I've been searching the boards looking for an answer to this but haven't had any luck, so if someone can point me in the right direction, or can supply an answer here, or correct my misunderstandings of this issue, I offer my semi-eternal gratitude... :)

If I'm understanding this correctly, BOINC project WUs run both in single-precision 32-bit float and double-precision 64-bit float modes on CPUs, correct? If so, then why are GPU-only WUs required to run in 64-bit double-precision float modes only?

For most of the science projects which run on the BOINC platform, they don't. There may be one or two projects (I believe MilkyWay is one) which have chosen to demand full 64-bit hardware, and haven't programmed their science application to use the well-documented mathematical techniques for emulating 64-bit accuracy on 32-bit hardware, but you'd perhaps be better asking at whichever project is causing you to ask the question.

Alternative answer: these days, even a 32-bit CPU has an internal IEEE-compliant floating point maths coprocessor. GPUs, in general, don't.

I'm thinking I need to do some more reading and look a little closer in my research here... My interpretation of everything I've read so far on the various project boards has led me to believe that any project offering GPU WUs required DP-capable cards. As my computer-science nephew has repeatedly pointed out...it's a "PICNIC", Problem In Chair, Not In Computer. :)


Anyhoo, thanks so much for the clarification on this, I do appreciate it! It would be nice that the various projects had a more unified method of letting potential users know not only what applications they offer, but also what specific hardware and OS is required...everything is very scattershot...and it confuses people like me. :)