Message boards :
Number crunching :
i7 Hyperthreading + GPU + Most Efficient Seti Workload
Message board moderation
Previous · 1 · 2 · 3 · 4 · 5 · Next
Author | Message |
---|---|
ausymark Send message Joined: 9 Aug 99 Posts: 95 Credit: 10,175,128 RAC: 0 |
I only have one nVidia 580 in the system, so max load from it would be around the 200W mark. The Power supply itself isnt operating above 55% capacity, so unless its faulty I doubt its an inability of it to supply the required power. |
ausymark Send message Joined: 9 Aug 99 Posts: 95 Credit: 10,175,128 RAC: 0 |
I've just written to Seasonic to make sure that where I have the cables plugged into on the power supply that I am not pulling too much power out of a single +12V channel. (I also just swapped one PCIe cable into the other PCIe socket) As a secondary step I am trying to see if running a single seti work unit on the GPU will cause the crash/reboot. I will let it run for 24 hours to test. If that works I will go back to two seti work units on the GPU and see if the crashing reoccurs. Not really the experiment I had in mind lol Cheers Mark |
ausymark Send message Joined: 9 Aug 99 Posts: 95 Credit: 10,175,128 RAC: 0 |
Update 22/9/11 Interesting observation, assuming the size of the gpu work units haven't changed I seem to be processing them at basically double the speed as when running 2 work units on the gpu. So besides the small 2 second lull when changing from a finished to new WU there may be little advantage to running 2 WU on the gpu This may also free up some cpu and ram resources..... Intersting.... Note: on the power supply side this is what i am using: http://www.seasonicusa.com/X.htm Cheers Mark |
LadyL Send message Joined: 14 Sep 11 Posts: 1679 Credit: 5,230,097 RAC: 0 |
Update 22/9/11 The file size of the units does not differ. However processing speed is a function of the angle range (AR) at which the WU was recorded - rule of thumb, the bigger the AR the faster the processing. We are currently recieving almost exclusively VHAR - spotted easily in the task list by the far shorter deadlines (~2 weeks instead of ~ 6 weeks for mid AR tasks). 2 units at the same time (on a Fermi) should take somewhat less than double the time of one unit. (on tasks of same AR) |
Spectrum Send message Joined: 14 Jun 99 Posts: 468 Credit: 53,129,336 RAC: 0 |
Hi Mark. I had the same problem with my rig a month back, after 4 or 5 weeks the brand new machine would lock up and at that point I was only running 1 fermi app per 560 card, ran memtest and it showed nothing, grabbed the DDR3 out of my Daughters machine and all works fine, had a faulty stick of Kingston ram,local store replaced it no problem but I got 2x2 gig kits of GEIL DDR3 anyway and has been fine for over 6 weeks now, may be worth a look. |
ausymark Send message Joined: 9 Aug 99 Posts: 95 Credit: 10,175,128 RAC: 0 |
Hi LadyL & Spectrum I have now restarted two work units on the PC. The Completion times are around 10min instead of 6 min - so the newer work units are different to what I was processing. Now to see if the thing reboots. As I have 12GB on this rig when it reboots I will pull one stick out. Then run the test again, if it reboots then pull another stick out, and so on until I locate the faulty stick (Assuming its just one). These are Kingston sticks too - sortoff assumed they would be rock solid. Anyway we will see how we go as I no longer believe its a power supply issue. (Even though I am still investigating it lol ) Cheers Mark |
ausymark Send message Joined: 9 Aug 99 Posts: 95 Credit: 10,175,128 RAC: 0 |
Update Time I have replaced a new stick of RAM that failed into its fifth week of operation and was causing the PC to reboot every 20 to 50 minutes. (Not fun when trying to get work done) Now running 'high load' tests over the next 48 hours. Essentially running 2 GPU and 3 CPU SETI tasks over that time. Assuming the PC doesn't crash I will then scale back to my regular 12 to 16 hours a day of running time. Most likely that after returning to normal operating hours the RAC may drop slightly. We will see, as it was still climbing when the rebooting issues started to occur. Combined with the inability to get/send work units hurt the RAC results this week. Anyway, lets see how things go. Cheers Mark |
John McLeod VII Send message Joined: 15 Jul 99 Posts: 24806 Credit: 790,712 RAC: 0 |
A similar experiment was done on earlier hyperthreading machines. It will be interesting to see if the results are about the same. The earlier results were an improvement of about 15 to 30% when hyperthreaded over non hyperthreaded. BOINC WIKI |
ausymark Send message Joined: 9 Aug 99 Posts: 95 Credit: 10,175,128 RAC: 0 |
Hi John This isn't so much a test of hyperthreading verses non hyperthreading as the hyperthreading feature will be on all the time. Its more about how much can you load the CPUs/GPUs before you either get diminishing returns, or negative improvement, or a very poor user experience (this is my daytime PC after all). Though atm I am hitting issues with supply of work units, which I haven't had happen in a long time - Murphy is looking over this experiment lol. Cheers Mark |
ausymark Send message Joined: 9 Aug 99 Posts: 95 Credit: 10,175,128 RAC: 0 |
