End of 2010!

Well, the time has come for my Christmas break ... and all is quite, nothing stirring not even a 3GHz CPU. I have turned it all off until I get back on the 12th January.

I did reach the 12 million mark last night, so that is a nice way to end the year...

Tiles@Home is a dinosaur?

I have been rendering for about 24hrs now and then spent some time on their forums. It would appear as if the Tiles@Home project has past its prime and now the overhead of creating download files exceeds the actual crunching.
I will watch this project and support it in the future if they have something useful for users to do.

Tiles@Home

For a moment I thought I had found a perfect example of convergence! Anyone following my mobile phone blog will know that I have been using a Nokia phone to create GPX files to update and add to the maps on OpenStreetMaps. I was wondering through their Wiki and spotted "Tiles@Home" and with a name like that jumped to the conclusion that it was a BOINC project.
Sadly it isn't ... however I have installed the client on the IBM x-series server (Ubuntu i386). It is running on the one core and seems to be generating some useful map tiles for them. See stats: http://server.tah.openstreetmap.org/User/show/byid/1834/
The down side is that this will mean one or more processors less contributing to my BOINC total, but I do feel that it is a worthwhile cause.

Hot weather

Everyone told me that Tasmania was cold and miserable ... well today, the local weather station said 33.8'C at 12:42. I have turned off the nVidia GPU on my desktop machine and have throttled back the CPU usage to 85% on the server. I fear that renting an office without air conditioning may have been a bad mistake!

Still no internet

I have a weird problem! I am in an office block that shares a connection via a managed router. It turns out that the previous tenants had a server in the building and strange IT set-up. Now the ports that they used to use will not work with the managed router's DHCP set-up.
As a result, still no internet at the office besides a mobile broadband set-up via a mobile phone and my laptop.
This means that the office machines are not clocking up many hours per day and the BOINC contribution is way lower than it normally would be. I am hoping t get this sorted out with the QLD based ISP today ....

EDIT: 16:15 and they have sorted the problem out! A night of crunching ahead for the "farm" :-)

I'm back!

Spent the day setting up more or less the same collection of PC's in the new office. The Internet connection will only be up and running on Monday but then I will be back in the game!

No more crunching ... for a while

I am currently packing up to relocate interstate, so far two of the office machines have been turned off and packed up including their annoyingly large and heavy UPS. In the next three days the "farm" will be progressively turned off. I am currently 3356th in the world rankings but I guess that will sink quickly.
At this stage, I'm not sure what the future holds in the way of crunching but I'm sure I will be back in the not-too-distant future ... hopefully with some new and efficient hardware (I can dream, can't I?)


“I'm leaving because the weather is too good. I hate Sydney when it's not raining.” ... apologies to Groucho Marx

Ubuntu/Realtek work-around

I got both machines working this morning with this combination of events:

1. Disconnect all network devices and delete all connection settings in the Network Manager applet.
2. Turn off PC, disconnect power supply and wait for about 10 seconds to ensure that residual voltage on the motherboard is gone.
3. Put a 10/100Mb/s hub between the affected PC and the giganet hub
4. Reboot and choose "Auto eth" on the manager applet.

This isn't a permanent or elegant solution but it will get you up and running if you need to.

Realtek Gigabit Ethernet LAN

Just when I am really impressed with Ubuntu, I have run into a problem on two machines, one is running 10.04LTS and the other is on 10.10. Both machines have Realtek Gigabit Ethernet LAN cards built onto the motherboards and use the r8169 driver.
In both cases, the ethernet port is just not active. I spent some time on the forums and it appears to be a well documented problem but there are very few solutions. The strange thing is that the machine that now runs 10.10 used to run 10.04 and  eth0 worked fine at 1000Mb/s. After the upgrade ... nothing!
The machine that currently runs 10.04LTS dual boots with Windows 2000 and the NIC works fine under Windows, although only at 100Mb/s.
This is a pain as I am now using USB to ethernet converters and they are s-l-o-w

BOINC with ATI on x64 Ubuntu

I mentioned in a previous post that I could not get the version of BOINC available on Synaptic to recognize the ATI GPU on my 64-bit 10.04 Ubuntu machine. If I downloaded and installed the same version of BOINC from the BOINC website, it worked fine but had to be manually started. I upgraded to 10.10 and after doing some reading on the forums, the following sorted it out:

In /etc/default/boinc-client set the user to "root"
# The BOINC core client will be started with the permissions of this user.
BOINC_USER="root"

All sorted out now ... apart from the same problem I have with the CUDA cards that it occasionally still doesn't see the GPU at start-up. This can be corrected by adding a "sleep" line to the start-up script.

Edit: I have just noticed that the ATI card in question now only reports a peak of 50GFLOPS instead of 100GFLOPS as it did under the last installation ... weird!

