Round Table Discussion B

1
Round Table Discussion B

• What are challenges for adoption of GPU/FPGA
  accelerated simulations?
• How far are we from a wide scale adoption?
• Wider impact
• Large machines (supercomputers) with GPU/FPGA?
• GPU-enabled Linux Clusters?
• GPU-enabled desktops?
• In addition to running MD trajectories, what can
  benefit
• Visualization?
• Coarse-graining, analysis?

2
Hardware

• All things considered, which hardware is best
  suited for MD CPUs, GPUs, FPGAs, other?
• Whats the top item on your hardware wish list?

3
Hardware predictions

• What will typical high performance computer
  hardware look like in 2020?
• Commodity or special purpose?
• How many cores? Nodes?
• Where will the bottlenecks be?
• How about compilers?
• How about MD software?

4
Language

• In general, do you prefer CUDA, OpenCL, other?
• Which language will be the favorite in 2020?
• How do you handle portability? Do you maintain
  two separate code bases for GPU vs CPU only?

5
Software

• Whats the top item on users software wish list?
• Which do we need more faster computers, or
  better force fields?
• How do you handle quality control? VV? Is this
  area neglected, or are we doing this adequately?
• What are the most important tricks that GPUMD
  developers should know?
• What will be the next disruptive technology
  change that will impact MD simulation?

6
Community

1. How can our community better leverage each
   others work?
2. What about common libraries? OpenMM?
3. More workshops? Open-source code? Sharing of
   lessons-learned?