Hi,
I want to setup a 4 nodes (each one with 4 cores) Raspberry pi cluster and install quantum espresso on it for simulations. Does anyone know how good it works compare with core i7 PC computer for simulating nanostructures?
any test on it?
thanks
Hi, it is a funny project, but I am afraid that you will get what you will pay.
1 GB of RAM each and a connectivity of 10/100 MBPS Ethernet. I am not sure it is feasible. And please consider that, depending on the length of your simulations, you will overload the Rpi, with the risk of burining them.
Give a look to this paper for benchmark on legacy hardware: Article Molecular dynamics recipes for genome research
Hi,
based on my hand-on expiriance, you will in general get 10 up to 100x less performances on your Raspberry pi 2 or pi 3 cluster, but it will for shure consume 30-50x less power.
Also full Raspberry Pi setup will cost you less then just fraction of a i7 CPU price.
If you need fast results and have PC with i7 then it is better to stay with it, if you want to play and do research with/for low cost then Pi is better.
Best regards,
Intel Core i7 with overlocking up to +/- 4 GHz is very performing for many DFT calculations. Do you have enough RAM? Can you also overlocked your RAM to high frequencies?
thanks every one. I just want to spend my money on i7 CPU or by few Raspberry Pi 3 and make a cluster with them. I hear about it RPIcluster, so I check if it is really possible to use. Based on what you mentioned, till now I think to spend money on i7 CPU, but anyone use RPIcluster for DFT?
Hi Masoud,
one more thing, you can get much better performances if you chose Odroid XU4 Cluster over one with Rasberry Pi's.
I have done some research ( and hoby-projects) with them and find them to be realy fast for SoC's dev.board and you can use OpenCL 1.1 with it. They cost 2x more and consume 4x power but they have much better performance. Even Hardkernel ltd. is now is offering pre-made 4x cluster as Odroid-MC1 with 32 CPU cores (and 24 OpenCL 1.1 CU's)
Happy new year and best regards,
I have suggestion, if you really want to try to make cluster of SoC for HPC simulations that will have more serious computing performances why don't you take instead of Raspberry much faster Odroid N2 with 4GB RAM and ARM Cortex A73/53 Hexa core (or older, and now cheaper ODROID-XU4 Octa core ARM Cx A15/A7 CPU with 2GB) SoC's?
Base on my personal research and benchmarks you can expect 6-7x greater DP floating point performances from then, without consideration of faster networking and RAM speeds/size.
- Badges
- Science topic
- Similar topics
- Chemistry
- Nanotechnology
- Nanostructures