Difference due to operating system

I have a PC in our workshop area that needs to run Windows 2000 to remain compatible with some older production equipment. I noticed that the OS doesn't even support SSE2 and hence this machine can't run any AQUA work units. I set up a dual boot with the current Ubuntu 64-bit desktop as the second OS. I am surprised at the difference in speeds this has highlighted for a single-core Celeron chip:

Windows 2000:
Measured floating point speed:    1512.17 million ops/sec
Measured integer speed:    2385.3 million ops/sec

Ubuntu 10.04 LTS:
Measured floating point speed    1309.06 million ops/sec
Measured integer speed    5868.99 million ops/sec

I will run a couple of CUDA work-units at some stage (probably Collatz as they appear to be fairly consistent) and see how the 8600GT GPU differs due to OS.

Edit: Under Ubuntu, on average Collatz CUDA work-units are 36% SLOWER than under Windows

10 Million ....

Finally reached the 10Million Cobblestone mark this morning, a couple of days earlier than I expected thanks to the recent abundance of AQUA Fokker-Planck work-units.

Collatz for CUDA 3 on x64 Ubuntu

As mentioned in my last post, I upgraded to the beta of Ubuntu 10.10 and in the process, the default drivers now support CUDA 3. I realized that the CUDA 2 application that I had copied across from the previous version of Ubuntu was crashing workunits with "directory not found" errors, so downloaded the CUDA 3.1 app from the "Power (Optimized) Applications" section of the Collatz website. I uncompressed the downloaded file, dumped the contents in the relevant directories and all is working fine! This stuff is getting really easy to use these days!

In terms of credit, the Collatz app produces almost the same credit/hour that the DNETC CUDA3 app does. I'm not sure but I think the new Collatz application is running the GP harder as I see the average temperature is now 70'C. I will watch that for a day or two and confirm the theory.

DNETC & CUDA on x64 Ubuntu ... update

Following on from my last post, I tried a few ideas to try and get the CUDA client working and after reading a few forum posts, realized that the problem was simply that my drivers were out of date and didn't support the cuda3.1 workunits.
I have upgraded to the beta 10.10 of Ubuntu and everything has worked straight out of the box. The BOINC client and the nVidia drivers that come with the distro appear to be fine for DNETC.

As a side note, I am pretty impressed with the Ubuntu 10.10 Maverick Meerkat release so far!

DNETC & CUDA on x64 Ubuntu ...

Well, everything has been pretty uneventful on the BOINC front for some time now. As a result, I got the urge to mess with something that is already working well (it's a major personality flaw, I will admit!).
I noticed DNETC seems to be generating a fair amount of of credit so decided to have a look at it. I am impressed that they support nVidia and ATI under 64-bit Linux, so decided to attach.
The "CPU only" machines, the 32-bit nVidia Windoze machine and the x64/ATI Ubuntu machine are all happy and generating some useful credit. The old 8600GT card is getting around 800 Cobblestones per CUDA work-unit which is pretty impressive.
The only problem is the 9800GT in the x64 Ubuntu box. It runs the CPU application happily but just gives a "process exited with code 195 (0xc3, -61)" error on all the CUDA work-units. I did some Google'ing and found some posts in their forum that suggest moving libcudart.so.2 into the project directory. That didn't work ... I also changed the file permissions in that directory as I was concerned that the output files weren't being created ... that didn't help either. I'm a bit lost at the moment as this machine happily runs CUDA work from Collatz, GPUGrid, SETI (optimized app) and PrimeGrid (but overheats).

PrimeGrid CUDA runs hot!

The PSU fan in my 64-bit Ubuntu work machine ran a bearing and I replaced it earlier in the week. The original fan had its leads soldered directly onto the PSU circuit board, as I was being lazy, I just didn't find a soldering iron and connect the new one in the same manner but rather plugged it into the chassis fan connector on the motherboard. I installed a "widget" to monitor the temperatures inside the case as I'm not sure if the M/B steps the case fan-speed down or not...
I received GPU temperature warnings for about 30 minutes at a time and then everything settled down only for the process to be repeated an hour or two later. I eventually tracked the problem down. It only happens when a "Proth Prime Search (Sieve) 1.21 (cuda23)" workunit is running. The GT9800 runs at around 63'C for SETI & Collatz. As I type this, the GPU is reporting 72'C on a PrimeGrid workunit.
I'm a bit concerned about the life-span of the GPU at these temperatures (as well as the irritating fan noise), so I have changed my preferences to not accept GPU work from PrimeGrid.

Nothing new...

Things have been fairly uneventful on the BOINC front as AQUA don't have any FP work-units. My average has fallen to a point where I am just keeping my position in the world rankings and occasionally going up a place or two.

