will benefit greatly from being able to produce more CPU cycles from far less power!
Heat dispensed by individual CPUs is another factor that seems to go hand-in-hand with power consumption.
Space is another factor that can be considered, I guess... I had 96 processors in 48 2U rack mounted pizza boxes. This neat little project fit into 3 packed racks including 3 KVM switches and two 48 port switches. Now I see that in that space with the Orion MultiSystems option, I can pack in 96 nodes in a fraction of one cabinet (1/3) and I have more room for much more computing fun. Also, no more need to keep a fire extinguisher on hand...
The Transmeta products show themselves to be an option that is sure to be considered more and more in the next year. It is truly in a different class and opens up many opportunities.
David