GPU computing with BOINC

E@H has several searches currently running. The only search that can use the processing power of a GPU is the Arecibo binary radio pulsar search (BRPS). There are two separate apps for this search, one that uses the CPU exclusively and a second that uses your GPU but also requires a significant amount of CPU support (1.0 GPUs + 0.2 CPUs). If you look back through your tasks list on the website, you will be able to see examples of both of these types of tasks. You will also be able to see a huge difference in the time taken to process a task when the CPU only is used (>2 days) as compared to the GPU + 0.2 CPU (2.5 hours). For that reason, you should consider changing your preferences to NOT accept CPU tasks for searches that have a GPU version of the app.

Apart from the BRPS, there are two other searches - for Gravitational Waves (GW) and for Gamma Ray Pulsars using the Fermi-LAT satellite data (FGRP). For both these searches, only a CPU can be used (at the moment), so your optimal settings would be to run BRPS on your GPU and GW and FGRP on your CPU cores.

A GPU core is nothing like a CPU core. Having 72 (or even much larger) computing units is simply a way of running a large number of threads of a single job in parallel and so drastically reducing the overall time taken. If you think about it, this is only going to work if elements of a job can be run in parallel. Some (parts of) jobs just can't be run this way.

So, your best option to make the best use of your machine in supporting E@H would be to change your preferences to stop the CPU only BRPS tasks. I'm not familiar with your CPU - core i7 X 920 @ 2.0GHz - but I'm guessing it's the precursor to the current Sandy Bridge i7 2635QM @ 2.0 GHz. Both are quad cores with HT - so 8 virtual cores. I have the 2635QM running in a Mac Mini with HT enabled and it is doing GW tasks in 30Ksecs and FGRP tasks in about 31Ksecs. For your machine the values are around 78Ksecs and 110Ksecs respectively, both of which seem to be unusually slow. It would be interesting to see what would happen to your times if you disabled HT and just ran with the 4 real cores. Maybe HT is giving you a disadvantage rather than the other way around.

This means that you shouldn't be getting tasks for CPU only applications. However you are getting such tasks so are you sure this is the current setting?? In any case, if you are trying to maximise support for the project, you really want this set to YES, seeing you have 14 computers attached which don't have a usable GPU. Even for the one machine with a usable GPU, you really would want the CPU cores to run CPU jobs, wouldn't you?

Fine.

All intervening settings are fine.

You must have had this as YES at some point because, in your list of tasks you do have BRPS tasks that were crunched with the CPU only app. However, as I mentioned previously, you are much better with this set at NO.

It would appear that you must have changed some preferences very recently, since a couple don't seem to agree with what your tasks list shows. The only one I would change is the very first one as indicated. These are all E@H specific preferences. You should also review your general computing preferences to make sure they are set appropriately. If you don't fully understand a particular setting, you should ask.