26 December 2010

Boxing day 2010

Milestones
With Seti now handing out work I managed to pass 10 million. Below is my Setificate :)




Weather
Well its been hot and humid. Until tonight it was dry, but then a thunderstorm came through.

That all adds up to not much running due to the heat. Maul has been going most of the week and today I also fired up Obi-Wan who is concentrating on Seti work.

19 December 2010

19th of December

Milestones
I hit 30 million for GPUgrid this week. My certificate is below.




Adventure 2800
I finally managed to get an answer out of LucidLogix. The price is very expensive for an expansion board with 11 PCIe slots single-spaced. All the decent graphics cards are double width so that would leave 6 slots usable. I still haven't got a price for the Adventure 2500, which has 4 PCIe slots double-spaced.

It would be more cost-effective to use a 7 slot PCIe motherboard. There are a few different brands of these available for around $400 to $500. I was thinking I might be able to stuff one of these with GT240's and it would make a great Seti cuda cruncher, however I think Nvidia has hobbled the GeForce drivers so you can only have up to 4 cards per machine, to make people buy the Tesla's


Climate Prediction
Well I finally finished off the last 3 work units. One errored out with 8 hours to go (after running for 200+ hours) and the other two managed to complete successfully.

I have downloaded a couple more Famous wu. I also got allocated 6 of the newer eu work units but they all failed to download. It was complaining about being unable to find a file.


Project news - Einstein
They are still only handing out gravity wave data. They ran out of binary pulsar data from Arecibo a couple of weeks ago.

They have managed to secure data from the Parkes observatory which is requiring a bit of development to format so they can use the same binary pulsar programs we currently use.


Project news - Seti
Maul currently has a few hundred multi-beam tasks to get through. Apparently they can't split the data from the "tapes" fast enough to keep up with demand.

They are also doing a general funding drive. They are getting additional data from the Green Bank observatory that will require some development work. There was some mention they might be able to get data from Parkes a while ago.

15 December 2010

GTX570 update

After yesterdays run showing that you can fry eggs on these things I decided to try Evga Precision. This is the default settings, with the fan speed on Auto. As you can see the GTX570 could double as a cooking appliance.


After taking this screen shot I decided to ramp the fan speed up to 75%, which brought the temperature down. I cut the fan speed to 70% and it seems to be holding the GPU temp at 77 degrees C. The room temp is 28 degrees next to the computer

This is the factory standard model. I didn't get an overclocked GTX570 because of the heat, so I hate to think how they perform.

13 December 2010

GTX570

The GTX570 arrived and has been installed in Spock.

It has run a GPUgrid work unit so far. The GTX570 completed a HIVPR_unbound work unit in 14,019 seconds compared to a GTX295 that did it in 27,728 seconds. The GTX295 can run two work units at a time. They used a different GPUgrid app (6.13/cuda 3.1 on the GTX570, 6.12/cuda 3.0 on the GTX295), so it may be more like comparing apples with oranges.

GPU temps got up to 85 degrees which was rather warm.

The GTX295 is still the best value for number crunching at this point, although that may change when a cuda 3.2 GPUgrid app becomes available.


GPU-Z details



GPU-Z Sensors tab with it under load



Installed in Spock



Another shot looking up



11 December 2010

11th of December

Milestones
None this week.


GTX570 ordered
I've ordered a GTX570 to see how it performs. It should arrive next week.


Project news - Seti
They are back up and running. Maul picked up a fair bit of cuda work and is busily working its way through it. Its got enough for a couple of days.


Maul
As mentioned above its got a heap of cuda work from Seti. Its also got a heap of Einstein cpu work to get through. Fortunately its finished all the Seti cpu work, so all cores can concentrate on clearing the backlog. It doesn't help that its usually off during the day due to the heat.

05 December 2010

5th of December

Milestones
Nothing new this week


BOINC testing
We're up to 6.12.8 this week. So far no issues. They are looking at this one being released soon.


Project news - Seti
New servers have been switched over, except for the science database one (they were copying it from the old machine just before the weekend). We got some resends of old timed-out work on Friday. I managed to get 2 work units, which were promptly done and returned.


Computing focus
Currently Einstein is the most reliable project.

Seti will come back on-line next week so I will be able to do some work on a couple of machines for it.

Climate Prediction I am trying to finish off the 3 remaining work units on 2 machines which usually get switched off during the day (due to the weather).

Orbit doesn't have any work currently.

GPUgrid have issues with their app that cause the machines to crash, so i'm not doing any work for them at the moment. Things aren't so good at the moment with their cuda 3.1 app crashing and their 10 hour work units. I might put old drivers on the 2 machines with GTX295's to see if that resolves their app issue.

28 November 2010

28th of November

Milestones
Nothing this week.


Future plans
At the moment I will look at getting a GTX570 to see how they go. If the GTX570 seems to be reasonable (price/performance) then I will look at one for each of the i7's except Maul. They will replace whatever each one has (2 x GTX295, 1 x GTX275 and 1 x GTX460). The GTX460 will get put into one of the quaddies.


Project news - GPUgrid
They have updated their science application program to cuda 3.1. However it crashes on my machines with multi-gpu cards (ie Kirk and Sulu). They both have GTX295's installed. I have ceased running GPUgrid on them for the time being. I suspect this is caused by a bug in cuda 3.1 where it gets confused as to which device is running the app. They are in the process of creating a cuda 3.2 app so this may resolve the issue, but its not ready yet.

They are also seeking donations so they can get a GTX580. They are available from their supplier (in Spain) from the 1st of December. I made a donation to help them get one.


Project news - Seti
Well their two new servers are in. One has been fired up as the BOINC database server and the other is set to become the science database server. They are expecting to copy the science database over to the new machine next week.


Weather
Its been fairly warm and humid most of the week. As a result all the machine have been switched off during the day. This of course means not much work is getting done at the moment.

Add to this some ClimatePrediction work that takes 250 hours a work unit and it makes things a bit more complicated. I have been favouring the machines with CPDN tasks to try and get them all finished.

20 November 2010

20th of November

Milestones
Nothing this week either.


Cuda 3.2 released
Nvidia finally released the cuda 3.2 toolkit. Its been in release candidate mode for a while. Now its available to the public.

The Seti optimisers have created a test app to try some different algorithms using the cuda 3.2 toolkit. As I have a few different types of graphics cards I ran it on all of them and reported the results. Hopefully this will lead to further optimisation of the science apps.


GPUgrid switching to cuda 3.2
The GPUgrid guys are also in the process of getting apps compiled using the cuda 3.2 toolkit. At the moment their cuda 3.1 app fails on my GTX295 rigs. I suspect this is the cuda bug where it gets confused as to which device is running the app, as it works fine on the single gpu machines. That means I can only run GPUgrid work on Spock (GTX460) and Chekov (GTX275) for the time being.


GTX570
No its not out yet. Supposedly coming in December. Its reportedly 480 cuda cores and based on the GF110 chip, which the GTX580 uses. Should be good for crunching if the price is right. I will look at getting one once the hype of a new release dies down and stock becomes readily available.


Maul HDD upgrade
I replaced the hard disk in Maul today. Not a big deal to install it, but rather the hassle of reinstalling Windows 7 the way I had it. Anyway its all done, so its got a little bit more disk space now. The old disk was a 320Gb and the new one has 500Gb, however I don't use much disk space on the crunchers.


