In collaboration with TensorWave
Rodinia Benchmarks: SCALE vs HIP on AMD Instinct MI355X
These are the original February 2026 Rodinia benchmark results obtained together with TensorWave, running SCALE against AMD's HIP toolchain on an AMD Instinct MI355X (ROCm v7.2.2). Across the Rodinia suite, SCALE delivered up to a 25.7× speed-up over HIP, with an average speed-up of 5.94× and faster execution on 10 of the 14 workloads measured.
Our most up-to-date benchmark results are live on the homepage. The February 2026 run on this page is preserved for historical reference.
up to 25.7x
faster than HIP5.94x
average speed up10/14
faster workloadsRodinia Benchmarks: SCALE vs HIP
Speed-up over HIP on AMD Instinct MI355X - ROCm v7.2.2 - February 2026
*The same measurement process was used to evaluate both the HIP and SCALE toolchains. The reported running time for each benchmark was measured from 150 total runs, split across 5 independent rounds, which were started after initialisation and after a warm-up phase. The reported execution time excludes CPU-only work such as file I/O or host-only memory allocations. The statistic used is the minimum recorded value, as the maximum likelihood estimator for the running time assuming that all measurements are subject to positive noise and follow a Pareto-like distribution. Other statistics including the standard deviation and median were also recorded to enable further analysis in future.