Folding Forum

Hello All,

Installation went fine... I want to set less than 100% to both CPU and GPU to keep heat down and to allow me to use my PC for other things. I figured out how to reduce the # of threads for the CPU, but I have two questions;

1) How can I commit less than 100% of my GPU to this effort?

2) Under Configure / Advanced there is a Percent CPU usage setting. However, this does not seem to affect anything at all. What is it for?

Thanks!
R

One more question....

Since I am running say only 30-50% of my CPU on folding (I have an i7-3930k), will it take too long for these work units to complete? And if so, is it a waste of time?

That is, does the managing system assume that the work units are lost and assign them to others so that when my data is submitted it is discarded?

Thanks!
R

Welcome to the Folding Forum, Rhyalus.

1. I'm not sure how to reduce the GPU %. I think it just runs all out at 100%.

2. I don't think the CPU % setting works atm and the suggestion here has been to reduce the # of cores/threads, which you have done. So you're good there.

3. The work should complete in time with 30-50% of the CPU usage be doing useful science. SMP Work units do get a Quick Return Bonus, so the faster you return the the work, the more points you get. Once you get a few % done you'll be able to get a decent time to completion. If you check the project ID here on this page you can see the preferred deadline and the final deadline. If you choose to contribute just 30-50% of your CPU then that's all good too.

http://fah-web.stanford.edu/psummary.html

There is no current way to use less than 100% of the GPU other than to manually use pause/unpause!

The %CPU is only useful for old uniprocessor cores. If you need to decrease the use of your CPU to less than 100% (For SMP) what you do is change the number threads that the SMP slot uses.

If the reason you want to use less CPU is because it interferes with your useage then my suggestion is keep it at 100% usage. It is folding on the GPU slot that causes problems rather than the SMP. The SMP slot runs at a very low priority and suspends itself whenever there is a higher priority application that runs thus rarely interferes with your normal useage.

The GPU has no such mechanism and as such typically causes lag especially when another application wants to use the GPU like watching videos. If you are noticing issues, most people simply pause and unpause the GPU slot as needed and that solves most issues.

If you change your view setting from Novice to Advanced, you'll see much more information about whichever WU is highlighted in the Work Queue, including when it was downloaded, the estimated completion time, and the deadlines. Timeout is the first deadline, and the point at which the servers send out that WU to someone else under the assumption that you're not going to return it. You can still return it for base credit by the Expiration, even if that other person returns it on time, but you won't get any bonus points for it.

In case you hadn't gotten one, it's strongly recommended to get and enter a passkey. They're required to earn bonus points, which are currently only garnered from most uniprocessor and all SMP work, but which are supposed to be implemented for GPU work in the near future. You have to turn in 10 bonus-eligible WUs before the timeout and maintain an 80% success rate to earn bonus points.

Thanks folks... I am learning a lot.

The reason I want to scale the speed back is not so much that it interferes with my other uses. I just don't want my system running all day at maximum heat generation, etc. This is my personal PC; while I am willing to commit unused CPU and GPU cycles, it is not at the expense of loads of heat, electricity and reduction of component life.

So if the developers are watching, please put in some options to reduce the impact on the GPU...

Casual helpers like myself might prefer not to tax the system to maximum heat load.

Happy Folding!

R

If your goal is to minimize heat and electricity, may I suggest that you drop your CPU voltage as low as it will operate and underclock your CPU rather than diminishing the number of SMP cores. Also, GPU folding typically uses a lot more heat and electricity than SMP folding, on modern machines, so not GPU folding will save more.

Rhyalus wrote:So if the developers are watching, please put in some options to reduce the impact on the GPU...

This enhancement has been requested many times. I know the developers are aware of the issues. Apparently it's a bit of a challenge to manage GPU utilization. In "normal" situations, the GPU was designed to process your video a fast as possible and the people who design GPUs and their drivers see no reason to ever limit the performance of a GPU.

GPGPU programming (including FAH) is a later add-on use and even if FAH figures out a way to limit how much work a GPU can do, it won't limit processing requests coming from other programs.

Underclocking or pausing should reduce the heat.

Rhyalus wrote:So if the developers are watching, please put in some options to reduce the impact on the GPU...

Yes, the developers are watching as this request has come up before. They may not be able to throttle the GPU right away, but are aware of the issue. In the meantime, they are in the process of coming up with Performance Levels to be used in future versions of the V7 client. GPU options may include pausing the GPU processing when the PC is not idle. So I would recommend watching for future releases to see how these get implemented.

Yes would be very nice to reduce GPU just enough to get rid of the video freezes, lag, mouse freeze etc.

road-runner wrote:Yes would be very nice to reduce GPU just enough to get rid of the video freezes, lag, mouse freeze etc.

Important words:

Level 3
Medium - All cores 90% (not all cores support throttling), GPU folding when it can be throttled.

The real question is when will it be possible to throttle GPU folding. Probably the GeForce 700 or 800 Series

Back in April Dr. Pande said he was thinking about adding a screensaver that would run the GPU only when the computer is idle. A number of donors, including myself, realized that it could really alleviate this problem and showed their support for the idea. There's also been some related Trac tickets, such as https://fah-web.stanford.edu/projects/F ... ticket/950

Hello - just returned and I downloaded the new version... so this was a small incremental change. Now it is easy to manually stop the GPU by sliding to medium, but this still misses the mark, at least for me.

I am just looking to control heat.

So wouldn't some sort of "duty cycle" control be easy to implement? For every three minutes, a 50% duty cycle would mean 1.5 minutes on and 1.5 minutes off... (just throwing a guess for a reasonable cycle).

Unless driving the card up and down was worse for the card in general?

R

The best I can suggest is to significantly under-clock the GPU while increasing fan speed to control heat. Depending on how much you are into HW, you can also clean the fans with compressed air, replace the thermal compound, or even upgrade the fan. If it is a notebook, lifting the notebook off the surface it is resting on may also significantly improve airflow.

And yes there are potential issues with constantly powering up and shutting down at any rate that could cause thermal expansion and contraction. See: all nvidia g84 g86s are bad as an example

Rhyalus wrote:I am just looking to control heat.

Pausing the GPU when the computer is being used has the goal of eliminating screen-lag which can become troublesome, depending on how the drivers manage the GPU's workload. It was not intended to manage heat, although heat is obviously not going to be a problem when the GPU is off.

Duty cycle management has been tried and the concept has been abandoned. It demonstrated a number of shortcomings and is no longer considered part of the plan for FahCores. Manufacturing and selling a GPU which has insufficient cooling is either a failure of the hardware designer to meet their responsibilities or the problem is caused by the person who insists on overclocking it beyond whatever the cooling subsystem can manage. Even if FAH were to reintroduce duty-cycle management, that doesn't prevent some other application from filling in the processing gaps with additional work, again causing overheating.

As has already been suggested, you have the options of underclocking / undervolting / increasing the fan profile / adding additional cooling equipment, etc.