would these cheap units run boinc and be cost effective ?
MeLE PCG03 Mini PC Intel Atom Z3735F 2GB 32GB Windows 8.1 4K

They will run BOINC/Rosetta just fine, and if you're getting one anyway then it will definitely be productive on Rosetta. If you're considering buying one just for Rosetta then my guess is that they won't be cost effective in terms of output per dollar - they're just too slow. My guess is somewhere around 600 credits per day. You'd get far more from a second hand Q6600 on that front (2000+ cpd), although it would use considerably more power.

That atom might do quite well on a performance/watt basis. I expect you'd see a better return with something like a Haswell G3258 (2000+ cpd but using less power than the Q6600) on either metric.

---

I was wondering I just moved both machines from q6600 to 5460s 775 to 771 conversion then I bought a s6 samsung and it bench marks high than the xeons for little power

.. then I bought a s6 samsung and it bench marks high(er?) than the xeons for little power

It doesn't look like the samsung benchmarks higher, but it isn't *that far* behind the Xeons either (considering one of them is a "phone", ha!)...

Here are the benchmarks of your Samsung vs your Xeon in a graph:


.. With that said, the ARM cpu in your samsung is most definitely more power efficient, but some of the newer Broadwell and Skylake based CPUs should close that gap.

Happy crunching!

I'm pretty sure the benchmarks wouldn't look like that for Rosetta - the Intel chip would be waaaaay ahead because of the ARM chip's memory bottleneck. Again, performance per watt might be a different story.

Let's hope Skylake and Zen are worth waiting for...

---

read somewhere that not all tasks can be 'parallelized' efficiently due to dependency on a processing part which is a bottleneck (e.g.a sequential part)
http://en.wikipedia.org/wiki/Amdahl%27s_law

in cases where there is complex dependencies on a serial bottleneck (likely in complex real world scenarios), the maximum speedup is determined by that bottleneck. but i'd guess cpu architecture can make a difference in speeding up that bottleneck itself (e.g. super-pipelining / superscalar processing, high performance cache designs and large cache, hyper threading etc), & that could well be where the higher end haswell chips considerably outperforms simplier designs even if the simplier chips could do well in simple but parallel computations.

it is somewhat unfortunate that more simplier 'cores' do not automatically equate to fewer higher performance (likely architecturally complex e.g. superscalar) cpu cores. and today being near the end of moore's law, it is also putting a limit of how much complexity it is possible to squeeze into a single chip surface. i.e. it would be increasingly difficult or even impossible to make the next better 'haswell' that outperform all the previous generations

just 2 cents