Validation
Independent output checks
Experimental outputs were checked against known prime datasets and independent prime-sieving tools to verify numerical correctness.
← HexCode
Research · Executive summary
P.A.T.H.
High-performance GPU-accelerated prime sieving on Apple Silicon.
Prime Accelerated Throughput Handler is an experimental, GPU-native prime-discovery engine designed around Apple Silicon and Metal compute. This page presents a non-confidential overview of the architecture, validation approach and selected measured results. Internal algorithms and optimisation details are intentionally omitted.
01 · Architecture
Heterogeneous compute by design.
P.A.T.H. was built to move a highly data-parallel workload away from general-purpose CPU threads and onto Apple's integrated GPU. Metal compute shaders perform the sieve workload while the CPU primarily handles orchestration, dispatch and output.
The implementation combines Metal compute, Objective-C++ bindings and arbitrary-precision integer backends. The research focus is not simply raw prime-counting speed, but the behaviour of a GPU-native numerical workload on a tightly integrated system-on-chip.
02 · Selected result
M1 execution with minimal CPU involvement.
Test platform
Apple M1 MacBook Air
Workload
~500 million odd-number candidates
Measured time
5.86 seconds
Observed CPU use
~2% during the reported run
Compute path
Metal GPU pipeline
These figures are implementation-specific measurements from the reported experimental run. They are not presented as a direct CPU benchmark, nor as a claim that P.A.T.H. outperforms specialised CPU prime-sieving tools in raw counting throughput. Their relevance is architectural: the test demonstrates substantial work being executed through the integrated GPU while the CPU remains lightly utilised.
Scope
Apple Silicon research
The project examines GPU offload, data-parallel execution and system behaviour on Apple Silicon rather than attempting to generalise the result to unrelated architectures.
Disclosure
Executive summary only
Internal kernel logic, optimisation techniques and implementation details remain withheld. The complete technical report is available for academic, engineering or research review on request.
03 · Access
Want the complete technical report?
The full paper, High-Performance GPU-Accelerated Prime Sieving for Cryptographic and HPC Applications, is available to academic, engineering and research reviewers on request.
Email to request accessPawan Sedara · 2026 · HexCode Research