On a non-BOINC related front, I bought a 1TB Seagate "Expansion" drive which is a USB2 device that has its own power-supply. I plugged it into my home PC to move some backup files around and to my horror, the brand new drive started emitting some troubling "click ... click" sounds followed by "Device Not Found". I was on the point of taking it back to the vendor when I had a quick look in Seagate's forums to see if the issue is wide-spread. Turns out that it is a problem especially with the 1.5TB drives.

A user on the forum stated that he had overcome the problem by using a higher grade USB cable with a ferrite core on it ... I thought that was worth a try and used a cable off a Garmin GPS. What do you know? That solved the problem for me! I won't trust anything too important to that drive but at the moment it is doing service on a digital video recorder and seems to be working OK

Update

I am admitting defeat! Despite trying for a couple of days, I can not get both the PCI-e and the onboard GPU to work together on the JW-A785GM-Fusion motherboard. They each work fine on their own but the best I can achieve is a hung GRUB when they are both enabled. I have even tried "long-shots" like starting the kernel with the "nopat" parameter.
On the topic of the JW-A785GM-Fusion motherboard, I am not very impressed with this item. The user-manual actually refers to a completely different board layout, their website contains 3 versions of the BIOS are either "Beta" or are not very well documented and in normal operation, the board hangs on POST about 1 out of every 3 start ups. Generally I would say to avoid this product for any application.

On a happier note, I have passed the 7 million credit mark today! :-)

It works! ... well, in principle ...

Ok, so if you are using 64-bit Ubuntu 10.04 (Lucid) and want to run Collatz on an ATI card, here is how it can be done. While it is really not the most elegant approach, it does the job.

1. Don't bother with downloading Catalyst from the AMD site, just go System > Administration > Hardware Drivers. After a search, there should be a "ATI/AMD proprietary FGLRX graphics driver". It turns out that this is Catalyst 10.4 anyway and gets rid of the issues with using .run files etc. As is also in the restricted drivers repository you will automatically get updates (time will tell if this is a good thing...)
2. Download BOINC in .sh format from Berkeley. Do not use GetDeb or the repository version, they will not work and this is what took me so long to get around. The GetDeb version should have worked as it is version "6.10.36". I used 6.10.17 which is the current recommended version. The only problem is that this worked in the middle of trying a large number of hit-and-miss options, so I have the installation on my Desktop and it requires root privileges. You may want to be a little more careful where you run the .sh file...
   
3. The only work available for 64-bit Linux/ATI is from Collatz and it is a manual install from their Power (Optimized) Application section. Read the good included "Readme" in the package if you are not familiar with where to put the library files for optimized GPU apps etc.

I still have a few things to do before this installation is complete, the most obvious one is to get the second (on-board) GPU up and running too. At this point, I think I might have a BIOS issue as the Catalyst Control Center reports it as an "Unknown Adapter"

Progress? ... Um, no...

Well, after 2 evenings of playing around with the new motherboard and ATI card, I have the appropriate ATI Catalyst 10.4 installed along with Boinc 6.10.36 and the Collatz 64-bit ATI application (along with its library files) ... and thats where things end. I can't see get Boinc to recognize either the PCI-e card or the onboard GPU. I have tried all the tricks I gathered when installing the nVidia cards and have added "boinc" to the "video" group etc
Reading a couple of forum posts I am getting the distinct impression that this might be as a result of an incompatibility between my hardware and the specific ATI driver or Linux kernel that I currently have installed. I may try older versions over the weekend to see if anything works.

Let the fun begin...

My new JW-A785GM-Fusion motherboard and XFX HD4550 graphics card arrived today. There seem to be a few options to play around with, the first is the motherboards ability to support "Hybrid Crossfire" between the built-in GPU and the card in the PCI-E slot. I assume that in BOINC terms this means both cards will land up working on a single work unit? While kinder in terms of memory usage, I think it will slow everything down to the lowest common denominator? Time will tell!
The other slightly alarming feature is the "128MB DDR3 Sideport" memory. While I understand that this makes the graphics card quicker as the system memory is not addressed every time, I just wonder if BOINC will see more than 128MB on the card?
I am planning to install Ubuntu's new 10.04LTS tonight along with the 10.4 ATI Catalyst drivers. I'm not convinced I will get it all up and working with Collatz work in one evening but as usual I live in my own little world of misguided optimism...

Hardware problems again...

A few months ago, I was unhappy to discover that my heat-pipe cooled 9400GT GPU was no longer visible to BOINC and as a result was not processing data-units. I assumed (based on the trouble I have had with nVidia cards in the past) that the graphics card was to blame.
When trying to boot the Athlon X2 this morning, it couldn't see the hard-drive ... panic set in as the back-up drive was cannibalized for a server a few weeks back and as a result there is a huge collection of photos that were not backed up anywhere.
I then had a revelation that it may be the mother-board, so tried another hard-drive and that didn't auto-detect either.
I brought the original drive into the office this morning and thankfully it works fine on another machine (and has now been backed up!)
I ordered a JW motherboard from my usual supplier and as this has a Radon ATI chipset have decided to give that a try. I also ordered a 4550 to go in the PCI-E slot. I know this isn't the ultimate in high speed cards but it is dirt cheap and has a low power consumption which is important for a computer that must survive summers without air-conditioning.
The only problem now is that the X2 will not be crunching AQUA FP units until the parts arrive and I find the appropriate Ubuntu drivers (I'm hoping this will not be too much of an issue...)

