06-10-2016, 08:49 AM
I've basically read through most of this Hardware forum, but I was wondering if we could talk about efficiency and scaling on larger systems.
Obviously there are a lot of parameters that could be maximized: Space, Performance, Hardware Cost, Temperature, Power Use, Reliability
Assuming physical space wasn't a premium, and Overall cost (both hardware and power use) per performance was the main consideration, what system designs are the most effective.
General things I have learned from reading past posts:
-with node computing, system design has shifted out rather than up
-power supplies are most effective at around 50% load
-power supplies get exponentially more expensive at a certain point
-3-4 GPU rigs seem to be relatively "easy" to put together with many motherboard / case options--not yet sure whether this makes it more cost effective
I know in the recent past it was a even harder than now to shove more than 3-4 GPUs in a system due to bigger power draws and thus higher temperatures. This seems to have mitigated somewhat with GTX 1080 only consuming 180W.
Anyway, I am not planning any system in particular--I was just wondering how to hypothetical scale efficiently. Any metrics advice or comments would be appreciated.
Obviously there are a lot of parameters that could be maximized: Space, Performance, Hardware Cost, Temperature, Power Use, Reliability
Assuming physical space wasn't a premium, and Overall cost (both hardware and power use) per performance was the main consideration, what system designs are the most effective.
General things I have learned from reading past posts:
-with node computing, system design has shifted out rather than up
-power supplies are most effective at around 50% load
-power supplies get exponentially more expensive at a certain point
-3-4 GPU rigs seem to be relatively "easy" to put together with many motherboard / case options--not yet sure whether this makes it more cost effective
I know in the recent past it was a even harder than now to shove more than 3-4 GPUs in a system due to bigger power draws and thus higher temperatures. This seems to have mitigated somewhat with GTX 1080 only consuming 180W.
Anyway, I am not planning any system in particular--I was just wondering how to hypothetical scale efficiently. Any metrics advice or comments would be appreciated.