4 ms·

Tech rants #1: PC-s use way too much power in 2021

1 points by hddherman 5y ago

dusted 5y ago
I disagree. The only limit to power consumption should be the practicality of cooling Power is too expensive in 2021 when it should have been too cheap to meter (tm). Plenty of people care about mobile, so I don't have to. There is plenty incentive to reduce power (cooling practicality included) in the mobile space, and this is good, but shouldn't take focus away from achieving higher performance on stationary and server PCs. There is a natural balance between power consumption and performance, and it bends to either side depending on target market, but they do drive each other.
- hddherman 5y ago
  It's a valid point if your goal is to get maximum possible performance. I have to admit that high power usage and the cooling challenges that follow have resulted in some cool solutions (no pun intended): https://youtu.be/7mDkvgO2ncY https://youtu.be/7mDkvgO2ncY Would a 20% drop in performance be acceptable to you if it meant a more than 20% drop in power consumption? Where do you feel the balance is right now and where is it moving towards? Appreciate your input!
  - dusted 5y ago
    I probably need to comment on mobile versus desktop to be fair. In mobile, I'd still rather prefer not to sacrifice performance, but, I'd be very willing to sacrifice features, they're strongly codependent. I'd be very happy to trade my Note 9 for a 5G version of a Nokia 3210 if it could have the original battery life of what.. a week? On desktop, I don't want to sacrifice performance at all, I'd be willing to accept a 50% increase in power consumption for a 20% increase in performance, easily, up until the point where I can't reasonably cool it. I'd be willing to go water-cooled with a custom, out of case, loop, but the chip must be designed for it, overclocking too much gives diminishing returns and decreases reliability too much. Immersion cooling I consider impractical on smaller scales, I'd not want to have to deal with that, and the infrastructure around it. I think there are great symbiotic effects of focusing on both.. The improvements in performance can benefit power-focused devices (smarter designs = more compute for the same joules), and improvements in power-efficiency can benefit performance-focused designs (lower consumption per compute = room for more compute before hitting thermal practicality limits) So I'm not against development, I'm against the reasoning, the way of thinking that I think we're prone to "in our time", of "everything is running out, be careful, use less" to save the planet (tm). I know it seems like an obvious solution "we're running out of X, be more efficient in using X". Well, up to a point, because, as we get more efficient in using X, using small amounts of X for more things are economically viable, and so, we end up using _MORE_ X. Imagine trying to build a a 2021 super computer using 1980s tech.. the resources required, both material and energy would make it unviable, so it didn't happen. But we're spending WAAAHY more energy and resources on compute today than in 1980 exactly because we're using less per unit.

1 points by hddherman 5y ago

I disagree. The only limit to power consumption should be the practicality of cooling Power is too expensive in 2021 when it should have been too cheap to meter (tm). Plenty of people care about mobile, so I don't have to. There is plenty incentive to reduce power (cooling practicality included) in the mobile space, and this is good, but shouldn't take focus away from achieving higher performance on stationary and server PCs. There is a natural balance between power consumption and performance, and it bends to either side depending on target market, but they do drive each other.

hddherman 5y ago
It's a valid point if your goal is to get maximum possible performance. I have to admit that high power usage and the cooling challenges that follow have resulted in some cool solutions (no pun intended): https://youtu.be/7mDkvgO2ncY https://youtu.be/7mDkvgO2ncY Would a 20% drop in performance be acceptable to you if it meant a more than 20% drop in power consumption? Where do you feel the balance is right now and where is it moving towards? Appreciate your input!
- dusted 5y ago
  I probably need to comment on mobile versus desktop to be fair. In mobile, I'd still rather prefer not to sacrifice performance, but, I'd be very willing to sacrifice features, they're strongly codependent. I'd be very happy to trade my Note 9 for a 5G version of a Nokia 3210 if it could have the original battery life of what.. a week? On desktop, I don't want to sacrifice performance at all, I'd be willing to accept a 50% increase in power consumption for a 20% increase in performance, easily, up until the point where I can't reasonably cool it. I'd be willing to go water-cooled with a custom, out of case, loop, but the chip must be designed for it, overclocking too much gives diminishing returns and decreases reliability too much. Immersion cooling I consider impractical on smaller scales, I'd not want to have to deal with that, and the infrastructure around it. I think there are great symbiotic effects of focusing on both.. The improvements in performance can benefit power-focused devices (smarter designs = more compute for the same joules), and improvements in power-efficiency can benefit performance-focused designs (lower consumption per compute = room for more compute before hitting thermal practicality limits) So I'm not against development, I'm against the reasoning, the way of thinking that I think we're prone to "in our time", of "everything is running out, be careful, use less" to save the planet (tm). I know it seems like an obvious solution "we're running out of X, be more efficient in using X". Well, up to a point, because, as we get more efficient in using X, using small amounts of X for more things are economically viable, and so, we end up using _MORE_ X. Imagine trying to build a a 2021 super computer using 1980s tech.. the resources required, both material and energy would make it unviable, so it didn't happen. But we're spending WAAAHY more energy and resources on compute today than in 1980 exactly because we're using less per unit.