I have taken the splash into the world of blockchains, multicore boards, and clustering.

I am wondering if anyone has recompiled BOINC using OpenCL?

Thanks

BOINC is a CPU-based 'command and control' program. Although OpenCL is available for CPUs, I think it's generally accepted that the added layers of complexity and runtime support would make it less efficient than the current native C++ code.

Where OpenCL comes into its own is in programming GPUs, as you've correctly indicated by choosing this message board area for your question. GPU programming is the role of the science projects that run under BOINC, not BOINC itself.

Parallellas were discussed when (and before) they were first launched, but IIRC the conclusion was that they could address too little memory per processing core for serious scientific processing to be attempted. Your best source for technical information and discussion is probably the thread PARALLELLA, RASPBERRY PI, FPGA & ALL THAT STUFF on the Einstein@Home message board - but be warned, there are over 900 posts spread over more than 5 years.

BOINC is primarily a task scheduler so it doesn’t actually do the computations, they’re done by the project developed apps. You can define the Epiphany as a coprocessor for BOINC so that it can schedule work to run on it. Unfortunately there’re aren’t any apps that I am aware of that managed to use it.

One of the Einstein volunteers tried using the Epiphany chip for their apps (as a coprocessor). It turned out the Epiphany didn’t have enough local memory to compute their FFT’s. See the message thread that Richard has linked to.

While it does have an OpenCL layer for the Epiphany it is only a subset of the standard so couldn’t run existing OpenCL code.

---

MarkJ