hi folks,

due to my rather high electricity bill I was wondering if the conclusion in this paper ( https://www.rechenkraft.net/wiki/images/a/a5/Distributed_Computing_in_the_Cloud_-_BOINC_and_Amazons_Elastic_Compute_Cloud_EC2_ENGLISCH.pdf ) are still true today. Or in other words: Is it still cheaper to run Boinc on my PC rather than the cloud?
I do not know if that could be easily answered in general so if it helps providing an answer: I use a I5-2500k for my Boinc projects.
To me it seemed plausible that unused cloud computing power might be sold cheaper than what I have to pay extra for electricity when using Boinc. I'd be happy if anyone could tell me which would be the most economic way to support Boinc! :-)

For BOINC yes. But don't confuse BOINC with distributed computing in general. BOINC is one type of DC and the work that utilizes BOINC is best served using the public's computers. Not all work will work / work well with this kind of DC and internal or external clouds may be required.

External clouds like Amazon AWS MAY be more cost beneficial than an internal system for an organization. If you read up on LHC you'll see that they have they're own internal system they utilize for tasks as well as also running other work on BOINC. For them, they have the funding and capability to have their own cloud system (which actually is made up of more than one institute's systems).

I think if you upgrade (and sell existing hardware) every one or two years, and therefore get a large percentage of your hardware asset cost/value back due to less depreciation, and you keep getting the most efficient hardware (via the upgrades) - then the cloud vs own hardware is substantially on the side of self owning and running.

Personally I would upgrade to a i7-4790 or i7-4790K or later so you can make savings in terms of watts/core, as well as a gain in frequency (speed) and new instructions sets such as AVX2.

i5-2500K
4 logical cores @ 3.3 GHz (base) for 95w

i7-4790K
8 logical cores @ 4.0 GHz (base) for 88w

I think the cost of your electricity is the biggest part of this question. Some people have cheap electricity. Some do not. If you get yours cheaper than what it costs for cloud computing...well...the math is there. The second big portion to this is whether you are supporting a GPU capable project. GPU's can do a lot more work than CPU's in most cases. So, it is typically (but not always) more efficient to go with GPU's rather than purchasing CPU processing from the cloud. If the cloud is offering you GPU processing cheaply, then go back to the cost of electricity comparison. Upgrading hardware is tricky. Just because a chip has less power draw doesn't mean it is financially smart to jump on the latest greatest. Especially if you are using hardware that is 5 or more generations old. The resale value of your chips isn't going to be that much compared to the cost of the latest tech. Therefore a 7w TDP is not going to save you very much $$ at all. It would take several years to even come close to justifying the purchase. I agree you would get better extensions such as the AVX2 support. But that also drives up the power consumption a great deal and not necessarily in a more efficient kind of way. It will finish work units quicker when supported, but so will over clocking the CPU. Again, more electric cost to do that. Financially, it may benefit you to downclock your system in efforts to find what is most efficient. That way you can maximize you money and resources. Yes you would process less work over a given time frame, but you would be paying less for each completed work unit as well...

---