Chapel Nightly Performance Graphs
The following graphs are used by the Chapel development team in order to track Chapel performance over time. It should be noted that the graphs show changes not only in the Chapel compiler, runtime, and libraries, but also to the tests themselves, and the systems on which the tests are run. To the extent possible, the graphs contain annotations indicating causes of nontrivial changes.
Since Chapel is open-source, community members are encouraged to submit their own Chapel performance tests to be included in these graphs. For an introduction to what is needed to submit a new performance test, see our GitHub How to Contribute as well as the developer's documentation for creating Chapel performance tests.
Primary Configurations:
chapcs runs our single-locale performance suite on a node with 24 cores and 256 GB RAM
16-node-xc runs our multi-locale performance suite on 16 nodes of a Cray XC with the Aries network. Each node has 48 cores and 192 GB RAM
16-node-apollo-hdr runs our multi-locale performance suite on 16 nodes of an HPE Apollo system with a 200 Gb/s HDR InfiniBand network. Each node has 72 cores and 512 GB RAM
comm-counts runs our communication count tests
chapcs/no-local runs our single-locale suite with --no-local
chapcs/c-backend runs our single-locale suite with the gnu backend
chapcs/clang runs our single-locale suite with clang as the backend compiler
1-node-xc runs our single-locale suite with intel, cray, and gcc backends on a Cray XC node
comp-default runs with compilation time tracking enabled
comp-no-local runs with --no-local and compilation time tracking enabled
comp-fast runs with --fast and compilation time tracking enabled
single-node comm=none benchmark results on a single node with 24 cores and 256 GB RAM
16-node-xc comm=ugni benchmark results on 16 nodes of a Cray XC with an Aries network. 48 cores and 192 GB RAM per node