Seti servers arrived
The 2 machines that we did the fund raisers for have arrived. They have got 1 up and running as a replica database and will see how it runs for the next couple of days. I also did a top-up donation so that they can get some additional hard disks.

13 November 2010

13th of November

Milestones
None reached. Things have been rather slow as most of the machines are off. What little crunching is taking place is on Einstein and ClimatePrediction as they have longish deadlines.


BOINC testing
We got up to 6.12.6 this week. They made a few changes to job scheduling in 6.12.5 around multi-threaded apps (MT). They have refined it a bit with this latest one. I've got it installed on Maul at the moment. Its working fine, but then I don't run MT apps.


GTX580 released
Well Nvidia released the GTX580. Its got 512 shaders (what the GTX480 was supposed to have) and runs somewhat cooler. They cost $680 (AUD) at the moment, so I won't be buying one anytime soon.


Weather
It turned hot this last week, so all the machines have been off during the day, with one or two running overnight.

Last week we had a thunderstorm blowing rain in through the loft window. It was only open 2 inches and its in the roof line (ie it slopes down at 35 degrees) so the rain was blowing up and in. The two computers close to the window were getting water on them before I got the window closed. I then had to shut them all off because of the heat.

30 October 2010

30th of October

Milestones
I reached 5 million for Einstein. Below is my certificate



Nvidia drivers
They released 260.89 last week and 260.99 this week. Both as WHQL. Unfortunately they down-clock on non-Fermi cards after 30 mins. As a result i've gone back to the 258.69 drivers for all my non-Fermi rigs (which is pretty much all of them).


Project news - GPUgrid
They are up and running with a bunch of longer running work units. They take around 10 hours each. Given the short deadlines (5 days) the lower-end graphics cards have problems completing these in the timeframe.

They are planning on introducing a cuda 3.2 based app, when Nvidia release the developer toolkit (its under a NDA at the moment). They will drop the cuda 3.0 and 3.1 apps. They will contine with a cuda 2.3 based app as well for those running older cards/drivers.


Project news - Seti
Well they are up in a very limited capacity now. No new work, but people can upload completed work and report it.

They've ordered 2 HP servers to replace the couple of faulty database machines. Expected delivery is 2 weeks, and then another 2 weeks to migrate stuff onto them.


Maul
Last week there was an annoying clicking sound coming from Maul about every 6 seconds. A couple of days later I got an error reading one of the data files. So putting 2 and 2 together I decided the hard disk drive might be at fault. I run some disk diagnostics and it doesn't find anything wrong. I decided to order a replacement HDD as I don't have any spare SATA drives. The one in it is from a few years back, in fact its the only original bit of hardware left in it now.

The next day i'm standing at the KVM and look over at Maul and notice the fan on the front isn't turning and the click sound comes each time as it tries to turn. I power it off and remove a power cable that was poking into the 20cm fan and preventing it from turning. Power up and off we go without the clicking sound.

I now have a few spare drives (500Gb WD Caviar Blue). I'll probably remove the old one and install the new one in about a week. I don't want to change the hardware too much in one go so that Windows activation doesn't think its got a new machine.

24 October 2010

24th of October

Milestones
Nothing reached this week


BOINC testing
We're up to 6.12.4 this week. It fixed a couple of things and introduced another annoying problem. It doesn't look like its ready for release yet, but not too far off it.


Maul returns
As mentioned last week Maul went back to the shop for a CPU upgrade. Its now back and crunching away.


Project news - Climate Prediction
They have space problems. They filled up one of their servers. They have ordered an additional storage system which is due to be installed this coming week.

They also have a bunch of new models being run as a continuation of one of their earlier experiments.


Project news - Seti
Well their BOINC database server was pronounced dead. Another one has been ordered. As a temporary measure they have turned what was the replica database machine into the primary. Hopefully the new server will arrive quickly and things can get back to normal.

With the rearrangements they aren't creating work (although there was some ready to send anyway) and they have placed restrictions on how much work one can get, Network connections are suprisingly quick as they backoff off for 5 minues after each scheduler request.

16 October 2010

16th of October

Milestones
Not really a milestone but I passed 45 million for all projects.


Maul in the shop
Yep, he's back in the shop. This time for a CPU upgrade. Going from a Core i7-920 to a i7-950. I'd really like a 970 but they are around $1300.00 (AUD) just for the chip. I'll pick it up next weekend barring any problems.


BOINC testing
We're up to 6.12.2 this morning. I have it installed on 1 machine at the moment and it seems fine so far.

I have been running 6.11.9 for a while on all the machines with some strange behaviour when running GPUgrid TONI_KKi4 tasks. As these are new tasks I think its just the tasks causing the problems and not BOINC. Others are also having problems with these tasks.


Seti donation
Last week they were running a donation drive as it appeared they might need to replace a 2nd server. I made a donation then. However someone else came along with a higher amount early this week. Not to be outdone I decided to match it. Anyway it looks like they have the extra funds if they need them.

The server survived the onslaught with the network maxxed out all week until it was time for this weeks outage. Its back up at the moment and as usual the network it maxxed out again and its hard getting the work. Its allocated to the machine, but because of the overloaded network it simply cannot download.


Temperature
We've had a warm week so most of the machines have been off. Yesterday it cooled down so everything got fired up (yes, even Maul) and they are crunching away with the cache set small in case the warmer weather returns.


Windows updates
Microsoft have been busy plugging security holes. 49Mb per Win7 machine of updates. I've been trying to stagger these so as to not blow my internet download limit and not max-out my network connection.

10 October 2010

10th of October

Milestones
Nothing reached this week, although I did pass 28 million for GPUgrid.


Project news - Climate Prediction
Well like all the other projects it seems its their turn to run out of disk space. They have managed to fix their initial space problem and got uploads going. I've completed a few Famous climate models but can't report them until they fully resolve the issue.


Project news - Orbit
They have work but some issue with the scheduler has prevented us getting it. Unfortunately Pasquale was out of town for a presentation of the results and wasn't able to fix it remotely.


Project news - Seti
Seems another of their servers is playing up, possibly caused by their air conditioner failure 3 weeks ago. One of the regular volunteers, msattler, started another fund raising drive. We are hoping the money from this one and the previous one will be more than enough to get two servers instead of one.

Meanwhile Seti came up on Saturday (Sydney time) and the downloads have been maxed out for the entire weekend. Getting work is almost impossible at the moment. I managed to report last weeks work and got a few work units but have finished them all off.


BOINC testing
On Friday Kirk started doing crash dumps when running GPUgrid work. I reinstalled the video driver (which hadn't changed in 2 months) and that seemed to get it going. On Saturday Sulu started doing the same. I'm not too sure if its the new work units from GPUgrid (TONI_KKi4) or a bug with BOINC or Windows.

While poking around I noticed that Control panel was telling me I had an old version of BOINC installed from August. It would appear that it shows the version installed as an administrator and not the version installed as a standard user (which asks for an admin account/password). So I gave it admin rights, uninstalled and reinstalled. That gets Windows thinking it was running the right version.

I also uninstalled and reinstalled the video drivers as they too appeared to be an old version. Control panel was once again showing an old version and Windows update keeps offering me an updated video driver, but i'm already running the latest release version from nVidia.

