Current performance benchmarks

[shyams2]

Commit-Hash: 54b559c

Machine: GPU node on Savio @ Berkeley

Setup:

• KH instability setup
• Fields step is not performed
• Strang splitting for advection in position space and collision steps.
• RK2 time-stepping used for collision step
• Benchmarks were obtained after running for 100 time-steps

Strong-Scaling:

Resolution	nGPUs	Time/Iteration
128 X 128 X 16^3	1	7.96s
128 X 128 X 16^3	2	3.81s
128 X 128 X 16^3	3	1.74s
128 X 128 X 16^3	4	1.77s

Weak-Scaling:

Resolution	nGPUs	Time/Iteration
32 X 32 X 32^3	1	3.88s
32 X 64 X 32^3	2	4.8s
48 X 64 X 32^3	3	3.48s
64 X 64 X 32^3	4	4.39s

Setup:

• KH instability setup
• Strang splitting for advection in position space, advection in position space and collision steps.
• RK2 time-stepping used for collision step
• Fields initialization: KSP solver
• Fields evolution: FDTD
• Benchmarks were obtained after running for 100 time-steps

Strong-Scaling:

Resolution	nGPUs	Time/Iteration
128 X 128 X 16^3	1
128 X 128 X 16^3	2
128 X 128 X 16^3	3
128 X 128 X 16^3	4

Weak-Scaling:

Resolution	nGPUs	Time/Iteration
32 X 32 X 32^3	1
32 X 64 X 32^3	2
48 X 64 X 32^3	3
64 X 64 X 32^3	4

[shyams2]

This numbers are now obsolete. Current numbers can be looked up from #39