3 ms·
So how many Pis can you cluster together to equal the performance of a standard i3/5/7 laptop (for non-GPU bound calculations)?
by ics 13y ago
So how many Pis can you cluster together to equal the performance of a standard i3/5/7 laptop (for non-GPU bound calculations)?
- tomrod 13y agoThat'd be a fun experiment. If only I had disposable income :). I wonder how much heat these things produce? Could they be used to make an efficient, small cluster?
- ics 13y agoMount an Arduino to measure the heat which bridges into your Mathematica program... world's most efficient micro-Mathematica cluster, world's least efficient radiator circuit.
- DanBC 13y agoPeople make clusters of them. Here are some links: 2 to 64 nodes http://www.southampton.ac.uk/~sjc/raspberrypi/pi_supercomputer_southampton.htm http://www.southampton.ac.uk/~sjc/raspberrypi/pi_supercomput... http://raspberrywebserver.com/raspberrypicluster/raspberry-pi-cluster.html http://raspberrywebserver.com/raspberrypicluster/raspberry-p... http://blogs.nvidia.com/blog/2013/07/19/secret-recipe-for-raspberry-pi-server-cluster-unleashed/ http://blogs.nvidia.com/blog/2013/07/19/secret-recipe-for-ra... Canonical had a 43 node Arm build cluster http://www.geek.com/chips/canonical-builds-a-42-core-arm-cluster-server-box-for-ubuntu-1390095/ http://www.geek.com/chips/canonical-builds-a-42-core-arm-clu... and there's a campaign for an RPi version http://www.indiegogo.com/projects/a-raspberry-pi-build-cluster-for-ubuntu http://www.indiegogo.com/projects/a-raspberry-pi-build-clust...
- noir_lord 13y agoYou could make a cluster with them (in fact for none-mathematica uses it has been done). Efficient not even remotely the performance per unit of a Raspberry Pi just makes it not worth it (note: it's worth it if you want to simulate multi-node programming, SMPP type stuff). For example a 32 node raspberry pi machine clocked 10.9Gflops on linpack[1], 32 * 35 is $1120 (excluding switches, cabling, power supplies which would add at least another $200 or so). A single Nvidia Titan clocks ~1.5 Teraflops in double precession mode for ~$1000 (a performance per dollar rate of 137 times better). According to wikipedia[2] a cheap Sempron 145 with 3 Geforce GTX760's at $1090 has a performance of 6.771 teraflops (so over 500 times the performance of the 32 node raspbery pi) 1 http://www.zdnet.com/build-your-own-supercomputer-out-of-raspberry-pi-boards-7000015831/ http://www.zdnet.com/build-your-own-supercomputer-out-of-ras... 2 http://en.wikipedia.org/wiki/FLOPS http://en.wikipedia.org/wiki/FLOPS