Folding Forum

Hi,

Apologies if this has been discussed before - I am aware that work is credited on the basis of benchmarks to ensure 'equal points for equal work'. However there's another variable - power usage. For example most GPU projects on Core 15 generate about 77% power usage (TDP) on my Kepler card. p7625 for example, however, drives my card to 103% power usage - which is about 33% more. Clearly, the same work on different hardware (or core) will end up in different power figures, but as there are projects which end up in different power usage on the same hardware and on the same core I'm wondering whether it's been considered to reward such more power hungry projects in terms of points. Or is it already the case?

Thanks

Donors to FAH choose to donate to science by running FAH. The Pande Group considers points to be a measure of the scientific work you've contributed. They make no allowances for what it costs you do make that donation -- either because you choose to buy certain equipment or it uses a certain amount of power.

People who live in a cold climate benefit from electrical room heating. People who live in a hot climate generally have to spend twice as much because they also power an air conditioner to pump the heat outside. Neither has anything to do with the amount you've contributed to science.

In addition to what bruce has stated, there isn't a universal price of power. It varies from country to country and even within a city, may vary on the time of day. Anything that involved money on the donor's side can't really be used in the calculation of points since there isn't a universal fix price thus could end up being unfair to some.

By having the system based on the scientific value, it will remain constant regardless of where you fold from. The WU will give you (virtually) same points from (almost) anywhere in the world as long as it has the same TFP.

Thanks. That's all clear, but it wasn't my point. The issue is that different projects have different watt per work ratio on the same hardware and core - sometimes quite substantial. I understand that from Standford's perspectives what matters is getting the most work done, the sooner for which donors are compensated accordingly in terms of points. I also understand the cost of kW/cold/warm climate argument, but that's a constant (relatively) and is not project dependant. However, from the donor's perspective, the same unit of work has different cost to the donor depending on the project and in that regard things are not fair. Hence I was wondering whether that's something taken into account - I understand not.

Short answer: NO

First step, of course, would be to start negotiating with your local electricity company and tell them that xx kWh of my electricity is going to charity, can you give me a discount?

Napoleon wrote:First step, of course, would be to start negotiating with your local electricity company and tell them that xx kWh of my electricity is going to charity, can you give me a discount?

That's how I look at it. I can contribute to some charity or other and not really have much idea where the money is actually going, or I can pay for the extra power used by leaving the computers on 24/7 (it's not very much) and know what it's being used for.

Breach wrote:The issue is that different projects have different watt per work ratio on the same hardware and core - sometimes quite substantial. I understand that from Standford's perspectives what matters is getting the most work done, the sooner for which donors are compensated accordingly in terms of points. I also understand the cost of kW/cold/warm climate argument, but that's a constant (relatively) and is not project dependant. However, from the donor's perspective, the same unit of work has different cost to the donor depending on the project and in that regard things are not fair.

If you get into what is "fair", you might as well re-write the tax code while you are at it. While you think that it is "fair" to reward projects that use more power, I think that it is merely rewarding inefficiency. There should be the same incentive for everyone to choose and operate the most efficient equipment they can or want to. So buy whatever equipment you want to achieve whatever operating costs you are willing to bear.

JimF wrote:

Breach wrote:The issue is that different projects have different watt per work ratio on the same hardware and core - sometimes quite substantial. I understand that from Standford's perspectives what matters is getting the most work done, the sooner for which donors are compensated accordingly in terms of points. I also understand the cost of kW/cold/warm climate argument, but that's a constant (relatively) and is not project dependant. However, from the donor's perspective, the same unit of work has different cost to the donor depending on the project and in that regard things are not fair.

If you get into what is "fair", you might as well re-write the tax code while you are at it. While you think that it is "fair" to reward projects that use more power, I think that it is merely rewarding inefficiency. There should be the same incentive for everyone to choose and operate the most efficient equipment they can or want to. So buy whatever equipment you want to achieve whatever operating costs you are willing to bear.

Not sure what you mean here. Are you suggesting to dump those watt-hungry WUs and concentrate on "efficient - most points per watt WUs"?

The biggest factor in determining what's watt-hungry is your hardware. The exact Project is much less significant. Every hardware generation changes the productivity factors. Generally this change when technology can replace XX nm traces with YY nm traces. Productivity also varies with clock speed, the number of CPU-cores or the number of GPU shaders. In the case of NVidia GPUs, there was also a major change from Fermi to Kepler, where the goal was to reduce power and heat rather than to increase throughput -- but changes like that are rare. Traditionally the manufacturers emphasize speed increase and don't remind you that you'll probably see it comes at the cost of increased power consumption. You can take advantage some degree of this concept by overclocking/overvolting (increased speed; increased power consumption) or underclocking/undervolting (decreased speed; decreased power consumption).

GPU technology progresses through multiple hardware generations more rapidly than CPU technology.

Let's not kid ourselves. There is no explicit relationship between power and points, but the QRB creates an implicit relationship. Overclock, use more power, earn higher bonuses. Replace a CPU or GPU with a TDP of X with one with a higher TDP, use more power, get faster throughput earn higher bonuses.

Breach wrote:
Not sure what you mean here. Are you suggesting to dump those watt-hungry WUs and concentrate on "efficient - most points per watt WUs"?

Oh god please don't joke about dumping wu's somebody will read this and think it is a good idea. If a WU is dumped then instead of one person having to do it multiple people do and that wastes a lot of time.

Breach wrote:Not sure what you mean here. Are you suggesting to dump those watt-hungry WUs and concentrate on "efficient - most points per watt WUs"?

No, the users should choose efficient equipment and crunch everything thrown at them (I sympathize with Core_16 users however).
Then, PG should just dump the users who dump the work units.