16 July 2017

16th of July

Farm status
Intel GPUs
All running Asteroids work.

Nvidia GPUs
Two AMD Ryzen's running Seti work.

Raspberry Pis
All running Einstein BRP4 work

Other news
The i3's and two new Ryzen's were dropped off at the computer shop last week. They have done the musical cases bit and installed the hardware, but they aren't open on weekends any more so I have to take some time off next week to go and collect them.

Asteroids ran out of work mid-week so I switched the Intel GPU machines to Einstein, except they were doing some server reconfiguring so I couldn't get any work from them either. I ended up running Seti for a day before picking up some Einstein gravity wave work. Since then Asteroids have put more work up. Asteroids is on 39M credits and Einstein is on 40.4M credits at the moment.

I don't currently have any plans for more Ryzen's, however I did have a brief thought of replacing the 8 i7-6700's with 4 x Ryzen 1700's. They'd use half the electricity (due to reduced numbers of machines). They are both rated at 65 watts. The i7's are of course faster than the Ryzen. I would also need to get a graphics card for each of the Ryzen's as they don't have built-in graphics.

Raspberry Pi news
I was contacted by an American user asking about running BOINC on a Raspberry Pi cluster that he had. The short answer was yes it can be done but its not cluster-aware so you have to run an instance of BOINC on each node. If the nodes are Pi2's or Pi3's then it recognises them as a quad-code CPU and will run 4 tasks at a time.

As for projects there is Asteroids, Seti and Einstein that I have run. Asteroids is very slow (they increased work unit size but didn't optimise their app for the Pi2 or Pi3). Seti is also slow but has been optimised to use whatever features the Pi has available and so has the Einstein app. There are others but haven't tried them. I tend to run Einstein on my Pi3's, but you need to select "run beta apps" on your project preferences on the Einstein website to get the optimised app.

On that note there was an article on the Rpi blog about OctaPi and how to create a Pi cluster. See https://www.raspberrypi.org/blog/octapi/ for the details. Now if we could get BOINC to run on the head node acting as the scheduler and use MPI (message passing interface) to communicate with the compute nodes...

No comments: