Modern components get hot. HOT hot. You've got water cooled NICs, transceivers, CPUs, RAM, you name it. H2O's becoming common as can be in a datacenter.
Well, I know between jack and all about thermodynamics. Next to nothing about flow, pressure, velocity, coolant loiter time, whatever the proper terms are. But it seems to me that you can't have 999PSI of pressure pushing the water through the whole rack, and you can't have 0.000001GPM moving through. There's gotta be some sort of balance.
1) Is this right? Do you need special pumps? Or can you run the entire water cooling loop for rows upon rows of systems off a single consumer-grade pump like you'd find in an AIO?
2) Assuming it's NOT right, how do you figure out pressure drop and all that? Is it measured per rack? Per row? Are there A and B sides for cooling, not just power? Is the pump normally a top-of-rack unit, or stashed with the CRACs?
Basically, how does water cooling work at scale?
Well, I know between jack and all about thermodynamics. Next to nothing about flow, pressure, velocity, coolant loiter time, whatever the proper terms are. But it seems to me that you can't have 999PSI of pressure pushing the water through the whole rack, and you can't have 0.000001GPM moving through. There's gotta be some sort of balance.
1) Is this right? Do you need special pumps? Or can you run the entire water cooling loop for rows upon rows of systems off a single consumer-grade pump like you'd find in an AIO?
2) Assuming it's NOT right, how do you figure out pressure drop and all that? Is it measured per rack? Per row? Are there A and B sides for cooling, not just power? Is the pump normally a top-of-rack unit, or stashed with the CRACs?
Basically, how does water cooling work at scale?