Optimizing throughput for the contest

I'm new to FAH and have gotten the SMP CPU client and the Windows GPU client running on my Linux box. Now I'm wondering about some of the options that the FAH config asks about and if they will affect throughput.

There is a choice for memory:

Is bigger better? I've gone with big and the memory footprint seems to be pretty small. (Smaller than X and the chromium browser.)

Then there are the advanced options:

Which, if any, of those will affect the points I earn and how should they be set?

If the answers to the above questions differ between Windows, Linux and Linux as VM guest on Windows, please specify.

As for Linux, I'm running Ubuntu 10.04 64 bit desktop. While working with Rosetta/CPU and SETI/GPU I found the best balance of performance was gained by running a real time (RT) kernel. I also tried the CK kernel but without further tuning it seemed not to produce as well for those tasks as the RT kernel. I also ran 2 instances of SETI which proved beneficial. The RT kernel kept the GPU well fed without having to sacrifice a CPU core to do so. The price paid was an approximate 15% reduction in CPU throughput due to the extra overhead in the RT kernel. I plan to do some testing of different kernels with FAH to see if there is anything better than the RT kernel. However, since my GPU (standard clocked GTX 460) crunches about 4-5 times as fast as my CPU (overclocked Phenom II x4 820) it seems like a small reduction in CPU performance to keep the GPU at 95% is reasonable.

Feel free to share any knowledge you have of legal ways to maximize points (or point to the appropriate resource.)

thanks,
hank

All SMP WUs are big, the setting has no effect.

SMP priority should be idle, gpu priority low, so that the smp client doesn't interfere with the more productive gpu client. No need for a RT kernel. The SMP client takes somewhat less of a hit from the gpu client with the ck kernel.

It may prove faster on to disable the cpu lock on the gpu client so that it is spread across all cores and doesn't slow one smp thread a lot,forcing the others to wait, but rather slows them all down a little bit.

The -advmethods flag affects WU assignment, which affects ppd. It could give you more or less ppd depending on the assignment. This flag can be used to avoid particularly slow WUs or increase your chances of getting particularly fast ones, depending on WU assignment with and without the flag. It's very difficult for someone folding on one machine to determine what's best. You have to poll someone running a lot of instances to see what they think.

Add the -verbosity 9 flag in extra parameters.

The answers are the same for Linux and Windows. If you aren't willing to flash your gpu bios to set your OC, you might find smp + gpu in Windows results in more ppd with an OC on the gpu.

Thanks for the info ChasR!