BeBOP SpMV Benchmark

Sparse matrix-vector mulitply (SpMV) is a common operation in scientific codes. It finds applications in iterative methods to solve sparse linear systems and information retreival, among other places. Thus knowing how well a particular machine performs SpMV is useful for both vendors and consumers alike. Our benchmark provides an effective way of evaluating a particular platform's ability to perform SpMV, and it can do so in only a few minutes.


Hormozd Gahvari
Mark Hoemmen
James Demmel
Katherine Yelick


We currently have available for download a benchmark for uniprocessor SpMV. We hope to have a distributed-memory parallel version available in the future.


H. Gahvari, M. Hoemmen, J. Demmel, K. Yelick. Benchmarking Sparse Matrix-Vector Multiply in Five Minutes. SPEC Benchmark Workshop 2007, Austin, TX, January 21, 2007.

H. Gahvari. Benchmarking Sparse Matrix-Vector Multiply. Master's Thesis, University of California, Berkeley, December 2006.


We maintain our lists of benchmark results here, and we encourage you to submit your own results to us. Please place platform information and benchmark numbers into the form below, and click on the "Submit" button. The results will be sent for us to look at and then post in our results section. Please fill in every entry.

Machine name:
Cache size:
Compiler name:
Time constraint: minutes
Memory constraint:

Benchmark output (MFLOP/s):