20th of October

Again not much to report this week. The weather turned hot so the farm was off for a couple of days. This of course makes a fair dent in the amount of work that can be processed. I give priority to the machines doing GPUgrid as I figure medical advances are more important than astronomy.

Today I finished installing all the GTX660's in the various machines. One each for the 12 core machines and last week I put two into one of the new 8 core machines. I also had to move a couple of machines around to fit the KVM cabling. The 12 core machines have PS/2 keyboard and mouse but the newer machines don't so I need to use USB connectors for them.

The file server was making various noises at the beginning of the week. This turned out to be a case fan that had died and was no longer able to spin up. It was supplied with the case, a Fractal Designs one. I had a couple of old CoolerMaster fans sitting in a box so put them into it temporarily. I have ordered a few Noctua fans. When they arrive I will remove the CoolerMaster fans (they are noisy and also tend to fail after a couple of years).

I managed to find a nasty little bug with the BOINC Manager by accident early in the week. The result is it will lock up the manager and use a whole CPU core. It then needs to be terminated by Task Manager. The developers have been made aware of how to do this and have reproduced it, so they will get a fix out soon.


Project news - Seti
They have commenced their annual fund raiser. The money they raise is used to fund the project rather than specific bits of hardware, which is what the GPU Users Group concentrate on. I will be making a donation, but probably not as big as usual seeing as I gave at the office, I mean the last hard disk fund raiser. As it is I have donated a 5 figure sum since joining Seti@home in 2008.


Adapteva on KickStarter
Adapteva are seeking funding to make their 64 core processor more readily available. This is a nifty way where people can give money (from $10 up) towards their target and as many people can do so as it takes to reach their target. Their target is $750,000. As I type this they are on $405,000. It closes on the 25th of October. The link to their project on KickStarter is here: http://www.kickstarter.com/projects/adapteva/parallella-a-supercomputer-for-everyone

6th of October

Farm News
Not a great deal to report this week.

Two of the four GTX660's have been installed in one of the P8Z77 machines. The good thing is the P8Z77 has two of the PCIe slots triple spaced, that is there is enough room to put two double-width graphics cards in there and still have a gap between them.

We had a couple of hot days this week, so the farm was powered off for two days. Its spring in Sydney so its probably going to be a long hot summer. Not good for number crunching.


GPUUG fund raising
One of the users has decided that for this weekend only he will match any donations up to a limit so the Green Bank transport fund raiser can be finished off. Well not one to miss an opportunity I made a donation and so have a number of other people. These donations have now completed this fund raiser.


Project news - Asteriods
This is a newish project that is calculating the shape of asteroids from multiple observations. Their science app is written for Linux, so I have setup a Virtual Box machine in order to run some of the work units. They were looking at using Vbox from within BOINC, but have now decided they will port their app to Windows (which represents about 60% of the user base). This is good news as BOINC running in Vbox is not as efficent as running it on the host operating system.


Project news - CPDN
Climate Prediction had one of their servers run out of disk space a fortnight ago. That then meant the upload server also filled up its disks. Its taken two weeks for the project staff to move the files around so we could get the uploads through. The climate models produce large files and with everybody uploading work its a constant battle just to clear space on the upload servers.

The new i7-3770 CPUs are much quicker, so much so that they get through a regional climate model in around 60 hours. The older i7's were taking 100+ hours. I have decided to only run CPDN on the new machines.


Project news - Seti
They have been having a lot of trouble with network bandwidth. This was made worse by all the short work units which were taking longer to download than run. They have moved onto some longer running work units so things have improved a bit. They are trying to get the v7 multi-beam app out, this will do an additional calculation thus taking longer to process and that will reduce the demand for work. Unfortunately its taking longer to get this app ready for deployment.

GTX660

This is the Palit GTX660 (no not the Ti). I bought these to replace the 560Ti cards. Its a factory over-clocked card.

 

In the cardboard box is a driver disk, power cable, DVI adaptor and the card itself. Its a bit shorter than the 560Ti.

 

 

 

This is a side-on shot. I was trying to show the heatsink in there.

 

 

 

And here we are installed in one of the Asus P8Z77 machines.

 

 

 

Here are the GPU-Z screen shots.

 

 

 

 

You'll need Nvidia driver 306.23 at a minimum as older drivers don't recognise the card.

 

Initial runs have been on Seti using Jason's x41z (CUDA 4.2) app. After that I let it loose on some Einstein work (CUDA 3.2).

 

The Seti work seems to be a bit faster. To be fair they have all been short work units recently so its hard to tell as they only run for 2 or 3 minutes. The Einstein work showed much better speed. I must stress that I have only run a few work units through it at the moment so its a bit early to tell, however Einstein appeared to be around 30% faster. The crunching will continue.

GPUgrid 100 million

I finally hit 100 million for GPUgrid. Below is my certificate. Crunching continues...

 

 

 

HD 7770
The old HD4850 that I have only supports OpenCL 1.0 and none of the projects use 1.0, so I bought a HD7770 during the week. The idea was it could run OpenCL apps as they are supposedly pretty good. Well they aren't.

I put it into one of the Asus P6T's which already had a GTX560Ti. After a bit of stuffing around to get OpenCL installed its ready. I start up BOINC and it gives a theoretical figure of 1,382 GFlops for the GTX560Ti and 3,520 for the HD7770. That suggests the 7770 is around 2 and a 1/2 times faster than the 560Ti.

I allow Einstein to download some work, the 560Ti gets 2 BRP cuda tasks and the 7770 gets 2 BRP OpenCL tasks. The 560Ti completes them in 2,000 seconds and the 7770 does them in 5,293 seconds. Okay its not a fair comparison as we are comparing a CUDA app with an OpenCL app. So to be fair I get some Milkyway work. They run OpenCL on both. The 560Ti takes 320 seconds and the 7770 takes 373 seconds. While the 7770 is theoretically faster, in real life it isn't.


Upgrades
As you would have noted in the last couple of weeks I have been updating some of the motherboards and prior to that I replaced the GTX570's with GTX670's. Now they have been completed that leaves the remaining GTX560Ti's to be replaced. I have been looking at the GTX660 (not the Ti). That would bring the peak power down from 170 watts to 140 watts and should give a performance boost as well.