03 October 2010

3rd of October

Milestones
None this week.


BOINC testing
Another couple of releases this week for testing. Current version is 6.11.9 now. I've got it installed on a few machines.


Project news - Einstein
They had a couple of glitches during the week but fixed them quickly. They have a problem with recent ABP work units not validating which they are working on. They aren't sure yet if they will have to redo the work or not.


Project news - GPUgrid
Their turn this week to run out of space on the upload server. They fixed it fairly quickly and we're back in business.


Project news - Orbit
We are just finishing off this weeks run. No new tasks available except resends. All machines except Maul have finished off their work. They are going to take a break for a week as they have a presentation of their results next week.


Project news - Seti
Seti spent all of last week off the air. They are now back online and i've managed to upload and report all my work from 2 weeks ago (along with the other 1 million or so participants). I've managed to get some work out of it since then as well.

26 September 2010

26th of September

Milestones
Not really a milestone, but I hit 27 million for GPUgrid.


Project news - Orbit
After last weekends outage they came back up around midnight Sunday (Sydney time). Since then we have finished off last weeks run, did another one during the week and judging by the number of work units ready to send we're on the 3rd run.


Project news - Seti
They went off the air due to filling up disk space on their upload server. They are in the process of rearranging servers, but will have to bring forward plans to next week. That leaves lots of work on peoples machines that we can't upload and report. There is no new work available to download.

The fund raiser was for a new science database server. They think they will be able to order it in about a weeks time and then a couple of weeks for delivery. After that probably another week or so for them to set it up the way they want.


Weather
Starting to warm up now we are in spring. Currently the loft is 28 degrees. As a result I have to cut back on processing.


BIOS updates
As I mentioned a couple of posts back I updated the BIOS on Chekov and Maul a couple of weeks ago. I ran around and did the rest of the machines this weekend to bring them all up to the same version.

19 September 2010

19th of September

Milestones
Nothing new this week.


Project news - Orbit
The completed their 1st run of 300,000 work units last week. The 2nd run of another 300,000 work units is underway.

Their server has crashed and the web site is down. For the 2nd run they are issuing the tasks to 2 computers and have a 3 day deadline (now likely to be missed). If not returned in time they will get reissued to another computer.


Project news - Seti
Their servers were off most of the week due to an air conditioner failure. They are now back online.

They have settled on a replacement science server that the fund raiser was for. I have put in $500 for them to get additional disk capacity (2Tb drives instead of 1Tb).


Nvidia beta drivers
They've been quiet for a while but have released new (beta) drivers. Version 260.63 supports cuda 3.2. They also have reworked their installer so that it can now do a clean install and not install certain components. Initial reports say it may be slightly quicker when running existing (cuda 2.3, 3.0 and 3.1) apps.


Cuda 3.2
As mentioned above the new drivers provide cuda 3.2 support. There is a non-disclosure agreement covering it, however rumours suggest the FFT libraries have been reworked for fermi-chips and provide substantial speed improvements.


GTX460
Last week the 2nd one was playing up. When I power up the machine it would tell me there wasn't a VGA card installed. Off to the shop it went. It of course worked for them, so I brought it back home today and will reinstall it later in the week.

14 September 2010

14th of September

Milestones
Passed 100,000 for Orbit.


Orbit
The farm has been concentrating on Orbit seeing as they had a small run of 300,000 work units, which seems to have now been completed. I was also running GPUgrid work as Orbit doesn't have a GPU application at the moment.

The SkySurvey app takes the i7's about 1 hour 26 minutes to complete. The project is set to replicate to 3 computers but the first 2 to return them get credit. If the last computer hasn't started it will be aborted by the project and if it has started it simply wastes its time. Not so good for the crunchers, but gives quicker turn-around for the project.

Due to the sheer number of work units flying around their server ran out of space for uploads last weekend. They managed to find room a few days later so we could upload completed work. Apparently they have requested extra funding for additional disk storage.

They are expecting to present their results in October and test it on real telescopes in mid-October, hence their need for results to be returned quickly.


GTX460 returned
I returned the 2nd card as it was giving errors when starting up the machine. Its symptoms were no display and the error beeps at power up indicate no VGA card found. I have put a GTX275 into Chekov to give it a usable GPU.

Of course it worked for the computer shop, but they are still trying to reproduce the error.


BIOS update
As part of the attempt to fix the problems with the GTX460 I downloaded the latest BIOS and installed it on Chekov. It made no difference. I also installed it on Maul as it has 3 x GT240 cards at the moment.

04 September 2010

4th of September

Milestones
Nothing new this week. Coming up on one for Seti, so I might concentrate on it for a little to get it "over the line".


Orbit
Remember them? Well they just dished out 300,000 work units and a new Sky Survey app. You don't hear a peep out of them for 6 months and then they just put out some work. The latest news item was in March.

The work units aren't too big to download, unlike some other projects. Current estimate to completion is 2 hours for my i7 machines, and they want to be back in 5 days. The last lot had a year to complete. Apparently they are planning on presenting the results at a conference in October.


Maul
Well I picked him up from the shop. He's been transplanted from a CoolMaster HAF 830 case to a CM Storm Sniper case. The power supply has been downsized from a Corsair 1000 watt to a Seasonic 750 watt. I still have to get him out of the box. I'm just waiting for one of the quads to finish off its Einstein work before I do that.


Quaddie trashes its work
For some reason BOINC decided to trash all the work I had on Qui-Gon. I had logged off as my normal BOINC user and switched to another user (to apply Windows updates) when it started up BOINC and deleted all the work units. Why it even started up I don't understand as the box for "always run at startup" was unchecked. When I log back on as the BOINC user its got no work. The logs have a heap of error messages.

Once its finished its Einstein work i'll detach and re-attach to release any work units that the 2 projects think its got. If I don't they will have to wait and time out.

29 August 2010

29th of August

Milestones
I passed 40 million for all projects this week.


GTX460
I removed the HD4850 from Chekov and installed the GTX460 into it. Its been running the usual mix of 4 projects on Chekov for most of this week. This makes 2 machines that now have GTX460's installed.


GPUgrid "fatty" tasks
They introduced a new batch of 500 workunits. These ones are twice the usual size. I managed to get one on a GTX295 which is looking like running for 12 hours. They are also in the process of ending one work stream and starting up new ones so there is a general shortage of work at the moment.

23 August 2010

23rd of August

Milestones
Nothing again this week. Crunching continues


Maul not home
I was supposed to pick up Maul after its case transplant, but the shop hasn't done it yet. It looks like next weekend now before it comes home. I did however pickup the 2nd GTX460 during the week.


BOINC testing
We had 6.11.6 released during the week, which incorporates a number of additions for the full functionality of notices. It looks pretty much the same as before from a users perspective. I have it running on Spock at the moment and will look at rolling it out to the others later.


Seti cuda optimised app
The guys at the Lunatics web site seem to be firming the x32f cuda 3.0 app for a release. They have asked us to concentrate on testing it. I previous tested the cuda 3.1 version and a later x32h version. I currently have 2 machines running it.

I managed to trash a cache of Seti work while updating the science app on Qui-Gon to x32h. Normally I wait until its finished all its work before updating a science app but because it was going to be almost a whole day before it finished I did it anyway, with the result that BOINC decided I didn't have the require apps and deleted all the work units.

