Folding Forum

Permit me to introduce myself. I am Gandalf the White or Gandalf to my friends . I have been crunching with WCG for a couple of years for the PCHF Team. The word has been bounced around that the WCG > hcc1 GPU tasking will be phased out in the next couple months. I have been advised that this might be a good home for my GPUs to work. I plan to leave my CPU tasking with WCG, but it is my understanding that I can redirect my GPUs to do work for Folding@Home. In the event that this is correct, I will be snooping around here to get a feel for F@H.

I welcome any communications from the local folks here. I have not yet decided on a team to join. I doubt PCHF will be very concerned about GPU tasking.

Respectfully,
Gandalf

It's wonderful to see you Gandalf!

Welcome to the forum. You are correct, Folding@home can use your GPUs, and there's a very significant scientific speedup in using them. The High Performance and GPU FAQs explain the benefits of GPUs for FAH. Setting them up in Windows is relatively simple, the installer should take care of everything for you. You'll want to first download the V7 software from the F@h homepage. During installation, just choose that you only want your GPU to be used, and things should start up automatically. The installation guides may be helpful: http://folding.stanford.edu/English/Guide#ntoc6

If you'd like to learn more about Folding@home, http://folding.stanford.edu/English/Learn is a great starting spot for all the resources. In particular, I'd recommend the Science FAQ, the Diseases Studied FAQ, as well as the Folding@home article on Wikipedia for more details.

Hello and thank you, Jesse_V. I guess I now have my weekends' homework assignment.
Do you have a F@H team you can recommend me joining. Here is a look at where I am at now with PCHF...
http://www.pchelpforum.com/xf/threads/p ... 13.149317/
With BOINC, I'm known as Gandalf the Grey.

No one here can directly answer your question as to a good team to join. We aim to be team-agnostic and to help all equally. To pick a few to suggest would be breaking that trust. If you want you can start out with the the F@H stats (Links on the top of the page) to get you a list of active teams. From there find one that matches an interest you have.

Question about how F@A uses a computers' capabilities. An Intel quad-processor has eight threads.
Does the F@A tasking configuration set one task per thread?
Can an ATI GPU configuration be set to run 10 tasks per ATI GPU with the support of two CPU threads (0.2C + 0.1ATI)?

Thanks for the info.
Gandalf

P.S. Modertor - Sorry about the Team related question. Didn't intend to break the rules. Thanks the for suggestion on how to proceed with finding a team.

SMP uses one thread per core; The GPU cores use one thread per GPU. ATI will use a full core of CPU in its one thread unless using Core_17 which is currently in beta testing. Nvidia will only use a fraction of a CPU core in its thread unless it is running Core_17 where it then uses a full CPU core.

It is better to run one slot per GPU then multiples. If you run multiple ATI WU's on the same GPU then each one will use a full CPU core (except for Core_17). If you are only using a fraction of the GPU on that one CPU core, you have a less than optimum driver. The last good ATI folding driver was 12.8 (and it is an outright pain to revert back from a newer driver), unless you are running the beta core Core_17 which seems to work better with more recent driver versions.

For folding running one WU fast is much better than several WU's running slow. Future WU's are generated from previous returned WU's so anything delaying a single WU's from returning quickly delays the entire project. Thus the idea of running 10 WU's on a single GPU does not work well because it slows down the calculation of ten WU's. However, it is possible to configure folding to do so.

Core_17 is being beta tested and seems to work well for ATI. If you are interested in running it then you can apply to be a Beta tester but you must accept all the duties and responsibilities therein.

P5-133XL - Thank you for the information. For clarity purposes, this is an image of how I am currently running my computers for WCG...

http://i702.photobucket.com/albums/ww27 ... 97cff8.png

It might take 15 minutes to run a GPU task, while it might take hours to run a CPU task.
I am running a Pre-Release version of BOINC's program.

Welcome to foldingforum.org, Gandalf.

FAH's GPU work assignments take about a day (depending both on the GPU and the particular project). SMP projects generally take more than a day -- for slow CPUs, they might take a week.

One critical difference between many BOINC projects and FAH is the deadlines. For Seti and other projects, the deadline is a guideline, not a requirement. For FAH, the points drop off radically if you pass the Timeout (Preferred Deadline) and a WU that exceeds the Final Deadline is discarded with zero credit. Moreover, bonus points increase for every minute earlier you finish them. Speed is more important than quantity.

Gandalf wrote:Question about how F@A uses a computers' capabilities. An Intel quad-processor has eight threads.
Does the F@A tasking configuration set one task per thread?

You are in control of how many threads are allocated to the CPU assignments. If you choose 4 threads, the operating system will choose which ones to use and recent versions of Windows do a pretty good job of assigning affinity. If you choose 8 threads, performance will increase by maybe 15% or 20% but certainly not double since most of FAH's CPU computations are limited by the 4 FPUs. Many people choose CPUs:6 or CPUs:8 for the type of system you're describing since the computations that support the GPUs use little CPU and (almost?) no FPU processing.

Hi bruce. Thanks for the info. One thing, FPU? That's new to me. What does it mean? F... Processing Unit?

All modern CPUs contain an integrated Math Coproocessr rather than an external one like the one shown in the beginning of the wikipedia article.

There are several different types of instructions. Integer instructions, Floating Point instructions, SIMD instructions, etc. They are often processed by portions of the CPU that are physically in different places. In the best of cases, different threads can be processed in parallel by separate portions of the CPU and you can get more total work completed than if all of the threads depend on circuitry that appears 4 times rather than circuitry that appears 8 times in your i7 or by running instructions that use 4 of this and 4 of that.

Don't worry about it if you don't understand it. You don't really NEED to understand it. It's just an explanation for why the recommended numbers work for most people.

It's the initials I didn't get. Back in the day, I used test those chips, when I was troubleshooting PCBs.
bruce, I'm a wizard. Remember? Understanding is not the problem.
I'm just to lazy to look FPU up in my Wizard Knows All book.