Graphical Client: potential energy of current fold.

[patfla]

I read here:

http://folding.stanford.edu/English/FAQ-main#ntoc64

that the Graphical Client should show the potential energy of the current fold.

Code: Select all

The molecule drawn is the current atomic configuration ("fold") of the protein being simulated on your computer and the graph on the bottom shows the potential energy in femtosecond increments.

I don't see that my display shows this value?

http://img690.imageshack.us/img690/1923 ... 2run66.jpg

I'm running FahCore_15.exe

I assume (question?) that the potential energy is a dynamic value over the time of a single folding session (work unit). And that the potential energy starts from some sizable positive value and then drives towards zero?

I think of GPU computation as a massively-parallel vector processor. And a MPVP is most obviously useful in terms of (large) matrix operations. Is there a FAQ somewhere that explains how F@H works at this level? (sort of: how is protein folding translated into matrix operations - and what operations of course?).

[7im]

Please check which section of the Main FAQ you are reading...

Screen saver (Windows: version 4 and earlier; OSX: all versions)

The v6 viewer image you linked is clearly quite different from the old v4 client screen saver picture in the FAQ. I'd guess the potential energy graph got dropped somewhere in the last 2 major version updates.

As for energy, most projects start unfolded and fold in. A few start folded and try to unfold.

I haven't see an FAQ on the mathematics behind molecular dynamic simulations at Stanford, maybe someone else has. You might check http://www.gromacs.org. Most of the folding fahcores are based on their simulation software.

[ihaque]

http://onlinelibrary.wiley.com/doi/10.1 ... 9/abstract

[patfla]

thanx - pat

[patfla]

I could have bought the paper online and while it may seem more troublesome we're not far from the Berkeley campus and what I prefer to do is to save up a list of papers I want and then one day drive to Berkeley and print out free online versions of the papers at Doe library. (Berkeley's library, like many I would assume, has free access to most academic journals).

I'm almost through the paper. I have many questions so I'll limit myself in this post to one. The papers describes how the ATI architecture of the time (published Feb 2009 although written prior to that) doesn't have a scatter capability then the paper illustrates this by saying that this means that the ATI architecture couldn't handle an expression like:

Code: Select all

a[i] = x

I think what this means - in my programmer's terms - is that, at that time, the hardware architecture wouldn't fully support pointer arithmetic? (and this may have had something to do with poor support of integers in general)?

I assume that now, something like a yr-and-a-hlf later, this is mostly of historical interest? Meaning ATI hardware and software do fully support pointer arithmetic?

What I would note (since I have some knowledge of CUDA programming) is that optimized CUDA programming has to pay very close attention to hardware constraints. In particular the memory hierarchy - e.g. register, shared, local and global. I don't see this going away for some time.

These articles are a good introduction to CUDA programming:

http://www.drdobbs.com/architecture-and ... /207200659

Next I'd like to learn more about the Six Forces that characterize molecular dynamics. There's a sort of dialectic between the computer science on the one hand and the 'domain knowledge' (molecular biology) on the other.

[patfla]

Didn't get that answered but that's OK because I researched my question more online and found such things as the differing computing capabilities of a fragment program vs. a vertex program

http://http.developer.nvidia.com/GPUGem ... ter32.html

(it seems this was written by Ian Buck)

'Program' here may be something of a misnomer (implying software that is) because the concept in this case is very (GPU) hardware dependent.

Reading on, I was struck by this encoding (in the nonbonded interactions exclusion matrix input):

2 X 3 X 5 X 7

It's isn't explained (one is supposed to know that already) but I assume the values 2,3,5 and 7 are used because these are the smallest prime values (> 1). And this relates to the Fundamental Theory of Arithmetic (?).

I thought: "why don't they just use a bit-wise encoding" but then realized, oh, GPU hardware probably doesn't support bit-wise operations.

GPU programming is very dependent on the lowest level specificities of the hardware and this is a good reason why one would want to carefully watch the latest small developments with Nvidia and ATI. For that matter it moves both ways. The GPU computing market would be watching Nvidia and ATI and ATI and Nvidia would be watching the GPU computing market which is getting larger all the time. I was made aware of GPU (CUDA) programming by a friend who's a digital signal processor designer.

I would give up asking these kinds of questions at folding@home except that I've had a longstanding interest in protein folding ever since reading the following:

http://www.amazon.com/Chance-Necessity- ... 0394718259

It seems that with proteins, form is function (think, say, hemoglobin).

Where though one takes the combination of such interests, I really don't know.

[codysluder]

would give up asking these kinds of questions at folding@home except that I've had a longstanding interest in protein folding ever since reading the following:

http://www.amazon.com/Chance-Necessity- ... 0394718259

It seems that with proteins, form is function (think, say, hemoglobin)

I have not read that book, but I did read the FAH website. From the second paragraph of this very brief summary I understand that form is function.
http://folding.stanford.edu/English/Science