I have been toying with the idea of building a small computing cluster out of ARM-based single board computers (SBC), and while that looks like it may be infeasible for a number of reasons, I'm confused by one assertion that I've seen a number of times: that an Intel CPU will almost always outperform the ARM CPUs on SBCs by a huge margin.
One person claimed that even an i3 would be "orders of magnitude" faster than a whole cluster of SBCs, but that is surprising to me given the availability of quad-core (1.5GHz) and even octa-core (2GHz) CPUs on pretty cheap SBCs, with the beefiest i3 having only two cores at 4.1GHz each. I am aware that Intel chips have other advantages like reliability and power consumption, and that for many tasks, single-core speed is important, but those are different discussions.
So my question is: for a highly parallelizable task, is clock speed times number of cores a useful metric for comparing CPU speeds (approximately, of course)? Or is the ARM architecture fundamentally inferior to Intel, even when the marketing numbers say otherwise?
No comments:
Post a Comment