4 ms·
I wonder if this board is cost-effective for non-gpu datascience, especially in terms of power consumption. I have read a paper on several single-board compute
by bayesian_horse 10y ago
I wonder if this board is cost-effective for non-gpu datascience, especially in terms of power consumption.
I have read a paper on several single-board computers, saying that they were almost there, despite being ARM processors.
- alexandros 10y agomind linking the paper?
- bayesian_horse 10y agoCan't find it. Here is something similar: https://pdfs.semanticscholar.org/645a/a7fdcccb1eb2287b7094a24d625b304b559d.pdf https://pdfs.semanticscholar.org/645a/a7fdcccb1eb2287b7094a2... There are several actual papers on google scholar though.
- rcarmo 10y agoYou can get the Deep Belief stack running on a Pi GPU: https://petewarden.com/2014/06/09/deep-learning-on-the-raspberry-pi/ https://petewarden.com/2014/06/09/deep-learning-on-the-raspb... Aside from that, I don't think the performance will be acceptable. I've been running Spark and Jupyter on a 20-core, 5 node cluster and it's acceptable, but only useful for debugging (because stuff happens slower).
- dogma1138 10y agoYou might want to look at the Pi compute boards they come with stronger cups and don't have any of the crap you don't need.
- bayesian_horse 10y agoThe problem with the compute boards is that they don't have anything you may need, like something to talk with each other. If I wanted to build a cluster out of them, I'd probably have to break out the USB connections in some way, probably meaning a whole custom PCB or something... Given that there are more powerful Pi clones out there I don't get the appeal for this particular usecase.