Is there a "get the most points guide"?

[jebo_4jc]

I'd like to know how to get the most points on a certain CPU, how much memory is required, and so forth.

 

[roftranspo]

I'd like to know how to get the most points on a certain CPU, how much memory is required, and so forth.

make sure you have the -advmethods flag on...that way stanford will (if they have any) send you the bigger and/or newer protiens worth more points. i think there are other flags for this too but -advmethods is the one i know of.

FOR THE HORDE

 

[tsuehpsyde]

I'd like to know how to get the most points on a certain CPU, how much memory is required, and so forth.

Easy Answer: borg alot of machines. Watch your points skyrocket!

 

[santaliqueur]

the advmethods flag caused my machine to put out less points per day. i ran it for a month, and my points went down, took it off, and they shot back up.

 

[RPhArrow]

I'd like to know how to get the most points on a certain CPU, how much memory is required, and so forth.

Ahh.... THIS is what makes it fun for anybody that likes to tweak . Before giving any advice I would remind all that the best settings for a given machine depend on what WU the Assignment Servers are passing out. Tinkers, Gromacs, gbGromacs, Gromacs33, Ambers, QMD's all perform relatively better or worse depending on your type of CPU, amount of ram available, OS which leads to what client you run, and finally which flags you set (which can help steer you to certain types of WU) . So getting the max ppd on a given machine or set of machines is definitely an art, not a list of instructions ! And as the type of available WU changes , so must you .

Here are a few of my own practices.... note that since my ppd is way down , I'm actively looking for improvements

P4's like QMD's (not currently available), & Gromacs.. They fare poorly against AMD in Tinkers & Ambers (I think..) . I don't know how P4 vs AMD64 works out on gbGromacs & Gromacs33. At this point, for a single core cpu, 1MB ram is plenty & often overkill, but now we get to flags.....Look at the Stanford projects page to see how big the proteins are for various projects/WU, and what their point values are. The small proteins will take less ram. If you want the bigger proteins (usually leading to bigger points) you will probably have to set the big packet flag in the console version of whatever client you run.

A perennial issue is whether to run 1 or 2 clients on a P4 which has HyperThreading available. Stanford wants HT disabled & a single instance of FAH running. Best points come from enabling HT & running 2 instances that don't create an overdemand for ram. (No problem if you have 2GB ram ! ) That's where the configuration of the newer clients shines because they allow you to allocate at maximum amount of ram to each instance. Dual core owners walk away the the points at this stage if they have sufficient ram !

edit: forgot to mention that even with lots of ram, it's easily possible to max out the memory controller with 2 WU both using big packet flag & advanced methods...

Hell, that's enough hot air for tonight Many old threads exist on the forum to research these issues, but if you have good questions, asking always helps out other people too !

RphArrow --\\\---------------------------->

 

[ChingChang]

Hmm if a CPU has a lot of cache, will the WU finish a noticable amount faster if all the computer is doing is working on that one protein? Faster than a computer with less cache?
I just randomly thought of that for some reason...

 

[RPhArrow]

Hmm if a CPU has a lot of cache, will the WU finish a noticable amount faster if all the computer is doing is working on that one protein? Faster than a computer with less cache?
I just randomly thought of that for some reason...

From what I've seen, read , & tried, cpu cache doesn't seem to make much difference.... cpu speed seems to make much more difference. Refering to P4's only, my 530, 531, and 530J all seem to do as well or better than my 630 with twice as much L2 cache.

RPhArrow --\\\------------------------->

 

[jebo_4jc]

Thanks for the tips.