Page Comparison

Dial in Information

Elsie Wahlig is inviting you to a scheduled Zoom meeting.

Topic: Elsie Wahlig's Personal Meeting Room

Join Zoom Meeting
https://zoom.us/j/6112761834

One tap mobile
+16465588656,,6112761834# US (New York)
+17207072699,,6112761834# US

Dial by your location
+1 646 558 8656 US (New York)
+1 720 707 2699 US
+1 877 853 5247 US Toll-free
+1 888 788 0099 US Toll-free
Meeting ID: 611 276 1834
Find your local number: https://zoom.us/u/aeGoL7KgTk

Agenda 

• Comments/Feedback on Jira Status

• Engineer allocation

• Collaboration

• Status / Roadmap

• AI on Servers

• Longterm Scope

• AOB

Minutes

• Presented standard 5 slides on Jira report.
• Received clarification comments for definition of LBI and whether HPC-164 remains blocked. Linaro Big Idea and HPC-164 no longer needs to be blocked due to change in infrastructure.
• Ben asked whether Dwarf was being considered for SVE optimisations. Dwarf as a fileformat standard for debugging and should in itself be platform agnostic however, tools based on Dwarf and work already started on Valgrind are limited in supporting the A64FX until Fujitsu release the technical specification. Likewise the performance profiling tool PAPI will require knowledge of the A64FX details. Expected 1Q20.
• Engineer numbers presented for 2019 (4) and expected for 2020 (3). There should be one extra from Arm (name to be supplied by Andy) to bring 2020 to 4.
• There was a request for more engineering resources to be made available by members.
• openHPC now have 6 dedicated TX2 nodes for CI.
• Larry to share details (where possible) related to ORNL Wombat evaluation for applications such as Gromacs to avoid repetition.
• AI on Servers LBI
• Suggestion made to review OpenMP > v4.5 as an alternative target for SVE/SVE2 enablement than OpenCL.
• Conclusion, given current resources and no more forthcoming then expectation of AI on Servers will only be able to deliver Arm optimisations for Tensorflow, perhaps PyTorch too.
• Ike reported that Fujitsu/Riken are no longer interested in Chainer, only Tensorflow.
• Additional features such as Spiking Neural Networks and development towards a true AI were sidelined.

Update

• Fujitsu have assigned Firmware Engineer Masato Fukumori as replacement for Compiler Engineer Masaki Arai.

Slides for Meeting

Recording Link

• Link here