Farm status
Intel GPU machines are running Asteroids@home tasks.
Nvidia GPU machines off.
Raspberry Pi's and Parallella's are running Einstein@home tasks.
We had a couple of hot days during the week where it got up to 41 degrees C. Everything was off that day.
Hardware upgrades
Three (out of six) i7-3770 machines that make up the Intel GPU part of the farm have been updated to i7-6700's. The other three are planned for next weekend.
Intel GPU drivers for the HD Graphics 530 produce invalid results. I've tried 3 different drivers so far without any success. I have suspended crunching on the built-in GPU for the moment.
Due to Win10 rebooting machines at 3:30am (after it installs updates) I have gone back to Win7 on the new machines.
I tried Linux Mint 7.2 on one machine but it corrupts the video display and that was just trying to get it installed. I could put Debian or Ubuntu on them but then I would lose the ability to use the GPU - not a great loss at the moment because they don't work anyway. However I expect Intel will correct their drivers eventually.
BOINC testing
We got 7.6.15 this week. The main tweaks are around the task CPU and I/O priorities under windows. There have been reports of increased task failures, BOINC manager not updating and the client taking a long time to start up. A number of us have suggested backing the changes out. I've gone back to an earlier build (7.6.9).
Pi woes
I have also been battling with the Raspberry Pi's. The B+ refused to upgrade to Debian Stretch. I managed to brick it a couple of times and had to go back to the official Raspbian Jessie image.
Meanwhile the Pi2's have got onto Debian Stretch but are refusing to put the latest BOINC client on, complaining about unmet dependancies, even though they've got everything on the list installed.
4 comments:
Hi,
your Parallela's really use the epiphany chip ? Or only the ARM ? Still searching for a BOINC application supporting epiphany. Saw some post of you in parallela community, but unfortunaltely boinc & parallela is like cats & dogs ;-)
To answer Gerhard, no the Parallella's Epiphany chip is not used. Unfortunately they haven't managed to get fftw to work on it (doesn't have enough local memory). If they could get fftw or even parts of it to work then it could be useful.
For my money these days I'd buy Raspberry Pi's as they are fairly cheap, quad-core ARM and they use less power than the Parallella.
Thx. I bought 2 of them more that a year ago, planned to use them for a boinc project, didn't manage to set them up yet. Thought someone must have achieved any boinc client meanwhile. Disappointing ....
thx for your response.
BOINC isn't the problem. You can define your own co-processor to it via the cc_config file. The problem is with getting apps to use the Epiphany. That requires the apps to have coding changes. A lot of the apps (depending on the project) use the fftw library for fourier transforms. If fftw could use the Epiphany then it would be useful. This is where the Parallella falls down, they don't have the software to make use of their hardware.
Post a Comment