Considering I just started Rosetta myself, maybe you are running out or RAM. I certainly did instantly that I started that project. Reduce the number of Rosetta tasks to leave yourself at least 3-4GB of memory for your other applications.
GPU#2 was the middle card that was cheek to butt with the other cards and was't cooling very well since it was clogged up with dust and hair after not being blown out in over 2 months.
De-dusted with the leaf blower fixed things right up.
Took out a hundred or so Einstein and GPUGrid tasks each before I caught the error.
=====================
I think you are right. I have had to take a gpu offline since this problem occurred on the Intel. I am going to go back to running "everything" and see if it throw more errors.
Tom
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
I don't have any experience running Android on this project @Link. But I run the cpu apps very well on my Raspberry Pi with 1GB of global memory. They do take up a bit of storage space though. I am using about 4GB of storage. I have a small cache setting of 0.2 days and that nets me 6 tasks in my cache.
I have done some Einstein earlier but now (and tried a couple of times in the last year) it is only problems with Einstein. The multi cpu wu is only making problems, gpu usage is very low, both cpu and gpu newer completes and all the crunching is for nothing. Milkyway was much faster on the old pc (10 years old but upgraded with a used Xeon X3670 and used AMD 7950) than the new one so I think I will only use the old pc for Milkyway and the occational Seti reruns. RAC will rise 8-10X compared to when both pcs was crunching Seti.
I have done some Einstein earlier but now (and tried a couple of times in the last year) it is only problems with Einstein. The multi cpu wu is only making problems, gpu usage is very low, both cpu and gpu newer completes and all the crunching is for nothing. Milkyway was much faster on the old pc (10 years old but upgraded with a used Xeon X3670 and used AMD 7950) than the new one so I think I will only use the old pc for Milkyway and the occational Seti reruns. RAC will rise 8-10X compared to when both pcs was crunching Seti.
Welcome back! To maximize RAC, you can try this approach, below, for your RX 580 host. I'm not familiar with best settings for the AMD Radeon HD 7900, but others here can advise. Running multiple projects on a host is possible, but it would be simplest to not do so initially while setting up E@H run parameters. The information below assumes the host is running only E@H.
In Account>Preferences>Project, use these settings:
Preference set: #set to match your RX 580 host location found in Account -> Dashboard -> Computers...Location
Resource settings
Resource share:100
Use CPU:yes #set to 'no' if not running CPU GW app, below & notes
Use AMD GPU: yes
Beta settings
Run test applications?:no
Applications #select only these apps
Gamma-ray pulsar binary search #1 (GPU)
Gravitational Wave search O2 Multi-Directional #see notes!
Allow non-preferred apps:no
GPU utilization factor of FGRP apps: 0.33 #see notes!
Default set
Set used for new computers: no
Save Changes
Notes: Multiple concurrent FGRP tasks will increase GPU temperature, SO BE SURE to monitor temps and adjust fan speed as needed. Otherwise, leave GPU utilization at the default setting of 1, or maybe try 0.5 (2 concurrent tasks).
If "Gravitational Wave search O2 Multi-Directional" app is selected along with the GRP app, then you likely will need to limit project_max in app_config.xml. Using app_config.xml can be very useful, but I'd not advise jumping right into those weeds yet. The simplest option is start off by not allowing the CPU GW app. However, if you want, a link to configuring that file is here in a related forum posting. Initially setting project_max to a value of 4 will allow 3 concurrent tasks on the GPU (determined by GPU utilization factor of FGRP apps, above) and one CPU GW task. Running 3 concurrent FGRP tasks will keep your GPU pegged at near 100% utilization with relatively low CPU utilization. One CPU GW tasks will keep one core/thread at 100% utilization and increase core temps. As you assess host performance on a daily or weekly basis, you can raise the max value by increments of one to allow more CPU GW tasks if desired. I've found on my 4-core/thread system, however, that it can handle only a single CPU GW task while running unrelated GPU tasks without degrading overall performance. As you have found, simultaneously running related CPU and GPU apps for the same data sets (e.g. GW) gives seriously degraded system performance.
On my host, running a single CPU GW tasks increases system power usage about 20 Watts and yields an output of only about 4000-5000 Boinc credits per day (~5 hours/task). So, obviously, running CPU GW tasks is not a ticket to high RAC, but does allow us direct contribution to the gravitational wave analysis effort. (I've temporarily backed off running CPU GW tasks, in case you look and see no recent GW activity for me.)
