Scaling your Workflows with Chapel

Workflows written using Chapel scale enormously. The same code written on your laptop can perform on however many nodes you can give it, be it one, dozens, or thousands. Applications written by Chapel users have successfully scaled to more than 8,000 compute nodes, 1,000,000 CPU cores, and 1,000 GPUs. The largest supercomputers in the world run Chapel. No matter the scale of your system, you can scale too.

Scaling Results with Chapel

Arkouda Argsort

Arkouda's argsort routine can sort 256 TiB of data in just 31 seconds, scaling to more than 8,000 nodes. All in ~100 lines of Chapel code.

Learn More

Navier-Stokes Simulations

Solve Navier-Stokes equations in Chapel, achieving comparable performance to C++ and MPI with half the code and improved clarity and flexibility.

Learn More

Bale Index Gather

Gather arbitrary elements from distributed arrays or tables in an aggregated manner to scale to thousands of nodes and hundreds of thousands of processor cores.

Learn More

STREAM Triad

The de facto industry standard benchmark for measuring memory performance.

Learn More

HPCC Random Access

Measures the performance of concurrent random updates to a distributed array.

Learn More

NAS FT

Computes the FFT of a distributed 3D array by computing 1D FFTs and transposing to localize other dimensions.

Learn More

PRK Stencil

Measures the performance of applying a radius-2 star stencil to a distributed 2D array. Emphasizes the performance of nearest-neighbor communication.

Learn More

Why Chapel Makes Scaling Easier

Global View of Data

Chapel has built-in support for distributed global arrays, meaning computations written for a single node can run on an entire cluster.

Multi-level Parallelism

The same language features support parallelism at the node and cluster level. Parallelize once and reap the benefits everywhere.

Data and Computational Locality

Control where your program’s tasks are executed using Chapel’s extensive locality features.