My computer cluster consists of four 2009, Mac Pros, 64-bit machines. I purchased them last year relatively cheaply – as they are older – and upgraded the firmware, allowing me to install 6-core X5680 Xeon CPUs (3.33GHz, 12MB cache) and 32GB of 1333MHz DDR-3 RAM in each, along with other touch-ups. One of these machines has 2 CPUs and 64GB RAM. They are again very capable machines.
While they may run twice as many threads as they possess cores, I’ve determined running the number of cores – 1 gives the optimal speed results since there appears to be competition for resources. (I suspect its L1 cache memory that is in short supply).
Using that rule, I have as many as 25 cores working on a given range of the number line.
Technically, they are connected by ethernet, but I’ve not pursued connecting them that way since their tasks are not directly interconnected. I use the KVM sitting on top to use a single monitor, keyboard and mouse.
The machines are, from right to left, J. K. Rowling,with 12 cores, is working on hexagon spin positions for (2.4 to 5.4)*10^16. Henrietta (Swan Leavitt) is working only one core – a single thread that started from 0 – but is now 4.3*10^15. Jocelyn (Bell Burnell) working on (5.4 to 6.9)*10^16 and finally, Vera (Rubin), is calculating in the range of (6.9 to 8.1)*10^16.
I’ll likely shut these machines down after I reach pi^34 or perhaps 10^17 — in the spring of 2017. More distant targets would be cost prohibitive and if I wish to see how if pi finally has a double, I’ll need to look for faster tools.