The Complexity of MP Benchmarking
* First hurdle, to classify the various forms of what is called MP?
* SoC Multiple heterogeneous processor cores (eg MPSoC)
* Uniprocessor with accelerators allowing synchronous offloading
* Homogeneous multiple processor ‘farms / seas / nets / etc’
* Multiple register-banked ‘threaded’ uniprocessors
* Symmetric multiprocessors (of various forms)
Conclusion, benchmarking is about the costs of running software, so need to focus around a
software model that encompasses MP - initial focus will be using the Symmetric Multiprocessors (SMP) software threading and shared data model. This also allows comparison with current uniprocessor solutions
* First hurdle, to classify the various forms of what is called MP?
* SoC Multiple heterogeneous processor cores (eg MPSoC)
* Uniprocessor with accelerators allowing synchronous offloading
* Homogeneous multiple processor ‘farms / seas / nets / etc’
* Multiple register-banked ‘threaded’ uniprocessors
* Symmetric multiprocessors (of various forms)
Conclusion, benchmarking is about the costs of running software, so need to focus around a
software model that encompasses MP - initial focus will be using the Symmetric Multiprocessors (SMP) software threading and shared data model. This also allows comparison with current uniprocessor solutions
Posts
No comments:
Post a Comment