It generally takes time (weeks, months?) to optimize systems to your liking, so be patient, search the forums, and ask questions.
Edit: A generally good practice is to set Computing preferences... in Boinc Manager to: Store at least 0.1 days of work, and zero days of Additional work. This will keep task queues short and nimble for configuration changes.
Ideas are not fixed, nor should they be; we live in model-dependent reality.
I have done some Einstein earlier but now (and tried a couple of times in the last year) it is only problems with Einstein. The multi cpu wu is only making problems, gpu usage is very low, both cpu and gpu newer completes and all the crunching is for nothing. Milkyway was much faster on the old pc (10 years old but upgraded with a used Xeon X3670 and used AMD 7950) than the new one so I think I will only use the old pc for Milkyway and the occational Seti reruns. RAC will rise 8-10X compared to when both pcs was crunching Seti.
Welcome back! To maximize RAC, you can try this approach, below, for your RX 580 host. I'm not familiar with best settings for the AMD Radeon HD 7900, but others here can advise. Running multiple projects on a host is possible, but it would be simplest to not do so initially while setting up E@H run parameters. The information below assumes the host is running only E@H.
In Account>Preferences>Project, use these settings:
Preference set: #set to match your RX 580 host location found in Account -> Dashboard -> Computers...Location
Resource settings
Resource share:100
Use CPU:yes #set to 'no' if not running CPU GW app, below & notes
Use AMD GPU: yes
Beta settings
Run test applications?:no
Applications #select only these apps
Gamma-ray pulsar binary search #1 (GPU)
Gravitational Wave search O2 Multi-Directional #see notes!
Allow non-preferred apps:no
GPU utilization factor of FGRP apps: 0.33 #see notes!
Default set
Set used for new computers: no
Save Changes
Notes: Multiple concurrent FGRP tasks will increase GPU temperature, SO BE SURE to monitor temps and adjust fan speed as needed. Otherwise, leave GPU utilization at the default setting of 1, or maybe try 0.5 (2 concurrent tasks).
If "Gravitational Wave search O2 Multi-Directional" app is selected along with the GRP app, then you likely will need to limit project_max in app_config.xml. Using app_config.xml can be very useful, but I'd not advise jumping right into those weeds yet. The simplest option is start off by not allowing the CPU GW app. However, if you want, a link to configuring that file is here in a related forum posting. Initially setting project_max to a value of 4 will allow 3 concurrent tasks on the GPU (determined by GPU utilization factor of FGRP apps, above) and one CPU GW task. Running 3 concurrent FGRP tasks will keep your GPU pegged at near 100% utilization with relatively low CPU utilization. One CPU GW tasks will keep one core/thread at 100% utilization and increase core temps. As you assess host performance on a daily or weekly basis, you can raise the max value by increments of one to allow more CPU GW tasks if desired. I've found on my 4-core/thread system, however, that it can handle only a single CPU GW task while running unrelated GPU tasks without degrading overall performance. As you have found, simultaneously running related CPU and GPU apps for the same data sets (e.g. GW) gives seriously degraded system performance.
On my host, running a single CPU GW tasks increases system power usage about 20 Watts and yields an output of only about 4000-5000 Boinc credits per day (~5 hours/task). So, obviously, running CPU GW tasks is not a ticket to high RAC, but does allow us direct contribution to the gravitational wave analysis effort. (I've temporarily backed off running CPU GW tasks, in case you look and see no recent GW activity for me.)
It generally takes time (weeks, months?) to optimize systems to your liking, so be patient, search the forums, and ask questions.
Edit: A generally good practice is to set Computing preferences... in Boinc Manager to: Store at least 0.1 days of work, and zero days of Additional work. This will keep task queues short and nimble for configuration changes.
I had given up Einstein for now but because of your reply I will give it another try on my newest computer. Happy easter (in these Corona times) to you all!
Everything looked nice. 13 cores with cpu and and 3 gpu task running nicely. Suddenly it goes from 90% complete on the GPU and starts again(?) "Waiting for memory" on the cpu tasks. I have 16 GB on the pc and 8 on the GPU. L leave it on and just see what happens...
Considering I just started
)
Considering I just started Rosetta myself, maybe you are running out or RAM. I certainly did instantly that I started that project. Reduce the number of Rosetta tasks to leave yourself at least 3-4GB of memory for your other applications.
