Folding Forum

I am new to folding and want to maximise my contribution.
I can leave my computer running overnight and during the day and as such
would like it to crunch at maximum when I am not using it.
I would like to be able to switch off the FAH when I am using it.

Can someone suggest the optimum clients/setup for this?

I have so far installed the GPU client which is running at 99% and
a CPU client which is only using about 12% of available CPU capacity.

I experimented with overclocking the CPU up to 3.8 GHZ from the stock 3GHZ before
but stopped this when the computer did not seem to offer much improvement in speed.
Would overclocking the CPU again benefit folding much? i.e. another 800mhz CPU speed?

I have the following System:
CPU: Intel Core i7 950 (quad core + HT )
GPU: nVidia GTX 460
Mobo: Asus Rampage Formula 3(nice overclocking features)
OS: Windows 7 ultimate 64bit
Internet: 4Mbs, I am willing to download/upload "large" files.

Regards
Will

SMP client v6.34 + GPU client v6.41 would be the way to go. However, SMP v6.34 is available as console version only. If that's a problem, I'd suggest v7 client. It's still in beta, but it's a very stable beta. It's a complete rewrite from v6, and can do both SMP and GPU in one client. Since easy pausing and restarting is a priority to you, I have a hunch that you'd prefer it over two v6 clients.

The ~12% CPU utilization is either because you're running one single-threaded uniprocessor client or you've forgotten to add -smp flag to the SMP client. Since it isn't a fully dedicated folding machine but could fold 24/7 nevertheless, I'd suggest -smp 4 flag (== running 4 FAH threads). This would increase the CPU utilization to "only" about 50%. Thing is, folding is extremely FPU (Floating Point Unit) intensive, and your CPU has only 4 of them anyway, so even 4-threaded SMP runs pretty close to maximum performance, despite the misleading 50% reading. That would still leave you 4 free threads on the CPU to run your other stuff. If that happens to be mostly non-FPU code, your stuff and FAH will run more or less concurrently, and the performance loss for both is reasonably small.

Also, FAH runs at idle priority, so whenever your other processes need CPU time and there aren't any free CPU threads, the OS immediately suspends one or more FAH threads until higher priority processes are done. Pre-emptive multitasking, that. In most cases you'd be hard pressed to notice SMP client running in the background, even with -smp 8. One example to the contrary is games, pausing SMP might make a difference there. Long story short, try running SMP with 4 threads 24/7 first and see if you can live with that. If so and there still is room to spare, we can discuss further optimizations.

GPU folding is a different story. GPUs aren't particularly good at multitasking, after all they were originally designed for a single purpose: putting graphics on your screen. You've probably noticed some lag when you use your computer while your GPU is folding, so you'll almost certainly want to pause the GPU client (or slot in v7) while you're at your computer, and restart it when you're done.

BTW, I'd recommend getting a passkey. Makes no difference to the science whether you have it set up or not, but it's good for you. For example, it's one of the prerequisites for earning SMP bonus points.

As for overclocking, FAH really needs to have the WUs (Work Units) returned as fast as possible. This is reflected in the nonlinear SMP bonus points system. The faster the return, the bigger the increase in bonus points. However, this isn't an OC forum. Suffice to say, stability is still number one priority for a scientific project. Marginally stable computers and marginally reliable results aren't of much use. Folding is heavy duty job for your CPU & GPU, so gaming/Prime95/whatever stable doesn't necessarily mean FAH stable. The tools link at the forum banner (top of the page) lists some CPU & GPU & memory stability testing utilities under "Testing Your System Stability". If you decide to OC, it is prudent to run a mix of tests for a "reasonable" period in order to ensure stability and sufficient cooling before going on with folding. Probably a good idea to run some even if don't OC.

Hope this helps, and welcome to the forum, willsan.

V7, while still in beta, should have no problems with a simple setup like that. I would advice you to give it a try and enjoy the future.

As Napoleon said, get a passkey. And yes, overclocking gains can be substantial, but loosing work units due to not being able to finish them offsets this. So, small steps and clock back at the slightest sign of instability. Use common sense and you'll get to the best point.

willsan wrote:I am new to folding and want to maximise my contribution.

Welcome to FAH and to the folding support forum, willsan.

Napoleon has given many excellent suggestions, including the one I've quoted below. My experience has been different, though, so I'll offer an alternative. Use -smp 8 (or smp:8), especially since you have an nvidia gpu. It provides a more "comfortable" reading of cpu utilization of 100% and though the actual improvement in throughput is going to be only about 15%. Eight threads do a better job of keeping the FPUs busy, and I never notice a lag because preemptive multitasking has always worked exceedingly well for me. If there's a noticeable desktop lag, it's almost certainly from the GPU. (For games, you will need decide what needs to be paused because I don't use my rig for gaming.) When using smp:8, you'll need to adjust the priority of the gpu slot/client.

Napoleon wrote:The ~12% CPU utilization is either because you're running one single-threaded uniprocessor client or you've forgotten to add -smp flag to the SMP client. Since it isn't a fully dedicated folding machine but could fold 24/7 nevertheless, I'd suggest -smp 4 flag (== running 4 FAH threads). This would increase the CPU utilization to "only" about 50%. Thing is, folding is extremely FPU (Floating Point Unit) intensive, and your CPU has only 4 of them anyway, so even 4-threaded SMP runs pretty close to maximum performance, despite the misleading 50% reading. That would still leave you 4 free threads on the CPU to run your other stuff. If that happens to be mostly non-FPU code, your stuff and FAH will run more or less concurrently, and the performance loss for both is reasonably small.

By the way, the original question of maximizing contributions is mostly applicable to V6. Its default settings ae designed to minimize FAH's impact on desktop activities, not to maximize contributions. When V7 becomes the mainstream client, I doubt many people will be asking this question. It does a reasonably good job of maximizing contributions on a lot of systems with default settings though there are still a few worthwhile tweaks, like gpu priority. When FAH was originally designed (ten years ago) computers were not very powerful and minimizing lag was really important. V7 is the first major rewrite and it's aimed at more modern systems which generally have excellent capabilities to simultaneously keep both FAH contributions and desktop responsiveness in reasonable ranges (except for games).

Thanks for the replies.
I have downloaded the v7 client and it seems to be working.
The CPU reads 100% usage across all 8 "cores" ( I have a quad core i7950 with hyperthreading enabled).
The GPU reads 99% usage.

Now I will experiment with getting a stable overclock for the system - I have this fancy
MOBO capable of fantastic overclocking which I have not really used to its full capacity yet.

Regards
Will

You should know that some of the more common OC stress tests like Prime95 may not test for FAH stability; FAH primarily uses the FPU, while some stress testers mainly use the integer unit and don't hit the FPU very hard. Many people have found that Prime95 stable does not mean FAH stable. StressCPU v2, available throughhere, runs the same GROMACS code that FAH uses, and is a much better indicator if your OC is folding stable.

This recent post by Bruce may be relevant: viewtopic.php?f=59&t=20092&p=200689#p200689