GPU acceleration, Avogadro 2.0

Greetings!

Could anyone tell me if Avogadro supports GPU acceleration for force field
calculations? If so, is it strictly OpenCL-based, and what graphics cards
would you recommend?

Also, I’m wondering about the feature list and timing for 2.0 - I know there
is 2.0 info on the web site somewhere, but I can’t find it!

Sincerely,
James

I apologize I didn’t have a chance to reply to your personal message earlier, but it’s better to reply on the list.

Could anyone tell me if Avogadro supports GPU acceleration for force field calculations? If so, is it strictly OpenCL-based, and what graphics cards would you recommend?

At the moment, no. It doesn’t even use OpenMP (for use of multiple processors) by default. Don’t despair. For one, the underlying Open Babel library, which handles force field minimization has a pile of performance improvements which will be incorporated into Avogadro 1.1.

There’s also a student working on parallelization / OpenCL and general algorithmic improvements this summer. I know there are many improvements coming.

Also, I’m wondering about the feature list and timing for 2.0 – I know there is 2.0 info on the web site somewhere, but I can’t find it!

I don’t think we’ve discussed 2.0 at all yet. Version 1.1/1.2 will be released RealSoonNow. If you’d like to test a “beta” Windows build, you can find one here:
http://avogadro.openmolecules.net/pettt/Avogadro-1.1.0-pettt.zip

Marcus promised a 2.0 discussion once v1.1.0 was released.

Hope that helps,
-Geoff

Thanks! Maybe I saw “1.2” and not “2.0”. I’ll try out 1.1.

As general feedback, the software is great so far (it does crash frequently,
but I just save often and I haven’t been bothered enough by it to track down
why it might be – perhaps my Nvidia drivers). My only real concern is that
my machine will end up not allowing me to work with the number of atoms I
want, although my impression is that this is frequently true of molecular
simulations in general.

I’m not a computational chemist and you can’t believe how satisfying it is
to just snap atoms together, hit “optimize” and have them assume the shape
you intended. And, the built-in force fields seems perfectly accurate so far
(which is why I want to be able to use them on lots of atoms). The times
I’ve looked at the resulting structures and thought “That’s not exactly how
it should look,” I’d misplaced a bond, or a hydrogen was in the wrong
conformation.

Sincerely,
James

-----Original Message-----
From: Geoff Hutchison [mailto:geoff.hutchison@gmail.com]
Sent: Wednesday, July 18, 2012 10:01 PM
To: avogadro-discuss@lists.sourceforge.net
Subject: Re: [Avogadro-Discuss] GPU acceleration, Avogadro 2.0

I apologize I didn’t have a chance to reply to your personal message
earlier,
but it’s better to reply on the list.

Could anyone tell me if Avogadro supports GPU acceleration for force
field
calculations? If so, is it strictly OpenCL-based, and what graphics cards
would
you recommend?

At the moment, no. It doesn’t even use OpenMP (for use of multiple
processors) by default. Don’t despair. For one, the underlying Open Babel
library, which handles force field minimization has a pile of performance
improvements which will be incorporated into Avogadro 1.1.

There’s also a student working on parallelization / OpenCL and general
algorithmic improvements this summer. I know there are many
improvements coming.

Also, I’m wondering about the feature list and timing for 2.0 - I know
there
is 2.0 info on the web site somewhere, but I can’t find it!

I don’t think we’ve discussed 2.0 at all yet. Version 1.1/1.2 will be
released
RealSoonNow. If you’d like to test a “beta” Windows build, you can find
one
here:
http://avogadro.openmolecules.net/pettt/Avogadro-1.1.0-pettt.zip

Marcus promised a 2.0 discussion once v1.1.0 was released.

