Opencl benchmark mac code#

(Execution timings reported, lower is better) Overall, the M1 Max is about 70% faster than the 5600X. Here, the PyPerformance benchmark, an official suite of real-world Python application benchmarks. In general, Python runs really fast on the M1 Max. 5600X, with MKL is faster on 3 tasks (SVD, VecMul, eigendecomp).Both are about the same on 1 task (stats).M1 Max, with Accelerate, is faster on 3 tasks (RNG, special, Cholesky).(Execution timings for function reported, lower is better) Tested functions include random number generation, special functions (exp/trigo/log/power), statistical (mean/median/std), vector multiplication, SVD, and more. For other common NumPy functions, the differences while benchmarking are not as dramatic.

Compared to the 5600X, the M1 Max CPU is generally at least 2x faster any data type and any matrix size, even with MKL enabled on the 5600X. Presumably, if you use Accelerate directly via a lower-level language, you can get even better performance. Mind you, this is via normal Python NumPy code. The results are quite stunning, especially for single precision, or FP32 (which is commonly used in machine learning applications), giving us about almost 2 TFLOPS for large enough matrix sizes (about the level of a GTX 1050 Ti).

Opencl benchmark mac code#

What I didn't know is that you can compile NumPy to work with Accelerate, which allows you to easily leverage the AMX instructions via normal NumPy code (NumPy installed via conda does not include Accelerate support, and instead uses cblas). We already know that the M1 Max CPU should have really strong matrix multiplication performance due to Apple's 'hidden'/undocumented AMX co-processor embedded in the CPU complex, and that it is leveraged when you use Apple's Accelerate framework. As a reference point for comparison purposes, I have my desktop AMD Ryzen 5600X CPU with DDR4 memory and not overclocked. There should be no difference between the CPU performance on the M1 Max and M1 Pro, barring slightly higher memory bandwidth available to the CPU complex on the M1 Max. The CPU in the M1 Max is a 10-core CPU, with 2 efficiency cores at 2.1 GHz and 8 performance cores at 3.0 GHz during all-core load. Benchmarking is complex with many different methodologies and configurations, and I am under no delusion that what I have come up with is the best way to do things. However, I still had questions, so here I am with some (casual) benchmarks that I hope add some additional perspective into interesting hardware capabilities on the M1 Max SOC, just for fun and out of my curiosity. Benchmarking is a complex task, and there are plenty of outlets like Anandtech that have comprehensively done different benchmarks. It's amazing for day-to-day use, but I was curious about the actual capabilities of the SOC. I recently got a 2021 16' M1 Max MacBook Pro.