Assessing Community Training Needs for GPU Computing in the Earth Sciences

JavaScript isn't enabled in your browser, so this file can't be opened. Enable and reload.

All responses are by default anonymous.

With the upcoming deployment of Derecho with greater capability of GPU computing, this survey aims to assess the training and learning development needs for GPU computing at NCAR and within the wider Earth Sciences community NCAR serves.

Prioritizing your own experiences and that of your immediate colleagues, please respond to the below questions, reflecting on your own interests rather than that of generalized ideas you might have about the needs of the broader research community. If a question does not apply to you or you feel you have minimal awareness to provide a meaningful response, feel free to skip.

How do you primarily identify?

Data Scientist or Engineer

Domain Scientist or Researcher

Research Software Engineer

Project Manager

Student

System Administrator Engineer

University Faculty Researcher and Educator

Other:

Clear selection

What programming language(s) do you primarily work with today?

Fortran

C/C++

Python

Julia

Matlab

Other:

Rate your current ability level with respect to GPU computing.

No previous training or experience

Extensive experience and expert level knowledge

Clear selection

Do you employ GPU computing within any of your current projects?

Yes

Clear selection

Do you anticipate a need for or substantial benefit by incorporating GPU computing into your work?

Yes

Maybe

Clear selection

Do you have a general idea and understanding to be able to assess when GPU computing might benefit or speedup a proposed project?

Yes

Maybe

Clear selection

Are there current limitations or what is preventing you from utilizing GPUs in your current projects?

Which programming language would you prefer when learning GPU Computing concepts or expanding upon already learned concepts?

Fortran

C/C++

Python

Julia

Matlab

Other:

Clear selection

When learning GPU computing concepts, would you prefer attending live seminars and training events with opportunities for direct interaction with presenters and peers or accessing these training events asynchronously to learn at your own pace?

Live trainings without live coding demonstrations

Live trainings with coding examples lead during training

Accessing archived materials and pre-recorded trainings

Clear selection

Would you benefit from and attend regularly provided office hours to answer questions or address issues for GPU users on Casper/Derecho?

Would attend office hours regularly

Would attend office hours infrequently

Unsure if I would benefit

I do not anticipate a need for GPU computing in future projects

Clear selection

What type(s) of GPU development would you prefer to learn and/or be the most effective use of your time?

CUDA Programming in Fortran

CUDA Programming in C/C++

CUDA Programming with Python

CUDA Programming with Julia

Directive based computing (OpenACC/OpenMP/std par/do concurrent)

Kokkos/RAJA

OpenCL

Specialty Libraries or Domain Specific Languages, GPU details largely hidden from developer

Other:

Which more advanced GPU development concept(s) would you most benefit from?

Profiler usage and optimization techniques

Multi-GPU programming

Verifying Correctness of GPU code

Specific usage techniques of libraries/packages like in following question

Other:

Are there specific libraries or packages, such as in Python or Julia, which you'd like to learn more with respect to GPU computing?

Examples may include Legate as NumPy, CuPy, TensorFlow, PyTorch, CUDA.jl, Oceanigans.jl, GPUifyLoops.jl

To avoid re-inventing the wheel with respect to materials already publicly available, how do you recommend tailoring GPU Computing training specifically for the Earth sciences community and addressing present unmet needs or pain points?

Any specific best practices/tools you've already learned under GPU Computing which would be useful to share with the larger community?

Any other feedback/requests you'd like to provide?

We anticipate starting the GPU training series relatively soon, starting with basic topics first then more advanced materials as we progress. Being aware of dates for conferences and holidays, when would you prefer a start date for this series?

Early/Mid December

Early/Mid January

Late January or early February

Clear selection

If you're open to be contacted further about your responses, please provide your email

Submit

Clear form

Never submit passwords through Google Forms.

This form was created inside of NCAR|UCAR. Report Abuse

Forms