4 ms·

Anton (computer)

54 points by luu 11y ago

cing 11y ago
This page does not reference Anton 2 (http://dl.acm.org/citation.cfm?id=2683599 http://dl.acm.org/citation.cfm?id=2683599), there's also a really technical presentation here on the hardware if you're into that (http://www.hotchips.org/wp-content/uploads/hc_archives/hc26/HC26-11-day1-epub/HC26.11-1-High-Performance-epub/HC26.11.130-Anton-2-Butts-Shaw-Shaw-Res-Search.pdf http://www.hotchips.org/wp-content/uploads/hc_archives/hc26/...)
gamesbrainiac 11y ago
When does a computer become defined as massively parallel? 100+ cores?
- andrepd 11y ago
  Maybe it's a qualitative difference, as in parallel by design, not just throwing a bunch of nodes together in a typical Beowulf configuration.
- jacquesm 11y ago
  The key is coordination between the cores.
- haxeny1 11y ago
  This is an example of parallel system that i like: http://librestats.com/2012/10/16/r-at-12000-cores/ http://librestats.com/2012/10/16/r-at-12000-cores/ . i haven't done anything near this scale, although, that would be a goal to aim for. Although, i am not entirely certain if the above link answers your specific query, because the cores in the link are not housed in what would be considered a single computer. Anton is a super computer unlike the libmpi based solutions that I linked to here. Additional links: mpi: http://www.mpi-forum.org/docs/ http://www.mpi-forum.org/docs/ http://r-pbd.org/ http://r-pbd.org/ Hope this helps. edit: typos, clarity etc.
- nickpsecurity 11y ago
  It was usually applied to at least 1,000 cores in the past plus ability of app to use them. First one was Thinking Machines CM line with 64,000+ simple CPU's. https://en.m.wikipedia.org/wiki/Connection_Machine https://en.m.wikipedia.org/wiki/Connection_Machine
HarryHirsch 11y ago
There are ways to manipulate the potential surface of a system so that your molecule of interest reaches all of available phase space within reasonable wallclock time (read: a few days at most). The keyword is "accelerated MD"; there are several methods. The McCammon group at UCSD has done much work in the field. This doesn't require an Anton in the basement, just a regular PC and a modern graphics card. The real question is: what does David Shaw, the old sphinx, propose to do with his system? I haven't seen much research published, but the common complaint is that he is very slow publishing. DESRES is a weird outfit on a sunny day.
- eggie 11y ago
  Beautiful. I just spent two weeks wandering around trying to figure out the name of accelerated MD, and suddenly you've got the boldness to educate HN about it. There is a fruitful space of combining good priors about the pattern of folding expected out of a given protein with MD simulations. I dream that it revolutionizes how we determine the structure of proteins.
- cing 11y ago
  You're right about enhanced sampling methods (of which aMD is just one). The D.E. Shaw lab actually presented detailed comparisons between conventional simulation and techniques like this at this year's Biophysical Society Meeting. I don't see why you couldn't also run accelerated techniques on Anton, or perhaps a related ASIC system though. Now, you don't have to look too far to see what scientists are doing with this system (https://www.deshawresearch.com/publications.html https://www.deshawresearch.com/publications.html, http://www.psc.edu/index.php/resources-for-users/computing-resources/anton http://www.psc.edu/index.php/resources-for-users/computing-r...). From the D.E. Shaw group alone, 7 papers in Science, 7 papers in Nature-family journals, and 13 papers in Proc. Natl. Acad. Sci. The publish slowly because they don't have any obligation to publish. I can't vouch for all of those papers, but it's definitely a step in the right direction for the field.

54 points by luu 11y ago

cing 11y ago

This page does not reference Anton 2 (http://dl.acm.org/citation.cfm?id=2683599 http://dl.acm.org/citation.cfm?id=2683599), there's also a really technical presentation here on the hardware if you're into that (http://www.hotchips.org/wp-content/uploads/hc_archives/hc26/HC26-11-day1-epub/HC26.11-1-High-Performance-epub/HC26.11.130-Anton-2-Butts-Shaw-Shaw-Res-Search.pdf http://www.hotchips.org/wp-content/uploads/hc_archives/hc26/...)

gamesbrainiac 11y ago

When does a computer become defined as massively parallel? 100+ cores?

andrepd 11y ago
Maybe it's a qualitative difference, as in parallel by design, not just throwing a bunch of nodes together in a typical Beowulf configuration.
jacquesm 11y ago
The key is coordination between the cores.
haxeny1 11y ago
This is an example of parallel system that i like: http://librestats.com/2012/10/16/r-at-12000-cores/ http://librestats.com/2012/10/16/r-at-12000-cores/ . i haven't done anything near this scale, although, that would be a goal to aim for. Although, i am not entirely certain if the above link answers your specific query, because the cores in the link are not housed in what would be considered a single computer. Anton is a super computer unlike the libmpi based solutions that I linked to here. Additional links: mpi: http://www.mpi-forum.org/docs/ http://www.mpi-forum.org/docs/ http://r-pbd.org/ http://r-pbd.org/ Hope this helps. edit: typos, clarity etc.
nickpsecurity 11y ago
It was usually applied to at least 1,000 cores in the past plus ability of app to use them. First one was Thinking Machines CM line with 64,000+ simple CPU's. https://en.m.wikipedia.org/wiki/Connection_Machine https://en.m.wikipedia.org/wiki/Connection_Machine

HarryHirsch 11y ago

There are ways to manipulate the potential surface of a system so that your molecule of interest reaches all of available phase space within reasonable wallclock time (read: a few days at most). The keyword is "accelerated MD"; there are several methods. The McCammon group at UCSD has done much work in the field. This doesn't require an Anton in the basement, just a regular PC and a modern graphics card. The real question is: what does David Shaw, the old sphinx, propose to do with his system? I haven't seen much research published, but the common complaint is that he is very slow publishing. DESRES is a weird outfit on a sunny day.

eggie 11y ago
Beautiful. I just spent two weeks wandering around trying to figure out the name of accelerated MD, and suddenly you've got the boldness to educate HN about it. There is a fruitful space of combining good priors about the pattern of folding expected out of a given protein with MD simulations. I dream that it revolutionizes how we determine the structure of proteins.
cing 11y ago
You're right about enhanced sampling methods (of which aMD is just one). The D.E. Shaw lab actually presented detailed comparisons between conventional simulation and techniques like this at this year's Biophysical Society Meeting. I don't see why you couldn't also run accelerated techniques on Anton, or perhaps a related ASIC system though. Now, you don't have to look too far to see what scientists are doing with this system (https://www.deshawresearch.com/publications.html https://www.deshawresearch.com/publications.html, http://www.psc.edu/index.php/resources-for-users/computing-resources/anton http://www.psc.edu/index.php/resources-for-users/computing-r...). From the D.E. Shaw group alone, 7 papers in Science, 7 papers in Nature-family journals, and 13 papers in Proc. Natl. Acad. Sci. The publish slowly because they don't have any obligation to publish. I can't vouch for all of those papers, but it's definitely a step in the right direction for the field.