Astropulse Opencl -- nVidia vs ATI/AMD


log in

Advanced search

Message boards : Number crunching : Astropulse Opencl -- nVidia vs ATI/AMD

Author Message
Profile Cliff HardingProject donor
Volunteer tester
Avatar
Send message
Joined: 18 Aug 99
Posts: 943
Credit: 50,357,636
RAC: 42,976
United States
Message 1331598 - Posted: 26 Jan 2013, 16:17:55 UTC

Please be tolerant with a senile old man, but since I have started processing nVidia_Opencl on one of my machines, I've noticed that I get very few of these work units, yet the scheduler always seems to have ATI/AMD units available. I know that the scheduler sends what is in the buffers at the time a send request is made, but it seems to me that my Opencl tasks are being discrimiated against.

1) If the scheduler has the ability to discern the distinction between the two, why is there more of one type being generated than the other? Doesn't the scheduler keep track of which type and how many of these tasks are sent?

2) If one is processing both types of AP tasks (CPU & GPU), shouldn't the scheduler generate/assign more of the Opencl tasks to a machine that requests them?

3) Is there a way to spoof the scheduler to send me more GPU AP tasks as opposed to the CPU type?
____________


I don't buy computers, I build them!!

ClaggyProject donor
Volunteer tester
Send message
Joined: 5 Jul 99
Posts: 4058
Credit: 32,795,739
RAC: 4,836
United Kingdom
Message 1331602 - Posted: 26 Jan 2013, 16:40:53 UTC - in response to Message 1331598.
Last modified: 26 Jan 2013, 16:49:14 UTC

Please be tolerant with a senile old man, but since I have started processing nVidia_Opencl on one of my machines, I've noticed that I get very few of these work units, yet the scheduler always seems to have ATI/AMD units available. I know that the scheduler sends what is in the buffers at the time a send request is made, but it seems to me that my Opencl tasks are being discrimiated against.

1) If the scheduler has the ability to discern the distinction between the two, why is there more of one type being generated than the other? Doesn't the scheduler keep track of which type and how many of these tasks are sent?

2) If one is processing both types of AP tasks (CPU & GPU), shouldn't the scheduler generate/assign more of the Opencl tasks to a machine that requests them?

3) Is there a way to spoof the scheduler to send me more GPU AP tasks as opposed to the CPU type?

There is no such thing as an AMD/ATI task, any task can be done by any suitable app, (ignoring the fact that MB VLAR tasks aren't sent to GPUs because they do them less efficiently)

1) Have you set the preference 'Use AMD/ATI GPU' to 'No'?, if you have, turn it back On, you don't have AMD/ATI GPUs so you can't ask for AMD/ATI GPU work,
all you get is a extra scheduler message saying there is work for a different GPU app but you're set your preferences to not receive it,
the message also doesn't say what type of work is available, it could be MB, it could be AP, or it could be something else.

2) Depends how many AP tasks the scheduler is allowed to send to a host in a single request, a year or so ago i could get 50 odd in one request, now i only get something like four or five AP tasks at a time,

3) the scheduler will fill the fastest device first, so make sure your host has a largish cache setting, so the GPU is always asking for work,
and that you're only got AP v6 selected, with 'If no work for selected applications is available, accept work from other applications?' also set to yes,
you'll get AP tasks in preference, and only MB tasks when there is no AP to send.

Claggy

Message boards : Number crunching : Astropulse Opencl -- nVidia vs ATI/AMD

Copyright © 2014 University of California