Just have one PC on the gpu app with 3 GTX295's, some have started to validate but about 30 to 40% fail due to errors
Einstein@home News Q1 2011
- Thread starter derk
- Start date
Just have one PC on the gpu app with 3 GTX295's, some have started to validate but about 30 to 40% fail due to errors
What about if you run just one wu per gpu? You don't even need the app_info to do that.
The GTX295 only has 895MB of memory per gpu so I wouldn't run more than 2 at a time.
Reading up a bit, it seems that the 2xx series gpus don't take kindly to running more than one app at a time so stick with one per gpu on there, also if you load up 6 E@H gpu apps you are going to have to start dedicating cores
The GTX295 only has 895MB of memory per gpu so I wouldn't run more than 2 at a time.
Reading up a bit, it seems that the 2xx series gpus don't take kindly to running more than one app at a time so stick with one per gpu on there, also if you load up 6 E@H gpu apps you are going to have to start dedicating cores
Last edited:
Quick Question RE: CPU Blocks - Specifically Watercool HK Rev. 3
Does it matter which way round the flow through the block is?
Hopefully, the wc'ing project should start tomorrow
MB Block & Rad arriving today - everything else tomorrow (UPS dependent).
Does it matter which way round the flow through the block is?
Hopefully, the wc'ing project should start tomorrow
MB Block & Rad arriving today - everything else tomorrow (UPS dependent).
Just realised I'm using that app info with cpu units running as well - could that be the problem
I have 6 gpu units running using 0.20 cpu resource each leaving me with enough cpu resources the run an additional 3 cpu units.
The cpu units do not have any problems.
Ooh, Einstein news! Cheers halz
I can also say that the new app is more efficient than the older one but there is currently an issue with validation in that, in some instances, tasks completed on different hardware (i.e. CPU vs GPU) will not validate against each other, even though both tasks are actually valid. As far as I know though, the Einstein admin are aware of the problem and are looking into it.
Could this be the problem?
The one failure I had was a GPU wu validating against a CPU wu......
217673412 3908045 2 Feb 2011 3:25:16 UTC 2 Feb 2011 22:41:12 UTC Completed, marked as invalid 5,312.70 1,357.80 10.27 0.00 Binary Radio Pulsar Search v1.05 (BRP3cuda32)
217835813 3864080 3 Feb 2011 13:07:46 UTC 5 Feb 2011 4:41:53 UTC Completed and validated 96,864.09 65,572.28 244.82 500.00 Binary Radio Pulsar Search v1.05 (BRP3SSE)
217673413 3710032 2 Feb 2011 3:26:58 UTC 3 Feb 2011 3:12:29 UTC Completed and validated 62,104.23 55,932.18 311.29 500.00 Binary Radio Pulsar Search v1.05 (BRP3SSE)
Mine is the first instance.
In every case they are all 'Error while computing' and the amount of time before the error occurs is always different.
Some last for a while before the error occurs and others error out almost straight away
EDIT: Some additional information on some of the error work units........
PM0032_03511.dm_328_1
[07:57:38][13868][ERROR] Error during CUDA HS/HSG kernel launch and/or device synchronization (error: 700)
[07:57:38][13868][ERROR] Demodulation failed (error: 1015)!
07:57:38 (13868): called boinc_finish
PM0033_00711.dm_272_0
[11:13:39][23440][ERROR] Error during CUDA device->host time series mean transfer (error: 700)
[11:13:39][23440][ERROR] Demodulation failed (error: 1008)!
11:13:39 (23440): called boinc_finish
PM0032_034D1.dm_420_1
[08:11:37][15584][ERROR] Error during CUDA host->device HS power spectrum data transfer (error: 700)
[08:11:37][15584][ERROR] Demodulation failed (error: 1007)!
08:11:37 (15584): called boinc_finish
PM0032_03511.dm_364_1
[05:45:36][11932][ERROR] Error freeing CUDA HS device memory (error: 700)
[05:45:36][11932][ERROR] Demodulation failed (error: 1010)!
05:45:36 (11932): called boinc_finish
Even more
Some last for a while before the error occurs and others error out almost straight away
EDIT: Some additional information on some of the error work units........
PM0032_03511.dm_328_1
[07:57:38][13868][ERROR] Error during CUDA HS/HSG kernel launch and/or device synchronization (error: 700)
[07:57:38][13868][ERROR] Demodulation failed (error: 1015)!
07:57:38 (13868): called boinc_finish
PM0033_00711.dm_272_0
[11:13:39][23440][ERROR] Error during CUDA device->host time series mean transfer (error: 700)
[11:13:39][23440][ERROR] Demodulation failed (error: 1008)!
11:13:39 (23440): called boinc_finish
PM0032_034D1.dm_420_1
[08:11:37][15584][ERROR] Error during CUDA host->device HS power spectrum data transfer (error: 700)
[08:11:37][15584][ERROR] Demodulation failed (error: 1007)!
