Fahcore_17 lags computer

[monkeyclaw]

Hello! I have searched this forum (Problems with AMD/ATI drivers) and I couldn't find any threads on this, so I apologize if this has been discussed.

I have a Radeon HD 6970 with 13.9 release drivers. Whenever a new GPU core, Fahcore_17 that is still in advanced testing before it is officially released I believe, is running, my entire computer environment is just slightly laggy. As I type these words, I can see the lag because it makes the cursor hang and stutter. It is most noticeable if I am watching any type of video, on the internet or local, because it just turns it out all choppy. Not a consistent choppiness, just a stutter every 1 to 3 seconds.

I remember when I first tried Fahcore17 with 13.4 drivers, and before when it was just in beta, I did not notice so much screen lag/stutter.

Do certain drivers help? Is there anything I can do to mitigate the screen lag while the GPU slot is folding?

Thanks in advance!

[7im]

Unlike CPUs that were designed to prioritize workloads, GPUs are first come first served and don't share well.

Driver versions can change this slightly, but not entirely and are not the answer. Depending on the browser, try disabling GPU acceleration in the settings. But that may not help with videos.

Alternately, set the GPU from full to medium power, and it will only fold when the system is idle. It will only activate when either the fah screensaver is set to engage, or if no screensaver, when the monitor powers down in to saver mode.

[Jesse_V]

Certain drivers can help, but the problem will still be there. There's not much F@h can do about this, it's something inherit to the design of GPUs. Systems that use GPUs assume that the GPU is at idle most of the time, and when it is used, only one application is using it. A video game can be safely written under this assumption: the renderer assumes that it has exclusive access to the GPU, so it can just send it all the work it wants. The renderer for the screen is the same way. Unfortunately, this assumption breaks down under GPGPU, where the GPU is used for general-purpose tasks (such as folding) and then the lag makes it obvious that GPUs don't have the advanced scheduling algorithms and interrupt instructions that CPUs do. It may not be worth it for a GPU manufacturer to fix this, and it may not even be possible. There is little F@h can do about it though. The best solution we've come up with is to introduce the Folding Power slider, which allows you to dial back the GPU utilization such that folding will only run on your GPU when your system is idle. Then F@h will have exclusive access to the GPU, and you won't have any lag problems. As soon as you use your computer again, folding becomes paused. It's a bit of a tradeoff between utilization/PPD and lag.

Here are some links that you might find helpful:
http://folding.stanford.edu/home/faq/fa ... ion/#ntoc3
http://folding.stanford.edu/home/faq/fa ... ate/#ntoc5

[monkeyclaw]

Thanks for the very detailed and informative replies!

I have been folding for about 7 years now, and I have been an ATi (or AMD) user for that entire time, so I am very glad to see these new gpu units come around that seem to even the playing field. The new V7 client that was introduced a long while back is very nice too, much easier to use the slider than click home-made service commands shortcuts to stop and start the GPU Client.

It is unfortunate that with the new core17 comes the price of it actually demanding ALL of the attention of the GPU! But from what you are saying, it seems like this is more of the inherent nature of GPUs and not much that can be solved through code or from F@H's side. This is a bummer, as I have been a 24/7 folder for many years now and I hate to turn things off because of lag, and am used to only having to do this during serious gaming.

So thanks for the info! My only question that remains is: why did the old GPU units for AMD/ATi not lag out the screen at all? Was it because of just how inefficient it was?

Thanks again!!

[Jesse_V]

So thanks for the info! My only question that remains is: why did the old GPU units for AMD/ATi not lag out the screen at all? Was it because of just how inefficient it was?

The old workunits used older FahCores which used different GPGPU APIs. FahCore 17 uses OpenCL on both Nvidia and AMD/ATi GPUs. Over time we'll see that core become the dominating core for GPU work, mainly because it has a cleaner and more universal codebase, which makes it easier for the researchers to maintain and use. Different APIs use the GPU in different ways. The lag will depend on the workunit, the driver, and the GPU, so it's kind of hard to say for sure.

[mdk777]

Was it because of just how inefficient it was?

yes.

while annoying...simply moving the folding control slider to free up the GPU when you need it is the only solution now.

[PantherX]

If you want your GPU to automatically pause when using your system while the CPU continues to fold in the background, you can do it with the following configuration:

Set the power level to full to ensure that the CPU folds 24/7
Open Advanced Control (AKA FAHControl) -> Configure -> Slots Tab -> GPU -> Under the Extra slot Options, enter the following:
Name: idle
Value: true

With this set-up, the GPU will only fold when the system is idle.

[monkeyclaw]

Okay cool. One last question: Is there any way to make sure it makes a checkpoint before you pause/stop a work unit? Right now it seems like the only way is to decrease the slider that tells it to make a checkpoint, but it says it decreases performance... by how much?

I have noticed that when I paused before, it sometimes goes back a percent or two, so I'm pretty sure it doesn't make a checkpoint automatically when paused. Is there a way to make a checkpoint manually?

[Joe_H]

Unfortunately there is no way to initiate a checkpoint manually. When a checkpoint is taken depends on which folding core is being used. Some cores write a checkpoint at a predetermined time interval that can be set in the client. Others do so on completing a frame, 1%, and in the case of Core_17 every 2%.

[monkeyclaw]

Okay, good to know. Thanks everyone for the answers to my many questions =)