Keith Myers wrote:Those are
)
=====================
I think you are right. I have had to take a gpu offline since this problem occurred on the Intel. I am going to go back to running "everything" and see if it throw more errors.
Tom
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
I don't have any experience
)
I don't have any experience running Android on this project @Link. But I run the cpu apps very well on my Raspberry Pi with 1GB of global memory. They do take up a bit of storage space though. I am using about 4GB of storage. I have a small cache setting of 0.2 days and that nets me 6 tasks in my cache.
OK, thanks @Keith Myers. The
)
OK, thanks @Keith Myers. The device will get a SD card tomorrow or so, than I can move NativeBOINC to it and see if it can work for Einstein.
.
Another SETI orphan here
)
Another SETI orphan here joining the fray. Hope I don't break anything :-)
I have done some Einstein
)
I have done some Einstein earlier but now (and tried a couple of times in the last year) it is only problems with Einstein. The multi cpu wu is only making problems, gpu usage is very low, both cpu and gpu newer completes and all the crunching is for nothing. Milkyway was much faster on the old pc (10 years old but upgraded with a used Xeon X3670 and used AMD 7950) than the new one so I think I will only use the old pc for Milkyway and the occational Seti reruns. RAC will rise 8-10X compared to when both pcs was crunching Seti.
I've just moved my small
)
I've just moved my small boinc operations here from SETI (exclusively since 2000). Hopefully issues obtaining and sending WU are now in my past.
Ole Kristian wrote:I have
)
Welcome back! To maximize RAC, you can try this approach, below, for your RX 580 host. I'm not familiar with best settings for the AMD Radeon HD 7900, but others here can advise. Running multiple projects on a host is possible, but it would be simplest to not do so initially while setting up E@H run parameters. The information below assumes the host is running only E@H.
In Account>Preferences>Project, use these settings:
Notes: Multiple concurrent FGRP tasks will increase GPU temperature, SO BE SURE to monitor temps and adjust fan speed as needed. Otherwise, leave GPU utilization at the default setting of 1, or maybe try 0.5 (2 concurrent tasks).
If "Gravitational Wave search O2 Multi-Directional" app is selected along with the GRP app, then you likely will need to limit project_max in app_config.xml. Using app_config.xml can be very useful, but I'd not advise jumping right into those weeds yet. The simplest option is start off by not allowing the CPU GW app. However, if you want, a link to configuring that file is here in a related forum posting. Initially setting project_max to a value of 4 will allow 3 concurrent tasks on the GPU (determined by GPU utilization factor of FGRP apps, above) and one CPU GW task. Running 3 concurrent FGRP tasks will keep your GPU pegged at near 100% utilization with relatively low CPU utilization. One CPU GW tasks will keep one core/thread at 100% utilization and increase core temps. As you assess host performance on a daily or weekly basis, you can raise the max value by increments of one to allow more CPU GW tasks if desired. I've found on my 4-core/thread system, however, that it can handle only a single CPU GW task while running unrelated GPU tasks without degrading overall performance. As you have found, simultaneously running related CPU and GPU apps for the same data sets (e.g. GW) gives seriously degraded system performance.
On my host, running a single CPU GW tasks increases system power usage about 20 Watts and yields an output of only about 4000-5000 Boinc credits per day (~5 hours/task). So, obviously, running CPU GW tasks is not a ticket to high RAC, but does allow us direct contribution to the gravitational wave analysis effort. (I've temporarily backed off running CPU GW tasks, in case you look and see no recent GW activity for me.)
It generally takes time (weeks, months?) to optimize systems to your liking, so be patient, search the forums, and ask questions.
Edit: A generally good practice is to set Computing preferences... in Boinc Manager to: Store at least 0.1 days of work, and zero days of Additional work. This will keep task queues short and nimble for configuration changes.
Ideas are not fixed, nor should they be; we live in model-dependent reality.
cecht wrote:Ole Kristian
)
I had given up Einstein for now but because of your reply I will give it another try on my newest computer. Happy easter (in these Corona times) to you all!
Regards
Ole Kristian A
Everything looked nice. 13
)
Everything looked nice. 13 cores with cpu and and 3 gpu task running nicely. Suddenly it goes from 90% complete on the GPU and starts again(?) "Waiting for memory" on the cpu tasks. I have 16 GB on the pc and 8 on the GPU. L leave it on and just see what happens...