With the Fermi-based cards you can run multiple instances of the app per card (assuming it has enough memory). I have been running 2 instances of the cuda 3.0 app at the same time, which finish marginally quicker than running a single instance.


GPU-Z 0.4.5
Below are a couple of screen shot of the GTX460 with the newer version of GPU-Z. This was while running 2 instances of the Seti x32f cuda 3.0 app.




As you can see from this last one the GPU is loaded up pretty much to the maximum.

15 August 2010

15th of August

Milestones
None this week. Number crunching continues for various projects.


Einstein finds a new pulsar
They seem to have found a new pulsar. Unfortunately it wasn't me, but it does show what distributed computing can do. There is also a suggestion there is a 2nd discovery which is currently being verified.

Link for details of the 1st discovery: http://einstein.phys.uwm.edu/radiopulsar/html/discovery_page/firstdiscovery.html


Windows updates
They went overboard this week. Around 29Mb of updates for the Windows 7 machines (each) and around 12Mb for the Windows XP machines (each). So a lot of bandwidth was used up on these, not to mention the time for me to install them on each machine.


2nd GTX460
Chekov currently has an ATI HD4850 installed and is mostly dormant. I've ordered another GTX460 for it and will swap out the ATI card.


Seti beta cuda app testing
This is fairly quiet at the moment. A bug has been identified in cuda 3.1 that it is unable to work out which card is doing the work for FFT calculations and thus they fail in mixed-card machines. This will supposedly get corrected with cuda 3.2.


Maul progress
Its been transplanted into a CM storm Sniper case, which will make it look the same as the others at least. Its been done by the shop, but I need to collect it, which means it will have to wait until next weekend before it comes home.

07 August 2010

7th of August

Milestones
This weekend I managed to hit 1 million for ClimatePrediction. Below is my certificate.




BOINC testing
I've been running 6.11.4 for a couple of weeks now. There have been a few changes but no new build for over a week. Apart from some differences in the messages (now known as the event log) and such (new notices tab) its behaving pretty well. I've got it installed on all the machines except the XP ones (they are running the last "recommended" build - 6.10.58). They are still working on some of the features.


Maul back in the shop
Well he's back to the shop. This time its getting a CM Storm Sniper case and a 750w Seasonic power supply, which will make it the same as all the others. I still haven't worked out why it does its random reboots so this is probably the last thing to try.


Obi-Wan flat
Its been sitting around doing nothing for a while. When I fired it up again a couple of weeks ago I noticed the clock wasn't working. It needed to get a new battery. It takes a fairly standard battery, but of course it had to be in the most inaccessable place possible. Next to the PCIe slot which was occupied by a GTS250. After much jiggling I managed to get the GTS250 out and pop the battery out of its socket (at least it was in one). Down to the electronics shop to get a couple of new ones, pop it in. Another lot of jiggling to get the GTS250 back in and away we go. Amazingly it decided to work after that. I should be right for another 2 and a 1/2 years, which is how long the previous battery lasted.


Drobo
Well I got it and a few 2Tb hard disks. I ran out of ports on the back of the gigabit switch so i've had to buy another couple of them too. Its working as expected.


Seti snap fund raiser
It seems the guys at Seti were going to remove a couple of their old servers due to reliability issues. One of the regular volunteers, msattler, ran a snap fund raiser with various guys donating $1000 each so they could get a replacement server. Unfortunately I missed it (only spotted the message thread this morning) but will see if they need some other bits such as disk drives or memory that I can help with.

31 July 2010

31st of July

Milestones
I passed 20 million for GPUgrid this week. Below is my certificate.


Seti cuda app testing
I've been beta-testing a revised Seti cuda application on various machines, that have different graphics cards. There are both cuda 3.0 and 3.1 versions. I ran the cuda 3.1 version on a GTS250, GT240 and the new GTX460 card. All worked fine for me and were an improvement on the older app run time.

There have been reports of problems with mismatched graphics cards and Jason G has been trying to improve that area of the application. I expect a revised version for retesting.

Drobo
I'm looking at a Drobo for some additional storage. They are basically an external hard drive box. However unlike a normal external drive some versions hold up to 5 drives. You can mix drives and capacities and it handles the data redundancy by itself. They seem to be around $550 for the USB and $950 (prices in AUD) for the ethernet version with no drives. This seems a much simpler solution to adding storage capacity without having to build your own RAID array.


25 July 2010

GTX460

There are a few varieties of the GTX460. The main difference is in the memory sub-system. The 768Mb cards have a 192 bit memory controller and the 1Gb cards have a 256 bit memory controller. The other difference is in the cooling system. Evga has a couple of cards designated as EE (external exhaust), meaning they blow the air down the length of the card and out the back.

This one is an evga 768-P3-1362 model, which is one of the 768Mb cards but factory overclocked. From looking at the stats from GPU-Z it seems the memory controller isn't being stressed much by the GPUgrid app. It hovers between 15 and 17% use. Temps are being reported as being 50 degrees C under load, so fairly cool when compared with the GTX2xx cards.

As mentioned in the previous post it seems most software doesn't understand it has 336 cuda cores. They think it has 224 and therefore its under utilised. Even GPU-Z 0.4.4 reports it as having 224.

It was straight out of the box, take a few photos and then off to work. I swapped out the GTX275 from Spock as its the nearest equivilent card I have. Installed the driver and we're off and running.


Specifications
GPU: Nvidia GF104 running at 763 MHz, 336 CUDA Cores, 400 MHz RAMDAC
Memory: 768 MB, 192 bit GDDR5 3800 MHz (effective) 91.2 GB/s Memory Bandwidth
Interface: PCI-E 2.0 x16 DVI-I, DVI-I, Mini-HDMI
Dimensions: Height 111mm x Length 210mm

The box



Lying down



Connector end
Showing the dual DVI plugs, a mini-HDMI and you can see some of the heatsink.



Power connectors
The other end. You can see the 6 pin PCIe power connectors. Its rated at 150 watts max power use.

25th of July

Milestones
None reached this week, but GPUgrid isn't far off one.


BOINC versions
We have the 6.11 family at the moment. 6.11.3 being the latest. Changes to the manager include a new Notices tab and the Messages tab has been removed. The messages have been relabelled as an Event log and open in a new window. The tasks tab columns have been rearranged.

Like anything new its causing some people angst, but I think its better. They will be adding RSS feeds into the Notices tab. Its still a work in progress.


GTX460
Well I got one of these. There are a few varieties of them. I will do a seperate blog post for this with some pictures.


GPUgrid science app
The guys from GPUgrid had a few problems with their app not working
a) for cuda 3.1
and
b) the GTX460

They have worked around the cuda bug and it will run with the GTX460's now. Performance is disappointing as it doesn't use all the cuda cores, but they will be addressing that after their summer break.

17 July 2010

17th of July

Milestones
I managed to hit 4 million credits for Einstein during the week. GPUgrid is coming up on another milestone too, but I will mention it when I get there.


Seti Limits
It appears the guys at Seti added limits on the number of tasks that can be distributed. This is so they don't overload their sever when the project comes up after a outage. They are currently having 3 day scheduled outages per week. They slowly increase the limits as the demand for work drops off. This way it allows all users a chance to get some work immediately following the outage.


