TR Forums

I recently became interested in the F@H project and decided to install the GUI client at home (for team #2630, naturally ), followed by the console/service client on my test machine at work (I figure it sits there unused for half of the day anyway, might as well put it to work...). The test machine has a 3.2 ghz P4 with hyperthreading, and when I check the task manager, only one thread is maxed out when the PC idles, the other is not utilized. Is there any way to utilize both threads? Perhaps installing two clients?

zer0 wrote:

I recently became interested in the F@H project and decided to install the GUI client at home (for team #630, naturally )

Welcome to the team.

zer0 wrote:

followed by the console/service client on my test machine at work (I figure it sits there unused for half of the day anyway, might as well put it to work...).

Do check with your company's policies about installing your own software and running the computer full tilt all the time. They may not like it and we don't want you to get into trouble.

zer0 wrote:

The test machine has a 3.2 ghz P4 with hyperthreading, and when I check the task manager, only one thread is maxed out when the PC idles, the other is not utilized. Is there any way to utilize both threads? Perhaps installing two clients?

Running 2x single core clients would max it out. Both clients will end up sending results slower but you will get more points. Stanford wants you to send results faster though. So it is up to you.

Flying Fox wrote:

Welcome to the team.

Thanks

Flying Fox wrote:

Do check with your company's policies about installing your own software and running the computer full tilt all the time. They may not like it and we don't want you to get into trouble.

I'm not too worried, the only people that would either monitor that or care work with me in my office, and they wouldn't mind. Afterall, it is just a test machine. Mostly I use it to test HDDs and RAM. When I'm off doing other things it just sits there, so I don't think there's any harm in letting do some good when I'm not at my workbench.

Flying Fox wrote:

Running 2x single core clients would max it out. Both clients will end up sending results slower but you will get more points. Stanford wants you to send results faster though. So it is up to you.

I'm not too concerned with the points at the moment. This is mostly a proof of concept...

(FlyingFox: See PM)

If it's hyperthreading, you're better off running a single thread IIRC.

One thread is good enough, I was just curious though about any way to optimize it for multithreading.

BTW: When I install the client as a service, how do I view the progress?

Try using FAHmon.

Nope. That didn't do it...

zer0 wrote:

ChronoReverse wrote:

Try using FAHmon.

Nope. That didn't do it...

You have to set up FAHmon to look into the working folder of the FAH program. Then it can do the monitoring. You can try FahSpy too.

Sooo.... How do I do that?

zer0 wrote:

Sooo.... How do I do that?

http://trac.fahmon.net/wiki/AddNewClient (screenshots here)

I too have a Pentium4 HT folding. Before, I was using the SMP client (it can make deadlines), but when I read that Stanford wanted the results faster, I switched to the single client and now the GPU2 client.

Supposedly, the maximum benefit you can get from a HT CPU is if you are running two clients, both with work units that use different parts of the processor (FPU, INT), and that can crank out a maximum of 20% more PPD. Running two clients, in most cases, can even make them slower than 50% of the normal.

Either way you choose to do it, welcome to the team!

(By the way, you will see your PPD increase by 4x with the addition of a 2600PRO, the fastest GPU that a P4 can feed with these work units).

Okay I have a new question


How should I fold with a DUAL PROCESSOR box where both processors have hyperthreading

Would SMP client cope with the hyperthreading well? or could i bind the HT to the two main cores and run normal clients on the HT-cores

I'd just run a single SMP client and call it a day. The client should use enough of the resources that there really isn't much wasted space that adding an additional client would take advantage of.

The SMP client actually runs 4 threads. So a single copy should keep a dual-core hyperthreaded system (or even a quad core system) busy.

IIRC it was in Stanford's FAQ where they said to enable the hyperthreading on dual core CPUs if possible. There were supposed to be some minimal improvement gains, but the only chips I can think of that do that (Pentium 4 XE comes to mind).