FMA (fused-multiply accumulate) Benchmark
Currently benchmarks pipelined FMAs on ARM and Intel chips.
To benchmark on a single thread,
git clone --recursive https://github.com/bwasti/mac_benchmark.git
cd mac_benchmark
make arm || make intel # only one will work
./bench
For multiple threads, rebuild the binary:
T=-DTHREADS=$(sysctl -a | grep machdep.cpu.core_count | awk '{print $2}')
make arm CFLAGS=$T || make intel CFLAGS=$T
./bench
Results collected so far:
| Hardware | Chip | Single Core GFLOPs | Cores | All Cores GFLOPs |
|---|---|---|---|---|
| 2020 Macbook Air | M1 | 91 | 8 | 460 |
| 2019 16" Macbook Pro | 2.4 GHz 8-Core Intel Core i9 | 135 | 8 | 800 |
Adding More Results
Please submit a pull request to update the README