08:11:37 (15584): called boinc_finish
PM0032_03511.dm_364_1
[05:45:36][11932][ERROR] Error freeing CUDA HS device memory (error: 700)
[05:45:36][11932][ERROR] Demodulation failed (error: 1010)!
05:45:36 (11932): called boinc_finish
Even more
Last edited:
I think you are trying to crunch the wrong type of wu's with the wrong client maybe?
Use loudbobs app_info if you want to do CPU as well, just make sure to remove the erroneous spaces
See I have my Einstein preferences I have the following set:
Use CPU: No
Use GPU: Yes
Run CPU versions of applications for which GPU versions are available: No
So I only crunch GPU
The other option as you probably don't want to run two apps on the G200's is to just delete the app_info.xml and let the standard client handle everything.
Use loudbobs app_info if you want to do CPU as well, just make sure to remove the erroneous spaces
See I have my Einstein preferences I have the following set:
Use CPU: No
Use GPU: Yes
Run CPU versions of applications for which GPU versions are available: No
So I only crunch GPU
The other option as you probably don't want to run two apps on the G200's is to just delete the app_info.xml and let the standard client handle everything.
Last edited:
Touch wood all seems ok now running 3 tasks per 470 and will note the completion times, watch how fast i change back to 2 tasks per gpu if they do take 20hrs
The few i done yesterday have validated ok, if memory serves me right i thought it used to be about 250 credits per gpu task they are showing as 500 am liking Einstein even more now
Seti units crashing all over the place, best leave that well alone while i have at least 1 project working.
The few i done yesterday have validated ok, if memory serves me right i thought it used to be about 250 credits per gpu task they are showing as 500 am liking Einstein even more now
Seti units crashing all over the place, best leave that well alone while i have at least 1 project working.
Much happier now, stopped all cpu work units - PC now much more responsive than it was.
Just watched a few more complete .......sweet
As things stand on gpu only I will get through 6 units in 1 hour 45 minutes - which aint to shabby
UPDATE: Well another 4 hours have passed and no more errors - but everything has gone to 'pending'
Just watched a few more complete .......sweet
As things stand on gpu only I will get through 6 units in 1 hour 45 minutes - which aint to shabby
UPDATE: Well another 4 hours have passed and no more errors - but everything has gone to 'pending'
Last edited:
Ta muchly Mr Loudbob sah, and dont think for one minute that lets you off the photo hook
**Edit** It's all coming back now, my memory aint as bad as i thought. Could not find any app_info so just changed it in client_state voila
Some bits'n'bobs arrived today:
EK-RE3 Motherboard Block & Koolance High Flow 360 Radiator & Fittings
The rest is still holidaying abroad - so probably take a few days to get here.
Once arrived I shall take a few photos - and some of the current set-up.
Was a little dissappointed with the rad - the 12 fixing screws that came with it were supposed to be 30mm long but they are just over 25mm (only just fit through a standard fan). Will need some new ones so they can accomodate the extras I got (anti-vibration neoprene gaskets).
Temps prior to wc'ing (i7 930 with H70 @ 4.2GHz):
CPDN: max 82
Einstein:max +95
Last edited:
Sounds real nice Mr Loudbob sah can't wait to see it all up n running
On Einstein front am a bit nervous, since i got the app_info correct (thank you Mr Biffa) everything is pending so far had no results since i applied it....all pending
**Edit** I take that back a few from yesterday and some from today have validated
Still at 500 credits per task i aint moaning
On Einstein front am a bit nervous, since i got the app_info correct (thank you Mr Biffa) everything is pending so far had no results since i applied it....all pending
**Edit** I take that back a few from yesterday and some from today have validated
Still at 500 credits per task i aint moaning
Last edited:
A wee taster:
http://img716.imageshack.us/img716/8080/bitsu.jpg
Rest of kit arriving Thursday - confirmed.
http://img716.imageshack.us/img716/8080/bitsu.jpg
Rest of kit arriving Thursday - confirmed.
The main reason was that it was only 30mm thick...
Once fitted, with fans attached it will still be out of view when looking throught the side panel window.
Availability also played it's part - alternative was the swiftech but stock was hard to find (when I orderd).
Hopefully, once I start to fit GPU's all will become clear - I just hope it works and as I don't know yet I ain't gonna disclose my plans.
Once fitted, with fans attached it will still be out of view when looking throught the side panel window.
Availability also played it's part - alternative was the swiftech but stock was hard to find (when I orderd).
Hopefully, once I start to fit GPU's all will become clear - I just hope it works and as I don't know yet I ain't gonna disclose my plans.
Team update
Highest Team RAC in the UK
At this rate we will be the number 1 UK team on total credit by the end of the year
Team RAC doubled since the beginning of January .....and rising
Ba now 3rd in the UK on RAC, 30th in the world on RAC and rising
All is good at Team OcUk
Highest Team RAC in the UK
At this rate we will be the number 1 UK team on total credit by the end of the year
Team RAC doubled since the beginning of January .....and rising
Ba now 3rd in the UK on RAC, 30th in the world on RAC and rising
All is good at Team OcUk