different folding cores

[lucifersam]

I dont suppose some kind person could shed some light on the different cores that folding uses, like which one offers the most credit for CPU time and uses memory etc
cheers

 

[BillytheImpaler]

There are a bunch of cores in use now. The most common is Gromacs which uses FahCore_78.exe. It uses between 10 and 100 MiB memory depending on the size of the WU. Double Gromacs are common now as well. They use FahCore_79.exe and use double precision floats. Memory useage is similar to Gromacs. Both of these are optimised so that they give pretty good cedit, especially Double Gromacs becasue they use SSE2. Gromacs33 is a very new core which uses FahCore_a0.exe. It is only available to Linux clients and, in my experience, has extremely poor performance. GBGromacs is a core that uses around 30-100 MiB memory and is FahCore_7a.exe. Amber core WUs are out there as well. Amber is FahCore_82.exe. It is faster, more recently written core that can do the same things as Tinker, FahCore_65.exe. Niether of these are optimised and Tinkers can run exceptionally slow. Athlon 64s usually do pretty well on these. They can use around 30 MiB memory. QMD core WUs, run on FahCore_96.exe are not available at the moment. They use a quantum method that involves solving Schrödinger's equation. They can use up to 350 MiB memory for the first few percent then they settle down into around 200 MiB. These have been on hiatus since January 2006 pending the analysis for the QMD WUs we've already crunched. These can give several hundred PPD but are only available to Intel Netburst based CPUs.

Does that do it?

 

[joeyjojo]

QMD core WUs, run on FahCore_96.exe are not available at the moment. They use a quantum method that involves solving Schrödinger's equation.

Whats this all about?

 

[BillytheImpaler]

http://folding.stanford.edu/QMD.html

What would you like to know about it? Ask and ye shall recieve.

 

[lucifersam]

cheers thanks very much. i was just wondering because i went from gbgromacs getting 226 credit taking 15 mins per step, to gromacs giving 192 and taking 27 mins per step

 

[rich99million]

cheers thanks very much. i was just wondering because i went from gbgromacs getting 226 credit taking 15 mins per step, to gromacs giving 192 and taking 27 mins per step

yep this is what makes folding interesting (and also a bit frustrating if you keep getting worse work than you could be getting) as different work will react differently to your computer than to the machine used for benchmarking

at least everyone is in the same boat so it evens out pretty well (except of course for QMDs only being available to Intel machines), you can always play around with the settings a bit and see if you get some more suited work from it

 

[lucifersam]

any idea when the QMDs will be back i got a netburst CPU and wouldnt mind one of these

 

[rich99million]

any idea when the QMDs will be back i got a netburst CPU and wouldnt mind one of these

no expected date yet - though sounds like there may be some internal trialing going on

there's a thread on the current status here although a lot of it is speculation and spam from people who are just after the bonus points they offer (the sort who'd probably delete their current WUs if QMDs were available and set multiple clients to cache them )

one post worth noting though is this:

Posted: Sun Mar 19, 2006 7:41 pm
Here's an update for those who are curious. I've looked over the first level analysis and I think the data looks interesting, but not what I expected. We're deciding whether we need to run the previous system with high level quantum mechanics (which would be the QMD's come back relatively soon, but perhaps with higher memory demands) or do we look at a different system.

We're debating the pros and cons of the variety of proposed plans right now.

and if you check the server status page and check the "memory" column you'll see that one server (171.64.122.135) is set to give work only to machines with 1500MB ram or more

I doubt they'd need quite that much, more likely it's to ensure there's plenty spare for the system or another clinet on the same machine. That server may even not be trialing QMDs - there was a monster beta WU a while ago which was so big that the finished file exceeded the max file size of the client so it couldn't send itself



/speculation

 

[lucifersam]

i didnt know you could cache im surprised they let people do this as i thought they prefered to get WUs back quickly. people like you mention kind of spoil the spirit i think. I got 2GB mem so would like a really big WU for the credit but wont get upset if i dont get one!

 

[BillytheImpaler]

i didn't know you could cache I'm surprised they let people do this as i thought they preferred to get WUs back quickly. people like you mention kind of spoil the spirit i think. I got 2GB mem so would like a really big WU for the credit but wont get upset if i dont get one!

You can cache WUs but only the deadlineless Tinker core WUs. Caching other work must be done manually and it's a big no-no because it slows down the project.

Some people wold love to see a WU us 1500 MiB memory whilst others would scream bloody murder, spam up the forums, and leave the project, just as they did with QMD using 350 MiB memory.

 

[lucifersam]

true ut from what i see the people that are intelligent enough to read forums, instruction etc can see there are ways to NOT get these WUs shame there are people like this as it harms the project in a way