GTX460 release
The long awaited GTX460 based on the GF104 was released last week. As expected the guys at GPUgrid are usually the 1st to rush out and get one. They are currently tweaking their science apps to work with it.

I was waiting for them and have ordered a Factory OC'ed one. Initial reports are promising about performance. They are also faily reasonably priced. The entry-level cards are around $200 USD.


BOINC versions
6.11 versions are now out for testing. Not sure whats in it as there hasn't been any mention of it in the mailing list. I'm still running 6.10.58 on my machines.

03 July 2010

2nd of July

Again no milestones reached this week. Crunching continues as normal.


Seti outage
Well the scheduled 3 day outage is over and we survived. There is still much angst over in the forums. This coupled with work fetch issues that have recently been intoduced by the latest incarnation of BOINC, which was being tested over at Seti beta, but suddenly got implemented on the main project. I'm sure Dr Anderson will have it sorted out in due course.


UPS
At the moment all the machines are on surge protected power boards. I've ordered a 700VA UPS to go onto the file server.


Network Reconfiguring
While I was at things today (applying windows updates on all the machines) I decided to try wiring up the network a bit differently.

All the number crunchers have 1Gbit LAN connections which then plug into a "Green" (ie low power) 1Gbit switch. After that it was going via the file server and then off to the router. Now the 1Gbit switch is plugged directly into the router. I'm hoping this will improve the comms speed.


GF104 due for release in July
Supposedly the official release date for the GF104 based Fermi cards is the 12th of July.

There are to be (initially) 2 versions. Both have 336 cuda cores. The reference design card having 768Mb RAM and a 192 bit memory interface. The other is a "nvidia partner" version having 1Gb RAM and a 256 bit memory interface. Expected power use is 150 watts for the reference design and 160 watts for the other version.

I would hope to try a "nvidia partner" version fairly soon after they are released. Rumour has it they have apparently killed off the GF100. Its being replaced by the GF104.

27 June 2010

27th of June

No milestones reached this week, just the usual crunching.

Seti@Home announces scheduled outages
They have announced that they will have scheduled outages from Tuesday through to Thursday each week, so they can concentrate on the science side of things.

This suggests to me that they have more crunching power than they require. At present I only have one machine doing S@H work so it will spend more time doing other projects. They already have issues supplying enough work to keep it occupied. We will see if this causes much grief. There is certainly much angst in the forums.


Internet throttling
This week I got my 1st throttle warning email from my ISP. Apparently they still had me listed as being on a 50Gb plan, despite my upgrading to the 100Gb plan when it came out over 15 months ago. Seeing I was no longer in a contract period and they now have 130Gb plans which don't count uploads, I switched to it. This should save me from future throttling.


Hardware issues
Maul still reboots from time to time. It seems to last about 3 days at a time. I still haven't managed to work out why. Since installing the newer nvidia 257.21 drivers it seems to be more stable, even though the times when it has rebooted it wasn't even doing any GPU work. Its also the only one running Seti work.

20 June 2010

20th of June

GPUgrid reaches 15M
This weekend I passed the 15 million milestone for GPUgrid. Things are bubbling along nicely there. See certificate below.




Seti and Seti-Beta woes
The BOINC developers were testing a number of changes to do with DCF (duration correction factors), work fetch and credit normalisation on Seti-Beta. They then seem to have implemented it on the main project without any warning. This is causing a number of issues with no credit being granted and machines not getting work any more. The DCF also varies wildly. All in all a total stuff up. From reading the forums a number of users are getting upset.

Fortunately I have only one machine doing Seti now, so i've simply set it to no new work and its concentrating on Einstein. I'll look at it again next week and see if they have managed to fix their issues. No point stressing about it and there are other more useful scientific projects that I can help.


CPDN
ClimatePrediction are now well into the Famous work units. I've managed to complete a few. They typically run for 260 hours on my i7 machines. I have disabled the smaller HADAM3P work units (they normally take 60 hours) as the project wants to get the Famous work completed in 4 months.


Maul
Well this machine is still playing up, typically rebooting at least once per day. I'll look into replacing the power supply and maybe the case next week.


Future hardware
At the moment i'm waiting to see the GF104 GPU's from nvidia. There are a number of rumours about their specs and pricing. The GTX460 sounds like it might be the card to go for when they become available. Expected release date is either June or July.

The i7-980x doesn't look like its going to drop in price any time soon. They are just too expensive at this point to bother with.

Still nothing from LucidLogix regarding their Adventure 2000 or Adventure 2500, so it looks like they are simply vapourware.

06 June 2010

6th of June

Its been a fortnight since the last update, so here's the news since last time...

GPUgrid errors fixed
The 2 machines that have GTX295's were getting errors when GPUgrid released their new 6.05 app. I tracked it down to having issues with SLI being enabled. If you disable SLI in the nVidia control panel they work fine, not to mention being a bit quicker.


Adventure 2500
I was keen to get hold of the LucidLogix Adventure 2500. Its basically an extension of the PCIe bus to an external box which provides an additional 4 PCIe slots. Emails to their sales department have gone unanswered.

We were discussing it and a similar product by Cubix over on the GPUgrid forums. The Cubix guys provided some additional information, but we can't get any information from LucidLogix.


BOINC versions
6.10.56 became the official version. Since then there is a 6.10.57 with a minor change to do with elapsed time calculations. I've put 57 on four of the machines and its working fine.


Mauls woes
He's paying the price for being bad. Well not really. It came back a fortnight ago. Since then its done its random reboots for no apparent reason. I've replaced the power cable and extension lead that he's plugged into but that doesn't seem to have made any difference. I even removed the 3rd GT240 from it in case that was causing problems.

I'm beginning to think it might have something to do with the case wiring as since the last time it was off at the shop the light in the power switch hasn't worked and the USB ports on the front panel have stopped working. Its a CM Stacker 830 case which is mostly aluminum and can be prone to short circuits. They are now discontinued unless I want the nVidia version (has an nVidia logo on the front and the grills are green instead of the usual black colour).


Weather
Getting cold (good for computers) and wet (not so good for computers). However managed to have 4 machines up and running for the last few weeks. This has improved the output from the farm quite nicely, with GPUgrid being the majority of it.


Milestones
I passed 500,000 for CPDN (Climate Prediction Network). It seems they've recently added a certificate shown below. I have also started running their Famous climate models. The HADAM3P models I have run up until now take around 60 hours. The Famous models take 165 hours. The good news is they don't have such huge files to transfer to get one going.

23 May 2010

3rd week of May

Another fairly quiet week on the farm.

Maul - he's back
Picked up Maul from the shop again. Its his 3rd trip. Got it home and lugged it up into the loft (climbing the ladder is always interesting with a big PC). As you can see from the previous pictures the case is a CoolerMaster HAF 932. Its not exactly light.

The shop replaced the motherboard for a new one. First bootup and it complained about the BIOS settings being invalid. So go and reset everything back the same as the other 4. This gets it booted and running. Previous Windows 7 still on there (fortunately not activated, only 2 days left).

Upgraded BOINC to 6.10.56 and got work and left it to run. A few hours later I noticed it had rebooted. Okay time to wipe it and reinstall. Formated the partitions, and reinstalled Win 7. Reinstall BOINC, copy backup of BOINC directory back onto it and its off and running again. So far its survived a full day crunching. Unfortunately Seti's upload server has gone off-line so can't upload completed work and it refuses to ask for new work (because it can't upload). The backup project on this machine (Einstein) is getting plenty of attention though.


