Re: TeraChem

From: McGuire, Kelly (mcg05004_at_byui.edu)
Date: Mon Apr 01 2019 - 13:25:39 CDT

Hey Gerard, does TeraChem us the RIJCOSX approximation for their DFT? I haven't used TeraChem, only ORCA. If so, I will be sure to do a simulation combining GPUs with and without RIJCOSX and post results here. I should be getting the TeraChem trial version, which is the full software version for 30 days, this week. When I get results, I'll post them here.

Kelly L. McGuire

PhD Candidate

Biophysics

Department of Physiology and Developmental Biology

Brigham Young University

LSB 3050

Provo, UT 84602

________________________________
From: Gerard Rowe <GerardR_at_usca.edu>
Sent: Monday, April 1, 2019 12:19 PM
To: Brian Radak; namd-l_at_ks.uiuc.edu; McGuire, Kelly
Subject: Re: namd-l: TeraChem

I'd be very curious to see the results of this. The benchmarks on TeraChem's website seem to be comparing conventional B3LYP in GAMESS with their GPU implementation. Modern DFT codes also include some flavor of the RI approximation (in Orca, RIJCOSX for B3LYP), which can result in speedups on the order of what they're reporting. What would be astonishing is if those effects can be combined.

-Gerard

________________________________
From: owner-namd-l_at_ks.uiuc.edu <owner-namd-l_at_ks.uiuc.edu> on behalf of McGuire, Kelly <mcg05004_at_byui.edu>
Sent: Saturday, March 30, 2019 3:21 PM
To: Brian Radak; namd-l
Subject: Re: namd-l: TeraChem

I wonder if anyone here has tried QMMM with GPUs and done some benchmarking. TeraChem offers a 30 day free trial, so I am going to do some benchmarking with my system. If I can get it to work, I will report it here.

Kelly L. McGuire

PhD Candidate

Biophysics

Department of Physiology and Developmental Biology

Brigham Young University

LSB 3050

Provo, UT 84602

________________________________
From: Brian Radak <brian.radak_at_gmail.com>
Sent: Saturday, March 30, 2019 1:13 PM
To: namd-l; McGuire, Kelly
Subject: Re: namd-l: TeraChem

This might have been discussed, but I don't know if this is planned in any official way - you are welcome to try scripting it.

It's also worth noting that fast QM codes do not always directly translate to fast QM/MM codes. Passing around a large set of MM charges (plus all of the ways that impacts the SCF calculation) can degrade performance in the range of 100-1000%. That's not to say it won't work, but it's not as guaranteed an outcome as you might assume.

On Fri, Mar 29, 2019, 7:31 PM McGuire, Kelly <mcg05004_at_byui.edu<mailto:mcg05004_at_byui.edu>> wrote:
Has anyone tried using TeraChem with NAMD for QMMM? They use GPUs, which some of their benchmarking showed 43x faster SCF calculations for about the same size QM region as I have. Anyone seen anything comparable to scalability with TeraChem and NAMD?

Kelly L. McGuire

PhD Candidate

Biophysics

Department of Physiology and Developmental Biology

Brigham Young University

LSB 3050

Provo, UT 84602

This archive was generated by hypermail 2.1.6 : Sat Dec 07 2019 - 23:20:36 CST