So who's running the SETI CUDA client then?

[TheTross]

Probably not many, but I'm going to start highlighting users in the news that are using the S@H CUDA client, so a simple yes + username will suffice. You could also report your experience of it so far if you want, as this may help others to decide to take it up or not (not least of all me ).

GPU user list (as of 06/02/2011)
Ba
Biffa
br83taylor
carbon8ed
Chasser
eggcup
FluffySheep
GrandMasterD
Grimgoth
halz
Jonathan Jeffrey Davis
k3v
kolala
loudbob
MadMaC
Marine Iguana
Mason-
Meridian
Odan
Pilgrim57
pnutty
ReaVerUK
Simon
the_running_man
ToxicTBag
wayne


Please note that any members that haven't shown any SETI activity for a month are removed from the list. Also, please help me maintain an up-to-date list by letting me know if you've started/stopped using the GPU app.

 

[ChrisLX200]

Yes ChrisLX200.

Not without problems, some CUDA workunits will crash out (VLARs) and the scheduler throws a fit as it can't seem to cope with both AP and CUDA workloads at the same time. You have to know how to spot the likely problem CUDA units and abort them. Right now I only have AP WUs to crunch on the CPU as the server will not issue any CUDA units to me. The problem is that I got 10 APs in the task list, Sceduler thinks that's enough work to keep me going, so the 280GTX will be idle until the task list is reduced.

On the plus side when it's working it's hugely faster than CPU (5mins c.f. an hour or so). The 280GTX gets very hot so good cooling required. Not had any driver crashes but video and mouse do stutter a bit when it's running 100% (hardly supprising I suppose).

I've only been running this rig a bit over a week, got 17,500 credits and another 15,000 pending.

 

[TheTross]

Added. Welcome to the team Chris, and thanks for the report

 

[edgedemon]

At the moment no, I have spent a lot of time playing with it (and learnt a lot about BOINC xml config files!) but until I install better cooling in my case, I am not running it for any extended period of time.
I have a package from OCUK arriving today so maybe this weekend I will rebuild and update this post with a yes soon

Chris - you can get past the scheduler issues by running a modified cc_config file, but you will have to spend some time playing with %CPU time to get a stable machine...

 

[EvilOne]

Hope they soon do a version of the client that will work on all graphics cards, and not just nvidia. Seems silly to make it using cuda now that the open sorce eqivent has being released.

 

[robinx]

I am, although also not happy about how its working right now, so I'm not happy leaving it running when I'm not at my house .. I don't know how to 'play' with xml files at all either, so need to do more reading up when I have time ... would rather they just released it when it was ready!

 

[ChrisLX200]

The CUDA app is really buggy though, I just got about 30 WUs and the first 10 all had computation errors. I aborted the remainder and probably won't bother again until an update is issued.

 

[Rojin]

Yes I am, Rojin. Only problems I have are WU's with low VLAR, but these can be found easily enough. Have a GX2 running at 80deg which chomps away nicely. I have to manage my task list by enabling\disabling Astropulse to ensure I don't get too many of them.

 

[TheTross]

Good man . I'll add you to the news now.

 

[robinx]

I found disabling GPU in prefences, let it download normal wu's, then enabled GPU again, set it to use only 3 cores, and it then used the CPU for normal wu's, and GPU on the cuda ones - running out of normal wu's again now, so will try and do it again to keep things going ... don't like having to mess around with it this much though

 

[TheTross]

That sounds like a right pita. Would editing your preferences so that you have a larger cache of normal WUs help?

 

[ChrisLX200]

Hmm. I just installed 181.20 drivers and it seems to have broken CUDA. I've got 3 AP units running but the CUDA units just say Ready To Start.

Edit: maybe a driver crash, reboot and it was processing them again (albeit 100% computational errors!). Aborted the remainder.

 

[robinx]

I've only had one computational error on the GPU when I had turned the fan speed down to 30% to sleep and forgot to turn back up when starting crunching again (as previously stated, not happy about leaving it crunching unattended atm) comp was on over night only so when I woke occasionaly I could refresh the Window 7 beta page!

 

[SimonB]

Hi,

Crunching on a pair of GTX260's running in SLi, hadn't been involved for several years but curiosity got the better of me.

Several issues to be sorted yet such as the computation errors with VLAR units and the scheduler needs to be able to differentiate between CPU/GPU profiles. Hopefully ver 6.5 will sort that when it's released from beta testing.

Cheers, Simon.

 

[loudbob]

Yes.

Usining:
Q6600 @ 2.4
3 x 9600 GT's

Works ok as long as not running anything else.
But crunching 8 AP's atm so GPU's idle.

 

[TheTross]

Cheers lads, both added

 

[loudbob]

Tross,

My seti username is 'loudbob'

original - i know

 

[Rojin]

Keep it down loudbob, you're making far too much noise as I issue a parp! on my way past

 

[ChrisLX200]

I've noticed whenever I get a 'compute error' with the CUDA client, although it doesn't completely crash the machine it becomes sluggish with jerky graphics, scrolling and mouse responses. A warm reset sometimes works (I think), but a cold reset definitely works. Might be corrupted GFX memory doesn't get cleared without a complete power down.

 

[SimonB]

I found disabling GPU in prefences, let it download normal wu's, then enabled GPU again, set it to use only 3 cores, and it then used the CPU for normal wu's, and GPU on the cuda ones - running out of normal wu's again now, so will try and do it again to keep things going ... don't like having to mess around with it this much though

Hi,

I may try this to keep the CPU busy rather than run AP's.

My rac which was steadily climbing is taking a nosedive at the moment due to too many AP's (41 left to do) and no cuda wu's, pending credit is now at 47,443.15 mainly waiting for AP's to be credited.

I have switched the idle GPU's to GPUGRID and left the remaing three cores to clear the cache which should take about 5 1/2 days.

Cheers, Simon.