BOINC testing
Well the testing continues. No new version as they are trying to get this one out as the "release" version. As mentioned last week pretty stable, haven't had any cause to complain.


Proxy Server
All the number crunchers communicate to the outside world through a proxy server. I last updated it about 18 months ago. I took the opportunity this weekend to update it to the latest production release. It was only 5 versions behind. Web access seems to be a little faster now too.

16 May 2010

Weekly report

Another week on the farm...


Maul ready to come home (again)
This time the they claim to have replaced the motherboard. The last motherboard was a "refurbished" one. When it gets back we'll see if it makes any difference. Half the city was closed off due to the arrival of Jessica Watson, so I couldn't pick it up this weekend.


BOINC versions
We are up to 6.10.56 this week. It seems that both the Nvidia and ATI drivers stop working if you keep polling them for free GPU ram details, so they've removed the checking from BOINC for now. Other than that it seems to be working fine. I would suspect this one will probably become the next release version.


Qui-Gon
Its an ACER veriton v7900. It seems to have well and truly died. Another dud motherboard. Still under warranty. So time to pack it back up and get it fixed. A visual inspection of the motherboard indicates that some of the capacitors appear to have cooked.

09 May 2010

May day

Maul back in the shop
It seems he just can't handle things when it gets under CPU load. Interestingly it works quite well running just GPU work, so something screwey with the motherboard. The shop seems to think that ASUS would have swapped the motherboard for a reconditioned one the last time it was in. Hopefully it won't take 2 months this time.


BOINC versions
We are up to 6.10.52 for Windows this week. The developers have been working on the file xfer limits. Unfortunately in its current form not much use to me. My ISP has a concept of peak and off-peak limits. Also i'd need to apply the limits across the entire "farm". I've put it on 2 of the machines and they have been working fine so far.


Obi-Wan returns
Given Maul is back in the shop and Qui-Gon is playing up I dug out another oldie and fired it up. All the quaddies have been sitting around gathering dust, so after reinstalling BOINC and then allowing Windows update to bring it up to date its off and running.

I hope this one lasts better than Qui-Gon, which works for about 2 days then starts beeping and needs 2 days rest before resuming work again. Another screwey motherboard.

02 May 2010

Weekly news

A few things in this weeks update:
  1. New BOINC versions, old one pulled
  2. Maul playing up
  3. Project realignment
  4. AMD releases the Phenom II x6
  5. GTX260 donated

BOINC versions
Bad news first. UCB decided to pull the current release version (6.10.43) due to a work fetch issue. To get the issue you have to be running a project that has a cuda app and your graphics card has less than the required ram (typically 256Mb cards). If you have one then they recommend you go back to 6.10.18. If your graphics card has 512Mb or better then you won't have a problem, so there is no need to do anything.

New version 6.10.50 was released with a bunch of changes and a few broken bits. The notable change they are adding is the ability to set a file transfer quota, and when exceeded it will suspend communications. Unfortunately the enable/disable comms options are now broken, but they are aware of the problem. In fact they have already released an updated Mac version. We're still waiting on a Windows build.


Maul playing up
Well it came back with much fanfare, but hasn't really behaved since. Firstly the shop hasn't got the case wiring correct and the power button is no longer illuminated and the PC speaker no longer works. These are just an annoyance rather than an issue. The main problem is it keeps locking up when using all 3 graphics cards. I suspect its a BIOS issue and the P6T cannot cope with 3 GPU's going at once. At the moment i've removed the load from GPU3 (which is in the PCIe x4 slot) and will see how it goes. BOINC is only using 2 GPU's for now.


Project realignment
After a bit of experimenting with running the GPUgrid work on the GT240 cards in Maul I've decided they are way too slow (about 10 hours a wu). So i've set Maul to just run Seti and Einstein work. The other machines with the more capable cards are concentrating on GPUgrid work.

Orbit would appear to be generating some beta-test work units while they sort out a new app. Once thats done they are expecting to generate real work. Its taken some years to get to this point, so i'm not holding my breath.


AMD releases Phenom II x6
While its somewhat behind Intel, the AMD doesn't do hyper-threading, so each core is an actual cpu. The release in itself this isn't significant, other than it may convince Intel to drop its prices for the i7 980x.


GTX260 donated
I've donated one of Maul's old GTX260 cards to one of the developers of the optimised apps for Seti@home. Most of the guys don't have a high-end card so cannot test their optimisations without help from others. I am hoping this will allow for more optimisation of their cuda app.

25 April 2010

Maul pics

While Maul was off in the shop I got them to replace the graphics cards. The new ones are the evga GT240 as you can see below.



The card is a single-slot design so its possible to put 3 of them into the Asus P6T motherboard. It doesn't use any external power connectors either.


And this is what it looks like once they have all been installed into the CoolerMaster 932 case.



Another shot taken at a different angle.




And another shot from the front of the case.




Here is the GPU-Z report for them. The first two cards run at PCIe x16 speed and the third one runs at PCIe x4 speed, other than that they are identical.


Speed-wise they are slower than the GTX260's but then you can't get the GTX260's any more. They only have 96 shaders (now called cuda cores) so they are theoretically less than half the speed, but the GDDR5 memory system helps to make up for some of that. I have run some GPUgrid work that comes in over 10 hours, where a GTX295 takes 6 hours.
Another "feature" to note is that under Windows Vista and 7 you need to put vga dummy plugs onto the 3rd card before it can be used. I asked in my computer shop for one and they had never even heard of them. All they are is a couple of resistors in a vga plug to simulate load.

21 April 2010

A quiet week

A fairly quiet week production-wise. I've been leaving one machine on during the day and the rest off. In the evenings I usually fire up a 2nd one, which operates until its time to head off for work.


GPUgrid
They released a new science app 6.72 which is meant to be faster. However the 3 work units that I got so far were the long variety that take 10+ hours, so its a bit hard to tell.


Maul ready to come home
Maul (its an ASUS P6T) has been off at the shop since February. Its motherboard has been back to ASUS twice. Anyway its come back. Funny thing though, the shop rang me to tell me Win 7 isn't working. I asked why as it was working when I dropped the machine off. They couldn't explain why.

Its had 2 old GTX260 cards removed and 3 GT240 cards installed. The CPU fan has been replaced by a Noctua and the old CPU fan is now blowing air out of the top of the case. The memory should be 6Gb this time.


Nvidia GTX470 and GTX480
It seems very few of the science apps will work with them. A couple of projects have managed to get them to work. There is a alpha-test Seti multibeam app that runs. The GPUgrid apps all fail, but a member has donated a 470 and a 480 to them so they can work out why their science app doesn't work.

There also seems to be some confusion with drivers. The latest 197.45 (WHQL) driver doesn't support them, but there is a 197.41 that does. No doubt nvidia will sort things out in due course, but I am not wasting the money on one at the moment.

09 April 2010

GPUgrid zooms past

This week GPUgrid cracked 9 million credits. It also passed my Seti credits (last week they hit 9 million). I've been crunching Seti since February 2008 and GPUgrid since December 2008.




