DrMrLordX
Lifer
- Apr 27, 2000
- 22,948
- 13,039
- 136
Just wanted to toss this in:
while I can neither prove or disprove my statements about OCing and power consumption with hard data, there are a few things that need to be taken into account. Even a "simple" OC is going to require you to stay at stock vcore, which prohibits you from undervolting your CPU. Results vary from core to core but it is reasonable to assume that if you can OC a chip at stock vcore, you can stick to stock speeds and undervolt your CPU. Also, power consumption does rise as you raise clock speed even if you leave vcore alone, and on some CPUs this increase in power consumption can be substantial. Take a look at what happens to X2s at around 2.8 ghz or higher . . . it's scary (especially the 90nm ones).
Lastly, consider the fact that CPUs do not always scale well when overclocked, or at least not perfectly. This is usually related to cache size, architecture, placement of the memory controller, memory used, etc.
Taking all that into account, you can run your CPU and memory faster (even keeping vcore/vdimm static, though you will probably have to raise vdimm) and chew up more power while not getting 100% of the extra performance you should be getting considering the increased clock speed. You will also burn up more power from pushing your motherboard harder, and you'll take an additional hit pulling power from the wall because even those 80-85% efficient PSUs waste power, thereby amplifying the effects of increased power draw. But why? Alternatively you could undervolt your CPU, tear out all drives except maybe a cheap HDD (or not even that if you boot off a network), maybe undervolt your RAM, take most if not all the fans out of your system and run with passive heatsinks, and save a ton on power.
Just as an example, do you think you'd get better performance/watt out of an undervolted Wolfdale or Yorkfield running at stock speeds on a barebones system using onboard graphics (or no graphics), an onboard NIC, no soundcard, no optical drive, etc etc powered by something like an Earthwatts 380W; or a Wolfdale/Yorkfield on a similarly stripped-down system with a nice set of case fans and a good HSF that's bumped up to 3.2 ghz on stock volts and has RAM running at 1:1 with a higher VDIMM, plus a motherboard that's getting a lot hotter thanks to the stress on the Northbridge?
It wouldn't be hard for that undervolted system to be running at half the power consumption rate of the overclocked system. Would the overclocked system be getting 100% more performance? I doubt it. You won't even get a 100% overclock, and the overclock you do get won't result in an equivalent amount of actual performance increase due to scaling issues.
Don't get me wrong . . . I love to overclock, but it does not improve performance/watt in most (if any) circumstances, particularly not when it comes to dedicated DC cruncher boxes. Overclocking sure is a lot more fun than being efficient, though . . .
btw, I really appreciate the work that the f@h crew at Stanford is doing. My step Grandfather succumbed to both Alzheimer's Disease and Parkinson's Disease years ago, and it was not pretty. He forgot who he was and couldn't swallow, and shook a lot. F@h's research may lead to cures for both diseases. So I say, have at it!
while I can neither prove or disprove my statements about OCing and power consumption with hard data, there are a few things that need to be taken into account. Even a "simple" OC is going to require you to stay at stock vcore, which prohibits you from undervolting your CPU. Results vary from core to core but it is reasonable to assume that if you can OC a chip at stock vcore, you can stick to stock speeds and undervolt your CPU. Also, power consumption does rise as you raise clock speed even if you leave vcore alone, and on some CPUs this increase in power consumption can be substantial. Take a look at what happens to X2s at around 2.8 ghz or higher . . . it's scary (especially the 90nm ones).
Lastly, consider the fact that CPUs do not always scale well when overclocked, or at least not perfectly. This is usually related to cache size, architecture, placement of the memory controller, memory used, etc.
Taking all that into account, you can run your CPU and memory faster (even keeping vcore/vdimm static, though you will probably have to raise vdimm) and chew up more power while not getting 100% of the extra performance you should be getting considering the increased clock speed. You will also burn up more power from pushing your motherboard harder, and you'll take an additional hit pulling power from the wall because even those 80-85% efficient PSUs waste power, thereby amplifying the effects of increased power draw. But why? Alternatively you could undervolt your CPU, tear out all drives except maybe a cheap HDD (or not even that if you boot off a network), maybe undervolt your RAM, take most if not all the fans out of your system and run with passive heatsinks, and save a ton on power.
Just as an example, do you think you'd get better performance/watt out of an undervolted Wolfdale or Yorkfield running at stock speeds on a barebones system using onboard graphics (or no graphics), an onboard NIC, no soundcard, no optical drive, etc etc powered by something like an Earthwatts 380W; or a Wolfdale/Yorkfield on a similarly stripped-down system with a nice set of case fans and a good HSF that's bumped up to 3.2 ghz on stock volts and has RAM running at 1:1 with a higher VDIMM, plus a motherboard that's getting a lot hotter thanks to the stress on the Northbridge?
It wouldn't be hard for that undervolted system to be running at half the power consumption rate of the overclocked system. Would the overclocked system be getting 100% more performance? I doubt it. You won't even get a 100% overclock, and the overclock you do get won't result in an equivalent amount of actual performance increase due to scaling issues.
Don't get me wrong . . . I love to overclock, but it does not improve performance/watt in most (if any) circumstances, particularly not when it comes to dedicated DC cruncher boxes. Overclocking sure is a lot more fun than being efficient, though . . .
btw, I really appreciate the work that the f@h crew at Stanford is doing. My step Grandfather succumbed to both Alzheimer's Disease and Parkinson's Disease years ago, and it was not pretty. He forgot who he was and couldn't swallow, and shook a lot. F@h's research may lead to cures for both diseases. So I say, have at it!
Facebook
Twitter