Hope that helps,
-Geoff


Live Security Virtual Conference
Exclusive live event will cover all the ways today’s security and threat
landscape has changed and how IT managers can respond. Discussions will
include endpoint security, mobile security and the latest in malware
threats.
http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/


Avogadro-Discuss mailing list
Avogadro-Discuss@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/avogadro-discuss

As general feedback, the software is great so far (it does crash frequently,
but I just save often and I haven’t been bothered enough by it to track down
why it might be – perhaps my Nvidia drivers)

I have heard anecdotal reports on Windows 7. My students (with integrated graphics) don’t usually have these issues, so I suspect updating drivers might help. If you’re willing to help us debug these issues, we’d be thrilled.

My only real concern is that my machine will end up not allowing me to work with the number of atoms I
want, although my impression is that this is frequently true of molecular simulations in general.

It just depends on how long you’re willing to wait. Certainly larger simulations (thousands of atoms) are usually run on clusters.

Our force field code certainly has room for optimization, and as I said this is underway. But yes, MMFF94 is highly accurate for drug-like compounds, and UFF is a useful alternative for a wide range of the periodic table.

In any case, besides our code, I’m actively investigating the OpenMM libraries, which are GPU-accelerated molecular mechanics methods out of Stanford. Certainly the performance looks excellent.

Thanks for the feedback!

-Geoff

I can pretty much crash it at will, so just let me know how I can provide
useful debugging info. I would have sent info in already, but all I get is
"avogadro.exe has stopped working…" There aren’t any details to copy.

One thing I’ve noticed that is repeatable is copying certain groups of
atoms. I don’t know what it is about them. It’s not a lot of atoms (< 100)
and it’s not all groups of atoms (even larger ones), but I have some that
will do it 100% of the time. This is on the 1.1 beta – I haven’t tried
1.0.3 (I uninstalled it prior to installing 1.1 – can they coexist?).

Sincerely,
James

-----Original Message-----
From: Geoff Hutchison [mailto:geoff.hutchison@gmail.com]
Sent: Saturday, July 21, 2012 10:29 PM
To: avogadro-discuss@lists.sourceforge.net
Subject: Re: [Avogadro-Discuss] GPU acceleration, Avogadro 2.0

As general feedback, the software is great so far (it does crash
frequently, but I just save often and I haven’t been bothered enough
by it to track down why it might be – perhaps my Nvidia drivers)

I have heard anecdotal reports on Windows 7. My students (with integrated
graphics) don’t usually have these issues, so I suspect updating drivers
might
help. If you’re willing to help us debug these issues, we’d be thrilled.

My only real concern is that my machine will end up not allowing me to
work with the number of atoms I want, although my impression is that
this
is frequently true of molecular simulations in general.

It just depends on how long you’re willing to wait. Certainly larger
simulations
(thousands of atoms) are usually run on clusters.

Our force field code certainly has room for optimization, and as I said
this is
underway. But yes, MMFF94 is highly accurate for drug-like compounds, and
UFF is a useful alternative for a wide range of the periodic table.

In any case, besides our code, I’m actively investigating the OpenMM
libraries, which are GPU-accelerated molecular mechanics methods out of
Stanford. Certainly the performance looks excellent.

Thanks for the feedback!

-Geoff


Live Security Virtual Conference
Exclusive live event will cover all the ways today’s security and threat
landscape has changed and how IT managers can respond. Discussions will
include endpoint security, mobile security and the latest in malware
threats.
http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/


Avogadro-Discuss mailing list
Avogadro-Discuss@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/avogadro-discuss

I can pretty much crash it at will, so just let me know how I can provide
useful debugging info. I would have sent info in already, but all I get is
"avogadro.exe has stopped working…" There aren’t any details to copy.

I’ll see if I can get you some sort of debugging version. The Windows development tools are new to me.

One thing I’ve noticed that is repeatable is copying certain groups of
atoms. I don’t know what it is about them. It’s not a lot of atoms (< 100)
and it’s not all groups of atoms (even larger ones), but I have some that
will do it 100% of the time.

If you can give me a repeatable sequence (e.g., open this file, select this, etc.) I’d definitely be able to find the bug. Chances are pretty good it may also exist on Mac and Linux too.

Thanks,
-Geoff