15 October 2017

15th of October

Farm status
Intel GPUs
Running Einstein work overnight

Nvidia GPUs
Off most of the week

Raspberry Pis
Running Einstein work


Debian point release
After last weeks Debian 9.2 point release I have been trying different Nvidia drivers. The 375.82 drivers that the point release provided switch into low power mode after about 15 minutes on my GTX1060’s. I raised a bug with Debian and was told its an upstream (Nvidia) problem. There are a few versions of the 375.82 driver in the different repos but they all seem to have the problem. There is a 384.90 driver in experimental that doesn’t have the issue so I have installed it on all of the Nvidia GPU machines.

In addition to the Nvidia driver I have been reinstalling all the Intel GPU machines with the point release as it gets them off Jessie, now that the desktop issue has been resolved by an updated kernel. I had the Intel GPU machines running Einstein gravity wave work overnight.


BOINC testing
There is a BOINC client 7.8.3 out for testing. The BOINC Manager still has the jumping tasks tab/transfer tab issue, as well as the event log losing its date format when it scrolls. I never experienced the buffer overflow issue because I am not running the CPDN project at the moment. BOINC Manager issues aren’t much of a problem for me as I use BOINCtasks for managing the farm most of the time.

08 October 2017

8th of October

Farm status
Intel GPUs
Eight running Einstein gravity wave work

Nvidia GPUs
All off. Were running Seti overnight

Raspberry Pis
All ten running Einstein BRP4 work


Other farm news
I couldn't ssh into Pi #9 so I had to get it to finish off its work and reimage it. Strangely enough I could access it via BOINCtasks but ssh somehow got corrupted.

I hit the 50 million credits for Seti earlier in the week. The Intel GPU machines are now working on Einstein tasks.

The weather is going to be hot tomorrow so everything except the Raspberry Pis will be off again.


Stretch-backports on Rpi updated
While updating Pi #9 I followed the instructions I gave in September 2017 about adding Stretch-backports. They didn’t work. It seems the Debian gpg key signing has problems. To access backports requires the Wheezy and Jessie gpg keys (Debian 7 and 8). I have updated the instructions.


Debian Stretch 9.2
They did a point release to Stretch (9.2) yesterday. There is good news and bad news.

The good news is it has a kernel update to 4.9.0-4 (4.9.51) that allows me to upgrade the Intel GPU machines. ie It resolves the bug of having a blank desktop. I have already upgraded two of them.

The bad news is it has a bug in the Nvidia driver 375.82 which has now crippled my Nvidia GPU machines. I've raised bug 877971 with Debian about it. The driver switches into P8 (idle) mode after about 15-20 minutes of compute work. A reboot gets it back into P2 mode for another 15 minutes or so.

02 October 2017

1st of October

Farm status
Intel GPUs
Five running Seti work overnight. Three off.

Nvidia GPUs
Three running Seti work overnight. One off.

Raspberry Pis
All ten running Einstein BRP4 work


Debian donation
In my previous post I mentioned trying to help their build farm, specifically the armhf part. Attempts to get the Maxwell development boards they use have failed. The local Australian distributor didn’t respond to email or phone calls.

I have donated some money to the project. It will probably go into a general fund as there is no option to leave a comment even.


BOINC 7.8
I updated 4 of the Raspberry Pi’s by using the stretch-backports repo from Debian. See last months post about how to add stretch-backports to a Raspberry Pi.

Due to a couple of bugs which were introduced with 7.8.2 I won’t do the rest just yet. The BOINC release manager at the time didn’t include the fixes even though they were available. We’re hoping there will be a 7.8.3 to address the most pressing issue to do with cleaning out slot directories.


Other news
There are some details of the 8th generation i7’s being released. Looking at the i7-8700 compared to my i7-6700’s it would seem they are a little slower (3.2GHz compared to 3.4Ghz) but have two extra cores while sticking to the 65 watt TDP. There is also talk of a Ryzen 2 and 3 but no details yet.

As you can tell from the farm status the weather is starting to impact the farm. I haven’t quite reached 50 million credits for Seti yet but will resume the other projects once I reach it.

16 September 2017

16th of September

Farm Status
Intel GPUs
Five running Seti work, three off

Nvidia GPUs
Three running Seti work, one off

Raspberry Pis
All ten running Einstein work


Other news
There was a Bluetooth bug discovered that effects pretty much everything - Computers, phones car audio even. Microsoft patched Windows last month and the Linux guys had fixes out a couple of days ago. As a general rule I disable WiFi and Bluetooth in my computers when I install them. I prefer wired connections and not getting irradiated. Wired connections are also faster and reliable.


Debian donation
I was in contact with the Debian build farm team regarding the armhf architecture part of the farm. I was going to try and get a couple of machines together to donate to them. The seem to have a bunch of Marvell DB-78x00-BP based computers, but after they wanted them as rack mount that pretty much killed off the idea. Rack mounts are too big and heavy to ship around the world.

I did try and make contact with the local Australian Marvell sales representatives. According to the Marvell website they have one in Melbourne. I sent an email but haven't received any response from them. I may still make a financial donation to Debian instead of hardware so the build farm can be expanded. My reasoning is that they have a long backlog (it looks like 35 days) to build apps so the armhf versions come out much later than the other architectures.


Seti push
Weather-wise its been warm during the day so the Nvidia GPU machines have been running overnight only. I passed 48 million credits for Seti last week and decided everything can run Seti until I hit 50 million. This coincided with Asteroids having a week-long outage for maintenance work.