00:16:24 Supreeth Madapur Suresh: CUDA fortran is simple and very similar to CUDA C. But it depends on the audiance
00:16:33 Supreeth Madapur Suresh: What they want
00:37:10 Daniel Howard (CSG, NCAR): Found the email: Complete the Scientific Software Development and Applications Census
Earlier this week, the NCAR’s Exascale Tiger Team (ETT) ran an announcement on Sundog that was sent out in Staff Notes Daily about the Scientific Software Development and Applications Census. Participation in this survey helps us catalog the broad range of ongoing software development and scientific application activities across NCAR and supports our mission of ensuring that NCAR’s computing, modeling, and data analysis plans remain well aligned. I hope you will help the team by taking the time to complete this survey by September 3 at 5 p.m.
00:51:34 Mick Coady: GTT email alias: cisl_gtt@ucar.edu
00:53:24 Supreeth Madapur Suresh: We have SWM code if you need a C++ code. Its around 600 lines
00:54:37 Brian Dobbins: I’d like to take a look at that, too. Can you point us towards it on Cheyenne, Supreeth?
01:00:38 Daniel Howard (CSG, NCAR): @John Dennis I can include in an email as well but here were some (now admittedly old) documents about CICE GPU efforts, which at the time then transitioned to KNL/many-core archs: https://www.nopp.org/wp-content/uploads/2017/02/Chassignet.APPIGO.2016-1.pdf I was also referencing this presentation about NVIDIA’s understandings of thrusts in GPU ESM efforts. I’m not a leader on any of these projects so what the current status/interest with any of these projects I beyond. But ESPC project in Navy was broadly still interested in GPU efforts, though halo exchanges often posed as a limiting factor. https://www.nopp.org/wp-content/uploads/2017/02/Chassignet.APPIGO.2016-1.pdf
01:01:06 Daniel Howard (CSG, NCAR): *these projects is beyond me
01:02:30 Daniel Howard (CSG, NCAR): @Jeremy I want to know that too! I can coordinate again with Brian who seems to have better current understanding on UCX
01:02:57 Brian Dobbins: I think the OpenMPI BOF at SC’21 will cover some of that, too.
01:08:42 Mick Coady: 👍 Brian
01:11:07 Brian Dobbins: Isn’t OpenMPI deprecating the OpenIB BTL? I think you might be stuck with UCX.
01:11:59 Ben M: yes, to go away in OpenMPI 5.x
01:12:08 Brian Dobbins: (Or is it ‘ofi’ now.. I always get their names confused.)
01:12:13 Brian Dobbins: Thanks, Ben. Thought I heard that.
01:12:26 Jeremy Sauer: Thanks all
01:12:29 Ben M: ofi is the OpenFabrics backend I think. That’s different
01:12:40 Ben M: (it is an alternative to UCX)
01:12:46 Brian Dobbins: Ah, got it. Thanks again.