BOINC versions
This week saw 6.10.43 pronounced the "release" version for Windows and Linux and 6.10.44 for the Mac. Following on from that 6.10.45 came out with a few minor fixes that just missed the 43 version. I've got 43 on two machines and 45 on another, the rest are switched off. Both versions are working fine.


Maul
He's still off getting fixed. Its only been 6 weeks so far. Apparently ASUS fixed it once, something to do with the BIOS (even though it had been updated a couple of times). As mentioned in my 1st of April post it was damaged in transit and has gone back to be fixed yet again.


GT240 cards
I was looking to get 2 or 3 of these to replace the aging GTX260's that were in Maul. Apparently the GDDR5 ones are about 2/3rds the speed of a GTX260, which is pretty good considering they have have less than half the number of shaders (now known as cuda cores) of a 260. They also use a lot less power and are single slot cards (well the evga ones are). The shop are supposedly finding out more details (like pricing and availability) for me.

05 April 2010

Seti passes 9 million

Despite concentrating on GPUgrid for most of the long weekend I managed to reach the 9 million milestone for Seti. Things have been returned to normal (ie no more manual intervention).




GPUgrid
As mentioned above I was concentrating on GPUgrid for most of the long weekend. The guys there have been trying to test some new applications that are a bit faster than the current one. They also tried a cuda 3 version to support the newer Fermi-based cards. Unfortunately the cuda 3 apps crashed straight away, so they have gone back to using the cuda 2.2 and cuda 2.3 versions.


Qui-Gon playing up
This is a quad-core machine with a GTS250 card in it. It was working most of last week without any problems. On Friday it decided it wasn't going to work any more and had a constant beep from its speaker and not even a BIOS screen. It did this a fortnight ago when I cleaned it out. I left it off for 2 days. Its now back up and running. I don't know why it does this. The CPU temperature seems to be around 43 degrees according to the BIOS readings.

01 April 2010

Fermi released

The offical launch of the Fermi-based cards was last week. I have been checking pricing on the GTX470, which seems to be as low as $500 AUD. Still about $100 more than it should be, and nobody has any stock until April the 12th at least! Its too expensive for me, so I will wait a bit in the hope the price will drop further.


Project news: GPUgrid
It seems a reviewer has been trying a GTX480 over at GPUgrid. The project has 2 science apps, one worked, one didn't. The developers over there been updating their apps to support the Fermi-based cards., which includes upgrading to the newer cuda 3 libraries.

They are also seeking a card (or cards) to be donated. The developers posted the reply from nvidia. It seems nvidia weren't interested unless GPUgrid had recently purchased a Tesla, which of course they hadn't.


Other news
Not much to report this week. The weather cooled down a bit so I have managed to get Kirk and Sulu running all day for the last couple of days. They both have a single GTX295 so they can do a fair bit of cuda work.

BOINC also had a minor issue with the newer Fermi-based cards as it wasn't correctly reporting some card details at start up. Its just a cosmetic bug which they have already corrected.


Maul loses capacitor
The shop told me Maul should be fixed by the end of the month. It still hasn't come back. Apparently when they shipped the motherboard back to the distributor a capacitor had come off. It was working properly, apart from the 4Gb memory limit, when I returned it to the shop.

Unfortunately all the work on Maul will have timed out by now. It was completed but not reported due to problems with the Seti servers at the beginning of March.

22 March 2010

Climate milestone

I managed to get to 250,000 for Climate Prediction this week. I think thats pretty good considering they take around 60 hours per work unit. Unfortunately they don't have a certificate.


Fermi
I've asked the computer shop to see how many Fermi based cards they are likely to be getting. They think it will be something like 20, but don't have a confirmed allocation at the moment, nor do they know what sort of cards they will be (GTX480 or GTX470). Add to that they don't know pricing either. Anyway i've asked for two GTX470's, depending on the price and availability of course.


Qui-Gon returns
Seeing as Maul is still off with the distributor I fired up Qui-Gon (an Acer quad-core machine). It was working fine for a while then it decided to reboot and keep the fans running at full speed. I went into the BIOS as it shows the CPU temperature. It looked fairly normal to me. I used a can of air and blew out the heatsink just in case. After that it refused to boot. That is until tonight when I tried it, a week after dusting it out. It booted up first go and is back to crunching some Einstein work.


BOINC versions
We had 6.10.39 last week and then it jumped up to 6.10.43 with a bunch of changes in the space of a couple of days. I've got it on two machines and it seems to be working fine. It has a number of improvements to the amount to CPU time the core client spends managing tasks.


Electricity
It seems the NSW state government has decided to bail out the electricity generators. They announced a 64% price increase over 3 years. This is on top of last years price increase to fund infrastructure upgrades.

11 March 2010

GPUgrid hits 8 million

Well I managed to crack the 8 million milestone for GPUgrid. Below is my certificate.




GPUgrid bonuses
They have tweaked the bonus so that its 50% if the work is returned within one day and 25% if returned with two days. This is to encourage more work to be returned as soon as possible. The results of one lot of work units are used to build the next set of work units, so they need them back ASAP.

07 March 2010

Another quiet week

Another fairly quiet week on the farm. The weather warmed up so only one machine was left running (Kirk). I was away for a few days and left it to do its thing. When I came back the room temperature was 33 degrees but it was still going :-) Anyway I turned it off and waited for things to cool down.


BOINC news
We got up to 6.10.36 earlier in the week. This has now become the recommended for all systems except Linux.

I like the resume of uploads. Climate Prediction and GPUgrid produce some big result files to upload and if the transfer is suspended it used to restart from the beginning, now it resumes from where its up to.

Another feature that may interest some people is the concept of a backup project (for those that want to crunch for a single project). It will only use the backup project when its out of work and can't get any for the primary project.


Maul off to be replaced
As mentioned last week Maul (its a ASUS P6T) went back to the shop. They have determined its a faulty motherboard and will be sending it off to the distributor. Hopefully it will get replaced fairly quickly. It still has a fair bit of work on it that was completed but couldn't be reported for Seti due to their last major outage.


Fermi firms up
Pricing it seems is based upon whatever they can get for it. Indications are the GTX480 will be $680 USD and GTX470 will be $299 USD. These are expected to be released in limited numbers only so the odds of getting one are fairly low. If certain web sites are to be believed then there will be less than 10,000 made.

01 March 2010

Autumn. Time for crunching

Not much to report this week. Weather was warm to hot until today, when like someone has thrown a switch its back to colder weather. Its also the 1st day of autumn. Being cool and wet has meant temps are down, so I have got 3 machines running as I type this.


Maul. No verdict
Checked with the computer shop on Friday to see if they have managed to find anything out about Maul and its 4Gb memory limit. They have some parts coming in so haven't managed to try certain things yet (like removing the cpu). They expect to swap the cpu for another one, if only temporarily, to see if that resolves things this week.

They did the obvious things like change the memory, but then i'd already ruled that out by using a known good set that I took out of Sulu, and have since put back.


BOINC news
I did offer my services as an alpha tester but never got a reply back. Hello Rom.

There are some rumblings of discontent by certain users. All very nice complaining, but none are volunteering to do any coding. If they really wanted to improve it then they should start doing something about it.

The current beta version is 6.10.35. Just a few minor changes in this one, so I haven't bothered putting it on any machine. In fact the most up to date machine was Maul and its off at the shop now.

