Earn Money with a Parallella Cluster

[AdrienChauvet]

Hi everyone,

I recently heard about the advantages of parallel computing compared to usual desktop PC's power.

I'm wondering if it's profitable to build a cluster of Parallella boards to sell the computing power on Gomez Peer or Slicify?

Has anyone tried it? Do you have some advice or experience to share?

Thanks for your time. Cheers from France,

Adrien.

[sebraa]

Check the specifications of the Epiphany chip.
Then, check the specifications of the Xilinx Zynq.
Then, think a while and answer for yourself.

[AdrienChauvet]

You're talking about this kind of board?


I don't know much about parallel computing.
All I know is that a friend of mine working in IT security heard about a team who broke some BCrypt encryption with a set of Parallella boards.
I guess it's a very specific use of Parallella boards. Anyway, it awakened my curiosity.

I found this link , I sent a mail to the authors.

[AdrienChauvet]

Here is a response of Brian Guarraci, who built parallella clusters.
I asked him if it was a good idea to built a cluster for earning money.

---

Hi Adrien,

Thanks for your email!

My observation is that Moore's law wins again... The nvidia Devbox is 28TFLOPs and costs about $10k. https://developer.nvidia.com/devbox

It's equivalent to about 1000 parallella boards, which would be 10x the cost... So I doubt it's going to be a good idea to invest in a parallella cluster for your stated purpose.

---

Now I have my answer. I hope it will help people on this forum too.

[piotr5]

I really am no expert there, but isn't comparing NVIDIA devbox with parallella like comparing apples with oranges? afaik parallella can only be compared to intel's knight's landing or other xeon-phi products. it is true that parallella is a bit old, a cheap amd apu can provide 4+32 cores clocked at 4 and 1 Ghz (although at much higher power consumption than a cluster of 3 parallellas). so either amd's or intel's chips can provide multiple cores, and you could lend their processing power to multiple users. I'm sure devbox plays in the same league. however, last I calculated amd's product has 48 Gigaflops mimd (as opposed to the 1 teraflop they calculated by adding the 512 simd cores to the equation). how many mimd cores and gp-gpu-wavefronts does devbox have? how many flops are possible using them exclusively? for example modern x86 processors can compute multiple floats simultanously with a single instruction, but this doesn't count as mimd...

in other words, I suggest distinguishing between simd flops and mimd flops in that the latter only counts operations happening in parallell if those operations are randomly distinct. i.e. it's the maximum flops you'll get if all your operations are impossible to run on simd. in pretty much all cases it's just min(number of instruction-queues , number of float operating queues) times frequency of operation-execution. for example amd's processor has 4 cores (at 4ghz) and enough ressources to perform their calculations. the gpu part of apu has 32 wavefronts at 1ghz each. a total of 48 Gflops mimd. now if there was a processor with madd taking 4 parameters, to effectively perform add and multiply for 2 distinct pairs of variables in parallell, and a possibility to perform 2 mul and 2 add instructions in a simd way, then this processor may double its mimd flops-count. same with intel-processors and their hyperthreading: there should be no bottle-neck if the flops happening in parallell are chosen badly...

needless to say I have no cs education at all, above words are just my half-informed opinion. I'm not even a hobbyist in computer science.