Update I have had to try to get seti work units uploaded/downloaded via a third party transparent proxy as it looks like my IP address is caught up in the HE connection issues that seti is experiencing. This has caused a severe drop in ability to process work units. So on Saturday I decided to change my main working environment from Mandriva Linux to Ubuntu linux - as it seems allot of development work is now occurring on the DEB line of things, I also wanted to try out the Unity interface under Ubuntu to see if simple is better. (Though I do enjoy the customisation involved in KDE4) The install was pretty painless, and for the sake of jumping into the deep end I decided to play with the Beta of 11.10 - which solved problems with printer drivers etc. Setting up Seti and CUDA was an interesting challenge however. I removed the nveau drivers to ensure Ubuntu was using the nVidia drivers. In Mandriva I had BOINC/Seti setup under my HOME directory. Trying the same thing under Ubuntu for some reason generated computation errors - and unexpectedly so on CPU tasks! It wasn't long before I used the DEB repositories to download the Boinc Manager and Seti program itself - in fact the updated seti also used CUDA! (This impressed me lol). So, it seems I dont have to mess around with seti/cuda anywhere near as much as i needed to in the past. However I am not sure how optimised the code is. So now I am waiting for CUDA tasks to see if they run, as I slowly upload more CPU tasks. Hopefully within a few days things will return to some form of normality for seti processing so that I can get some form of baseline measurement. Which I think will be around the 14,000 RAC mark. Cheers Mark |
arkayn Send message Joined: 14 May 99 Posts: 4438 Credit: 55,006,323 RAC: 0 |
|
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
At this moment, you have no CUDA work on the machine. All the work Mark has is Stock 5.28, Claggy |
ausymark Send message Joined: 9 Aug 99 Posts: 95 Credit: 10,175,128 RAC: 0 |
Hi Clagy & All The new Ubuntu / SETI system has been requesting CUDA work units, however either there have been none available or the server has denied sending them due to the maximum allocation of work units per PC having been hit. From your comments above I am guessing the 5.28 versioning is not as advanced as the 6.X that we run using the optimised clients. What is interesting though is that in Ubuntu there seems to be several include files (/usr/include/boinc/) relating to BOINC, hmmmm but I guess that isnt the seti core processing file. (For a moment I was entertaining the idea that it compiled a version of the Seti client to the machine specific capabilities.) Ive been able to find where the respective SETI and configuration files are stored within Ubuntu so shortly I may try superseding them with the enhanced Seti apps. Hopefully this time around it may stop the CPU computation errors which forced me to go for the standard Seti app in the first place. I am however surprised that the standard Seti app is also requesting files for both Astropulse and CUDA - is this normal? Cheers Mark |
arkayn Send message Joined: 14 May 99 Posts: 4438 Credit: 55,006,323 RAC: 0 |
BOINC knows that you have a CUDA capable GPU but does not know that there is not a application available on the server for you to use. Astropulse is available for Linux though. |
Spectrum Send message Joined: 14 Jun 99 Posts: 468 Credit: 53,129,336 RAC: 0 |
Hi Mark. You may want to take another look at your setup as currently there is no GPU indicated on the details page of your machine! |
ausymark Send message Joined: 9 Aug 99 Posts: 95 Credit: 10,175,128 RAC: 0 |
Hi Spectrum Thanks for the warning. I'm moving over to Ubuntu from Mandriva Linux and am coming to grips with its repository based BOINC/SETI. Initially it seemed that it could do CUDA tasks as it was requesting CUDA work units. I have been trying to confirm but getting work units to/from SETI/Burkeley is proving to be problematic atm - and hence up until this morning wasnt even sure that the client, or the enhanced seti apps that I have configured into Ubuntu were even being allocated - which atm seems like they are not. I may try running seti again from the home directory - but the reason why I stopped doing that under Ubuntu was that every work unit was getting computation errors after a few seconds..... Aint computers fun? lol Cheers Mark |
arkayn Send message Joined: 14 May 99 Posts: 4438 Credit: 55,006,323 RAC: 0 |
|
ausymark Send message Joined: 9 Aug 99 Posts: 95 Credit: 10,175,128 RAC: 0 |
Hi Arkayn Yes, the script on the system that I came across seems to do that, as well as to run the BOINC user at a low priority "2". Ive copied the optimised binaries to both the project directory and to the usr/lib/boinc-app-seti directory. I have also created symlinks from those binaries into the /var/lib/boinc-client/projects/setiathome.ssl.berkeley.edu and /var/lib/boinc-client/projects/setiweb.ssl.berkeley.edu The project seems to be trying to download Astropuls and CUDA work units (as per my app_info.xml file. I guess we will see what it does with them once they download. Cheers Mark |
ML1 Send message Joined: 25 Nov 01 Posts: 20084 Credit: 7,508,002 RAC: 20 |
On Linux: Anyone know what the boinc-client actually looks for to determine if there are any GPUs that can be used? Happy fast crunchin', Martin See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
ausymark Send message Joined: 9 Aug 99 Posts: 95 Credit: 10,175,128 RAC: 0 |
Im no expert, but learning more every day. But you need to have CUDA libraries installed and the video drivers must match that and the kernel modules. What I am not sure however is if the standard seti client does CUDA/GPU - otherwise u need to use the optimised CUDA libraries available from: http://www.arkayn.us/forum/index.php?action=downloads;sa=view;down=5 Cheers Mark |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.