21 February 2010

20 million and counting

Despite a quiet week on the computing front I managed to hit the 20 million milestone this week. Below in a FreeDC image of my latest stats.




Maul in the shop
Given all the problems getting Maul to recognise 6Gb of memory its back at the shop. A google search turned up some forum post mentioning "bent pins". This would suggest a bent pin or two on the cpu socket. Seeing as the shop put the machine together they can sort it out. ASUS provide a 3 year warranty as do Intel, so the components are under warranty.


Seti long outage
I had some Seti work on Maul which I have been trying to get off it before sending it to the shop. Unfortunately they have had a few problems, one after the other. I managed to upload the results, however couldn't report them. Its been like that for about 4 days now. I have set the other machines to No New Work for Seti so they don't bother until the project is back to normal.


Fermi
While I was in the shop dropping off Maul I had a discussion regarding the upcoming Fermi (aka GF100) gpu chip. They have been advised by their supplier to expect some stock at the end of March. This initial chip is supposedly 20% faster than a ATI 5870. It is expected to be followed in May with another version that is 30% faster. After that they are expecting a dual-chip card to come out. All are meant to be faster than a GTX295. No indication on if they will actually have stock or how much they will cost.


Intel i7-980x
According to web reports these are due out on the 16th of March. Apparently slated for the Mac. Its a hex core with hyperthreading, so 12 cores on a chip. Its clocked at 3.3Ghz. Indicative pricing is $999 USD. Given the pricing I don't think i'll be getting one anytime soon.

13 February 2010

More memory woes

Well I got 4 sets of memory from the shop this week, replacements for the 2nd lot of duds. I put one set into Sulu and left the memory tester running. All good, so onto Maul. Not so good. All 3 sets failed. Came up as 4Gb instead of 6Gb.

At the moment what I might do, when I can run its cache of work down, is to take the memory out of Sulu and stick it into Maul. If that works then its probably the memory. If it fails then its probably just Maul.

Update: Swapped memory out of Sulu into Maul and it still reports as 4Gb. So its something about Mary, I mean Maul. Swapped it back, put original 1Gb x 3 back into Maul and we are back at square one.


Weather
Or should that be whether to run or not? Its been hot and sticky again this week, raining in the evenings, so all the machines have been off for most of the week, with some occasional runs overnight. Not much science done this last week.


Project news: GPUgrid
New faster app released to the public. Its claimed to be up to 60% faster, but I think thats if your cpu has nothing else to do. Unfortunately mine are busy running cpu tasks as well as the gpu tasks, so its nowhere near that figure.

It also didn't resolve the FFT issues with the 65nm GTX260's (which Maul has a pair of). I didn't expect it to as its really an Nvidia issue.


Project news: Einstein
They have release a new app that takes 4 ABP work units at a time, so they take 4 times as long to run. This in turn relieves some of the stress on the server.


BOINC testing
Some changes are afoot. They are planning on doing nightly automated builds. They are also looking for some more people who'd like to do alpha testing, preferably with a fair bit of experience with BOINC. That reminds me I should install 6.10.32 on one of the machines to test it.

28 January 2010

Einstein reaches 3 million

Finally I managed to get Einstein to the 3 million milestone. Below is a rather faint copy of the certificate.




Memory (or lack of)
The shop still hasn't replaced the memory. It was meant to be available and tested by Monday. Monday came and went and when I rang they admitted it hadn't even arrived. Tuesday was a public holiday. So the other couple of machines (Sulu and Maul) await their memory upgrades.


BIOS upgrades
I've been on the ASUS web site tonight checking for what BIOS version is the current one. Apparently the one I installed (ver 0202) dates back to January 2009. The current one seems to be 1004 and is dated January 2010. It still doesn't support the i7-980x, so I may as well wait until they have a version that does.


Progress report
Well not much to report due to having machines off. Its hot and humid here in Sydney. As mentioned above Einstein got to its milestone.


Project news: Einstein
The guys at Einstein have been busy developing a new ABP (Arecibo Binary Pulsar Search) science app. This uses less double precision maths and is therefore somewhat faster.

They have also indicated there is a new Gravity Wave search program in the works. This is expected to cut run times down from around 8 hours to 3 hours. It won't be deployed until the next search run (S5R7).


Project news: GPUgrid
The guys at GPUgrid have been optimising their science app. There is a newer version in the works that is supposed to be up to 60% faster that the current one. They have also been trying to move away from using the CuFFT libraries provided by nvidia.

Unfortunately their testing of an OpenCL version of their science app proved that OpenCL on the ATI cards is not ready yet. Apart from being somewhat slower than a CUDA app, the ATI software causes system hangs or reboots. Hopefully later versions will correct these issues.

18 January 2010

Out of memory

Memory
I bought 4 sets of RAM upgrades. However when I tried them all four reported 4Gb instead of 6Gb when installed in Maul. Thinking this was a little suspicious I installed one set in Sulu, which found 6Gb, however when I ran a memory tester on it, I started getting errors. So back to the shop they went.

I have suggested I will start charging them if I get another faulty batch. No news yet from them regarding replacements. If there is no output from a memory test with the next lot i'm not even taking them out of the shop.


Graphics cards
Maul has a couple of old GTX260's (the 65nm variety). Unfortunately these don't work with the GPUgrid science app due to bugs with Nvidia's FFT cuda routines. They are also over a year old and starting to get clogged with dust. If I can't clean them out, which requires some disassembly, i'll need to replace them. My options at the moment are:

  • Replace them with newer GTX260's (55nm). Approx AUD $300 each
  • Replace them with a GTX295 (if I can get a single PCB card). Approx AUD $600

I was hoping Nvidia would have the GT300 (aka Fermi) cards available by now, but they still seem to be having issues. Rumours suggest that they get too hot so they have had to reduce the number of shaders. Don't expect to see these coming off the production line any time soon.


BOINC versions
6.10.29 was released with a number of modifications, most notably to the work-fetch side of things. I've installed it on 3 machines, which would appear to be working fine so far.


Plans for 2010
As far as future (potential) upgrades this year go its currently looking like:

1. Upgrade Intel i7-920 CPU's to Intel i7-980x (aka the i9)
2. Upgrade graphics cards to GT300 based ones, if Nvidia gets its act together


Progress this fortnight
Not a great deal. The machines have mostly been off during the day and only a couple running overnight due to the heat. I have been running a number of climateprediction work units which take 60 hours each. This is in addition to my usual mix of Einstein, GPUgrid and Seti work.


Einstein news
I still haven't managed to get Einstein to the 3 million credit milestone yet. Its currently on 2.98 million. Hopefully by next week i'll be able to get it to this milestone.

They have released a new ABP science app that uses a lot more single-precision maths, thus making it a lot faster and run times have reduced quite dramatically.

01 January 2010

Another year gone

Well another year has gone by. Below is my BOINCstats image of where I ended the year. A bit shy of 20 million credits. I wonder if I can double it by the end of 2010?




Laptop screen
The screen in the laptop seems to have died. At the moment i've plugged in an old monitor to get things going. I bought another laptop (different brand) and once i've got my stuff off this one its off to be fixed.


Wether or weather
Its hot and humid here, currently 31 degrees in my loft where the machines live. I only have one running due to the heat. I don't expect much progress this month until things cool down. One good thing though, the electricity bill has halved because the machines haven't been running as much.