F@H 100x faster!?

[RMouse]

Anyone look at project 7032? VERY exciting stuff. Possibly increasing the power of Folding Home by 100 times!

http://fah-web.stanford.edu/cgi-bin/fah ... ned?p=7032

I hope it is a great success.

Keep folding!

[iceman1992]

It's not the only one.. There are tens of projects with the same description.. from 7000-7030 something.. And the 10084 group of projects too..

[Jonazz]

I'm folding one now

They've been around for a while. See viewtopic.php?f=17&t=18884&p=188919#p188919

But don't forget: while 100x faster is a big increase, FAH has already come a long way! Back when this project started, they could fold proteins on the nanosecond timescale. Around 2008, they could fold proteins on the milisecond timescale (http://folding.stanford.edu/English/Papers#ntoc29). That's 1 million times faster! They're probably working on proteins wich take tens of milieseconds to fold today.

[Jesse_V]

But don't forget: while 100x faster is a big increase, FAH has already come a long way! Back when this project started, they could fold proteins on the nanosecond timescale. Around 2008, they could fold proteins on the milisecond timescale (http://folding.stanford.edu/English/Papers#ntoc29). That's 1 million times faster! They're probably working on proteins wich take tens of milieseconds to fold today.

There's a blog post that talks about F@h's capability growth over time: http://folding.typepad.com/news/2012/06 ... -come.html

[Jonazz]

That's what I based my post on

[TonyStewart14]

If you look at the description for Project 7035 here, you'll see that the project is testing "a novel means to significantly accelearate [sic] Folding@home Molecular Dynamics calculations by 100x without any additional increase in hardware." How cool would that be if it were true! A hex-core i7 could be as productive as 10 quad-16 bigadv systems, for example. They have a paper linked there too, but I'd be really interested to see how this project progresses.

[Jesse_V]

It's even more amazing you realize how accelerated F@h already is.
1) By using adaptive sampling and Markov state models they get something like a six orders of magnitude improvement over standard molecular dynamics.
2) The work gets done in parallel.
3) They use OpenMM for further hardware optimization.
4) Gromacs is widely considered to be one of the most efficient molecular dynamics software packages on the planet.
5) For certain calculations GPUs and PS3s can offer additional orders of magnitude speedups.
6) And this whole thing is running on the world's most powerful distributed computing project.

And they're looking into a 100x speed increase on top of that!

[Stonecold]

100x seems like a bit of an over-exaggeration. It just seems a little too large of an increase (and by such a generic number) to be completely true. I may be wrong though, but this is just my speculation.

[k1wi]

100x seems like a bit of an over-exaggeration. It just seems a little too large of an increase (and by such a generic number) to be completely true. I may be wrong though, but this is just my speculation.

Would you have been more convinced if the figure was 97x?

[Stonecold]

100x seems like a bit of an over-exaggeration. It just seems a little too large of an increase (and by such a generic number) to be completely true. I may be wrong though, but this is just my speculation.

Would you have been more convinced if the figure was 97x?

Not that I won't believe it, it's just so impressive that I'm skeptical that FAH could improve 10,000% in one go.

[Jesse_V]

100x seems like a bit of an over-exaggeration. It just seems a little too large of an increase (and by such a generic number) to be completely true. I may be wrong though, but this is just my speculation.

Would you have been more convinced if the figure was 97x?

Not that I won't believe it, it's just so impressive that I'm skeptical that FAH could improve 10,000% in one go.

From the paper's abstract:

We present numerical results showing that time steps of up to 1000 fs can be used, with real speedups of up to 200 times over plain molecular dynamics. We present results of successfully folding the Fip35 mutant of WW domain.

[7im]

It's probably okay to be skeptical about PG's ability to apply this technique to all proteins of all sizes. But it is not okay to question stated facts. That's like not believing in gravity.

[Stonecold]

It's probably okay to be skeptical about PG's ability to apply this technique to all proteins of all sizes. But it is not okay to question stated facts. That's like not believing in gravity.

Oh OK. I was under the impression that this was a universal thing that applied to all proteins.

And I don't believe in gravity. It's actually the Flying Spaghetti Monster holding us on Earth with his noodles.

[bruce]

I understand quite a bit of the underlying mathematics of this new methodology. There is a sound basis to believe that accurate results can be achieved with significantly longer time-steps. I don't know enough about "all proteins" to predict an answer to 7i'm speculation, however, so there will need to be enough beta testing to ascertain any practical limitations of the methodology.

For those of you who are expecting this announcement to imply you'll earn 100x as many points, that's simply not going to happen. This methodology will still be benchmarked by the traditional methods.

[Jonazz]

100x seems like a bit of an over-exaggeration. It just seems a little too large of an increase (and by such a generic number) to be completely true. I may be wrong though, but this is just my speculation.

I suggest you read this:
http://folding.typepad.com/news/2012/06 ... -come.html

100x actually isn't that much if you realise where we came from in 2000.