Page 1 of 1

v7 GPU Client types

Posted: Thu Jan 10, 2013 1:36 am
by lundrog
When folding with V7 and GPU's, do any of the client type command's apply to the GPU slots?


"
-advmethods --client-type=advanced
-bigbeta --client-type=bigbeta
-bigadv --client-type=bigadv
-betateam --client-type=beta
"

Taken from https://fah-web.stanford.edu/projects/F ... ncesV6ToV7

If so, what are the differences?

Thanks


Roger Lund

Re: v7 GPU Client types

Posted: Thu Jan 10, 2013 5:11 am
by bruce
Each new project and each revision of a FahCore is tested before it is distributed. The first stage, of course is testing in the Pande Lab. Once a project or a core becomes a release candidate, it is distributed to a group of volunteers commonly called the Beta Team where a significant percentage of problems are found an reported. More information can be found here: viewtopic.php?p=102182 The types beta and bigbeta are supposed to be limited to those who volunteer to participate in the beta team. Once the project or core seems to be stable on a wider variety of equipment and operating systems, the project and/or core are moved to advanced or bigadv which are open to anyone who accepts a certain level of risk prior to the project and core being ready to release to the entire FAH community.

The "big" projects are restricted to machines with a high number of CPU cores, 16 being the minimum. See viewtopic.php?f=24&t=10697 and viewtopic.php?f=85&t=23245 and http://folding.typepad.com/news/2012/02 ... llout.html. Even on systems that report that they have 16 cores, the deadlines are often too short unless the system has a pretty good clock rate and has 16 unshared FPUs, not virtual cores (Intel) or Modules (AMD).

The sequence of Lab testing/beta testing/advanced testing/full FAH release applies ot all types of hardware, including GPUs.

F=

Re: v7 GPU Client types

Posted: Thu Jan 10, 2013 5:15 am
by P5-133XL
The client type determines a WU's preference type you are requesting.

First there is general release and that is specified by no client type. These are the most stable and tested projects.
Beta is the newest, least stable, least tested projects.
Advanced is intermediately tested and stable projects

All three of the above are applicible to all slots including GPU's.

Bigadv are SMP projects requiring a minimum of 16 CPU cores with deadlines to match.
Bigbeta is the beta version of bigadv.

You request the group of projects, but you are not guarenteed to get one. If the group you request does not have a WU applicible, then you'll get one from one of the other groups. bigbeta failure may get a bigadv; bigadv failure produces a standard SMP project; Beta failure will likely give an advanced project; and advanced failure will likely produce a general release project.

Re: v7 GPU Client types

Posted: Thu Jan 10, 2013 5:56 am
by lundrog
Interesting , thanks for the info.

I have two gtx 680 and I'm getting tiny work loads that only take two hours , doesn't seem it matter if turn it off and delete the work folder , remove the slot and change Client type get the same work back again....

Same if I delete work folder change client type in XML and re start client.

Re: v7 GPU Client types

Posted: Thu Jan 10, 2013 6:31 am
by art_l_j_PlanetAMD64
lundrog wrote:I have two gtx 680 and I'm getting tiny work loads that only take two hours , doesn't seem it matter if turn it off and delete the work folder , remove the slot and change Client type get the same work back again....

Same if I delete work folder change client type in XML and re start client.
Please do not delete Work Units you have received. This slows down the project because the servers will wait for the Timeout before re-issuing the WU to someone else. You can click 'Finish' so that the WU will be uploaded when it is completed and a new WU will not be downloaded. It is then that you can shutdown the client, make changes, and restart the client. It should never be necessary to delete the work folder, unless you are instructed to do so by someone at Stanford/Pande Group in order to solve a problem.

Re: v7 GPU Client types

Posted: Thu Jan 10, 2013 7:58 am
by Zagen30
As to the WUs you're processing taking only a few hours, that's not an issue. Different projects have different length WUs, for reasons I'd imagine only the researchers know. For the past few weeks many Nvidia cards were getting a steady diet of some relatively long, taxing WUs. The other day many of us saw a different set of projects, probably the ones you're seeing, crop up, and they're both shorter and somewhat less taxing. There's basically nothing you can do to control it, so just let the client work on whatever it's assigned.

Re: v7 GPU Client types

Posted: Thu Jan 10, 2013 11:31 am
by lundrog
Ok, wanted to make sure I did not have a problem .


Roger

Re: v7 GPU Client types

Posted: Thu Jan 10, 2013 9:45 pm
by bruce
lundrog wrote:Interesting , thanks for the info.

I have two gtx 680 and I'm getting tiny work loads that only take two hours , doesn't seem it matter if turn it off and delete the work folder , remove the slot and change Client type get the same work back again....

Same if I delete work folder change client type in XML and re start client.
As has already been said, it's bad for FAH when people delete things if there is any other possible choice because it causes significant delays in the scientific progress.

Perhaps you were confused about getting multiple WUs from the same project. There are often tens or hundreds of thousands of WUs in a single project. They will have different Run/Clone/Gen numbers and every one of them needs to be completed. Sometimes a single project is the priority and other times there are wider varieties of Projects being assigned.

Re: v7 GPU Client types

Posted: Sat Jan 12, 2013 6:16 pm
by lundrog
It was only a couple times, I thought it was stuck.