It was nice while it lasted...

Over the 5 days when my "farm" was able to get Fokker-Planck work from Aqua@home, I chalked up 147,415 credits. Not bad for one quad, three dual cores and a single core...

Aqua@Home has more Fokker-Planck work units!!

Aqua@home has not had any Fokker-Planck work for a couple of weeks and as a result, my daily average has dropped to around the 10,000 cobblestone mark. This has been exacerbated by Milkyway having periods of no work either. When the last lot of FP work was available, I was seeing around 45,000 per day...
At around 14:00 today, 49,000 work units were loaded onto the AQUA server. I have the Linux machines at the office running this weekend to try and "catch up". The only problem is that the fastest server is busy with an AQUA MT work unit that has 154 hours still left on it. This means that it will probably miss out on the FP work units.
The older work stations will also miss out as FP requires SSE2 or better.

I'm back ...

I've been in the UK and the Netherlands for the last two weeks and the two AMD X2's have been turned off, including the one with the NVidia GT9800. As a result, my daily totals have dropped somewhat. Everything is up and running now but we are into the Easter long weekend so machines will shut down until Tuesday. I should be up to full steam again next week.

Weird Seti@home error

Now that AQUA seems to be out of Fokker-Planck work units, I decided to attach the Intel quad to SETI and run some Astropulse work using the Lunatics SSE3 optimized client. The weird thing is when I attach to the project, all is as normal but when I go on their website to change the default location, I just get a blank screen with "Couldn't find computer" when I click on the "details" link.
The problem seems to be the allocated Host ID (in this case 5328134) ... the strange thing is that no machines around this number seem to exist either (5328133/5328135 etc). When I spotted this yesterday, I thought it was probably something to do with a delay in details getting to the database but now we are 24hrs later and the problem persists.
There doesn't seem to be anything in the forums related to this. Guess I will wait and see if it comes right by itself ... not too concerned as my XP laptop has been waiting for Astropulse v5 work units for over a week now anyway.

EDIT: Ok, 3pm now and the computer in question is showing on the SETI@Home website and I am able to change locations etc. Still no Astropulse v5 work units however.

Top 100 in Australia

Another milestone (of sorts!) for me, finally in the top 100 crunchers (combined projects) in Australia as of this morning. Actually in position #99!

1 Million in Milkyway

Ok, I know its not a huge milestone but I was impressed to see the 1,000,000 mark with Milkyway. This was achieved using the 32-bit optimized client of the old "zslip.com" website. It still still seems to be at least 30% quicker than the standard 64-bit Linux client.

5 Million!

Sunday afternoon and I see I have just passed the 5,000,000 mark! A couple of Fokker-Planck work-units from AQUA did speed things up as I was earning over 300 Cobblestones for around 50 minutes work on most my CPUs. These were "multi-thread" work units, meaning that they use all the available cores for just one work-unit ... still over 330 Cobblestones/hour was very generous for ageing X2 dual core AMDs. They have now released an updated application that delivers notably less credit.

The 9400GT does appear to have given up, the BIOS doesn't even recognise it as a device. As I was about to bin it (have no plan to argue about warranty after the 9800GT saga), it did occur to me that the motherboard's PCI-E slot may have died. At some stage I will try the card in a different machine.

Another GPU bites the dust!

There was a heat-pipe cooled Leadtek Nvidia 9400GT in my home computer, this machine is on about 5 hours per day mainly in the evenings. I started getting "[error] Missing a CUDA coprocessor" messages yesterday. It seems that the card has died. There goes my theory that a GPU that doesn't rely on a cooling fan will last longer...

Still ticking along...

Nothing to report at the moment, the very warm weather here in Sydney is making its mark on my credit as I am turning everything off over the weekends. Temperatures over 35'C with no air-con will not be a good environment for Nvidia cards.

On closer inspection, I see that I have just passed 1,000,000 credits for Collatz Conjecture ... not bad as I only signed up on the 8th October 2009. The joys of a GPU!

I'm back!

All but one P4 were turned off while I was away on holiday and even that machine seems to have fallen over just after Christmas (probably a power glitch at the office). I will be turning most the machines on this morning and will start the long process of catching up on two weeks of almost no credit.
For once, my luck seems to be holding, most the projects are up and running when I booted my home PC earlier.