Jump to content

  • Log In with Google      Sign In   
  • Create Account

#ActualKing Mir

Posted 25 August 2013 - 01:23 AM

Power consumption is proportional to how much work the processor is doing. Optimizing for speed does that, and tries to increase the number of things the processor does in parallel. Instruction Level parallelism does not add any additional overhead, that is not already incurred by virtue of the processor supporting it. So as long as you're not optimizing by making a program multi-threaded, optimizing for speed will optimize for power.

Optimizations that reduce access to main memory or improve cache performance may also reduce the power consumption of memory, I think. Certainly reducing page faults and Hard-disk IO will give power savings.

Optimizing by moving computation to the GPU is, IMO, likely to reduce power when it also improves speed, because these kinds of computations tend to use memory less efficiently when run on the CPU, but I'm not confident of that.

On the other hand, making a program threaded, or use threads more efficiently may consume more power, because there is a cost to synchronizing threads. The same applies to multi process or multi machine "programs", though power consumption tends not to be a dominant consideration in network communication.



Others have pointed out that if you optimize for speed, but increase the frame rate, you're not reducing power consumption.

#1King Mir

Posted 25 August 2013 - 01:21 AM

Power consumption is proportional to how much work the processor is doing. Optimizing for speed does that, and tries to increase the number of things the processor does in parallel. Instruction Level parallelism does not add any additional overhead, that is not already incurred by virtue of the processor supporting it. So as long as you're not optimizing by making a program multi-threaded, optimizing for speed will optimize for power.

Optimizations that reduce access to main memory or improve cache performance may also reduce the power consumption of memory, I think. Certainly reducing page faults and Hard-disk IO will give power savings.

Optimizing my moving computation to the GPU is, IMO likely to reduce power when it also improves speed, because these kinds of computations tend to use memory less efficiently when run on the CPU, but I'm not confident of that.

On the other hand, making a program threaded, or use threads more efficiently may consume more power, because there is a cost to synchronizing threads. The same applies to multi process or multi machine "programs", though power consumption tends not to be a dominant consideration in network communication.



Others have pointed out that if you optimize for speed, but increase the frame rate, you're not reducing power consumption.

PARTNERS