Could smp folding support GPGPU?

[P5-133XL]

Turbo clock rates are determined by the number of cores in use, not temperature. While folding, all cores are in operation and as such the turbo speed is significantly lower than if you are running an application that is only using a single core.

but. that does not make sense, if all cores are at 100% load, wouldn't that be a reason to increase the turbo frequency? and considering i have a good 20 degrees before the cpu starts getting uncomfortably hot, it should be at 3.7 at the moment.

The chip is keeping the amount of heat generated within spec. The two variables it uses are the number of cores in operation and the clock-rate (Voltage is not considered, it is assuming that the voltage is staying at the default). The amount of heat generated increases as you use more cores and if the clock rate goes up. If the spec says the chip will not generate more heat then X watts then as you increase the number of cores running the clock rate must decrease to keep the chip within spec. This has nothing to do with the cooling system but rather how much heat the chip will be allowed to produce.

In theory the clock rate will change smoothly. However, that is not what is observed because it changes in big chunks. That is because, in the chip itself it has an algorithum that says 1 core means boost to x clock rate; two cores boost to y; three cores boost to Z; etc.

If you want to increase the clock rate beyond what the chip spec says is its maximum wattage then you have to OC and not turbo-boost. Then you can take advantage of your better than average cooling system.

[pipnina]

people talk about things like: setting maximum wu size. for example, the wu i have at the moment, has taken me 2 days to get to 40% and every night the points i get goes down by 500, so can i set my client control to prevent wu's with a base point of over 800 Points?
i find that when its most efficient

[7im]

Turbo clock rates are determined by the number of cores in use, not temperature. While folding, all cores are in operation and as such the turbo speed is significantly lower than if you are running an application that is only using a single core.

but. that does not make sense, if all cores are at 100% load, wouldn't that be a reason to increase the turbo frequency? and considering i have a good 20 degrees before the cpu starts getting uncomfortably hot, it should be at 3.7 at the moment.

It does turbo up under load, but only to a certain level when all cores are in use. Change your smp to run on only 2 cores and see what happens. Also, google turbo mode profile for your chip to see the parameters for scaling up or not.

[P5-133XL]

You can't. Your choice consists of SMP or multiple uniprocessor slots. In either case you take projects that the servers give you. There is no capability to limit what is given by base point values. If you are having problems completing SMP WU's then go uniprocessor for those tend to be smaller with longer deadlines.

If your PPD is dropping overnight, I wonder if the machine going to sleep rather than actually folding, when not actively used?

[pipnina]

You can't. Your choice consists of SMP or multiple uniprocessor slots. In either case you take projects that the servers give you. There is no capability to limit what is given by base point values. If you are having problems completing SMP WU's then go uniprocessor for those tend to be smaller with longer deadlines.

If your PPD is dropping overnight, I wonder if the machine going to sleep rather than actually folding, when not actively used?

i have a base point wu on my smp;4 of 2200 at the moment, i have currently been running it for 3 days, i dont run it t night because i am not allowed.
this is why it drops overnight, and the reason i wanted to limit it to 800, is because even on a school day, my i5 2500k can complete a wu with that base point in less than 6 hours.
the ones with base points of 100 or less, they only ever get me 3k/5k ppd estimates. whereas the 800 base point gets me over 20k ppd estimates, the 2200 base gets me 5k/6k ppd estimates.
i have tried using 4 uniprocessor slots. but this offers even worse ppd estimates.

[bruce]

Turbo clock rates are determined by the number of cores in use, not temperature. While folding, all cores are in operation and as such the turbo speed is significantly lower than if you are running an application that is only using a single core.

but. that does not make sense, if all cores are at 100% load, wouldn't that be a reason to increase the turbo frequency? and considering i have a good 20 degrees before the cpu starts getting uncomfortably hot, it should be at 3.7 at the moment.

No. It works the other way. If only one (or a few) cores are active, the CPU is running at a lower power level than if they're all busy so the hardware has been designed so that the idle cores can "lend" some of their power to the active ones while keeping the total power at a lower level than you'd be running with everything at 100%. The hardware is allowed to do some dynamic overclocking when the chip is not fully utilized and then back down to a more standard clock rate when fully loaded.

A lot of people run software that never takes advantage of multi-core hardware but this way the manufacturer can overclock their uniprocessor for them. It's a useless feature for folks who run FAH.

[P5-133XL]

i have a base point wu on my smp;4 of 2200 at the moment, i have currently been running it for 3 days, i dont run it t night because i am not allowed.
this is why it drops overnight, and the reason i wanted to limit it to 800, is because even on a school day, my i5 2500k can complete a wu with that base point in less than 6 hours.
the ones with base points of 100 or less, they only ever get me 3k/5k ppd estimates. whereas the 800 base point gets me over 20k ppd estimates, the 2200 base gets me 5k/6k ppd estimates.
i have tried using 4 uniprocessor slots. but this offers even worse ppd estimates.

I'm going to give you a process for you to test to see if this helps. It will not be perfect, you will get some very bad performing WU's occasionally but you may end up with many more of those small SMP WU's you so desire

You will need to be in expert mode (upper right drop down box of FAHControl). Click configure, go to the slot tab and Remove the SMP slot and make exactly one uniprocessor slot. Next, go to the expert tab and click on the add extra core option (Bottom right) where you will add -forceasm then click OK, then again click on the add extra core option and in the text box add -np 4 then click OK and click save, finally restart the client.

Uniprocessor WU's are universally smaller with longer deadlines than typical SMP WU's. There are two types of uniprocessor cores FAHCore_78 and FAHCore_A4. The FAHCore_78's are older cores and are true uniprocessor and since you only have one slot when you get one, you'll get really bad PPD till it passes. But those are older projects and are not as common as they used to be. However FAHCore_A4 can operate as an SMP slot as well as uniprocessor. So what I just did is tell using the extra core option "-np 4" to use all four cores in your single uniprocessor slot. The "-forceasm" only applies to projects for FahCore_78 and makes it run faster. The 4 on "-np 4" refers to how many threads to run, never exceed the number of cores reported by your OS and you may want to use less if there is any significant non-folding CPU usage.

With a little luck you will get far more small FAHCore_A4 WU's and very few uniprocessor FAHCore_78 and your total long-term PPD will be higher than a traditional SMP slot. Just as a warning: you may end up with lots of FAHCore_78's rather than FAHCore_A4's and then you'll be really unhappy with the results. You just don't have any control over what gets sent your way, so you'll have to accept the 78's as well as the A4's.

The other alternative is not to do this test and accept whatever smp WU's come your way as your maximize PPD.

[pipnina]

i will try this when my current wu is finished
thanks for the idea