// DBOINCP-300: added node comment count condition in order to get Preview working ?>
Ron Kosinski
Joined: 23 Mar 05
Posts: 57
Credit: 1,018,740,952
RAC: 939,052
12 May 2020 19:06:38 UTC
Topic 222451
(moderation:
)
I will be changing the GPU card in one of my boxes from a NVIDIA GTX-760 to a Radeon RX-580. Will I need to clear out my GPU work unit cache before the change?
yes, set NNT, finish your work. then close BOINC, purge your nvidia drivers, install your new card, install AMD drivers, and re-open BOINC. you'll have the smoothest transition that way.
One additional suggestion I'd add to the advice given by Ian&Steve C. is greatly to reduce your requested work fetch queue depth. I'd suggest 0.1 + 0.1 days. There is plenty of time to raise it once things have settled down after your transition.
If you want to install the new card before your old tasks run out, you can manually abort the old GPU ones. Just make sure you've downloaded at least a few batches of tasks for the new GPU. All the aborted tasks will count as failures; aborting them in large numbers will cause the server to temporarily stop issuing you more work until you start returning successful tasks again. If you've got some new GPU tasks on hand the recovery will go much faster than if you're stuck until a few CPU tasks finish.
One additional suggestion I'd add to the advice given by Ian&Steve C. is greatly to reduce your requested work fetch queue depth. I'd suggest 0.1 + 0.1 days. There is plenty of time to raise it once things have settled down after your transition.
+1
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
If you want to install the new card before your old tasks run out, you can manually abort the old GPU ones. Just make sure you've downloaded at least a few batches of tasks for the new GPU.
A very good suggestion but there's just one small thing to consider. You won't be able to download tasks for the different flavour of new GPU until you have actually installed that GPU.
So if you need to do things quickly and have a lot of old tasks to abort, the best plan is to abort the majority, leaving a small number - 4 would be more than sufficient. Completing those 4 after a big abort (but before swapping the GPU) will restore more than enough of your daily allowance so that there will be no risk of not being able to get sufficient tasks for the new GPU immediately after it is installed, recognised and the new app version downloaded.
If you want to install the new card before your old tasks run out, you can manually abort the old GPU ones. Just make sure you've downloaded at least a few batches of tasks for the new GPU.
A very good suggestion but there's just one small thing to consider. You won't be able to download tasks for the different flavour of new GPU until you have actually installed that GPU.
So if you need to do things quickly and have a lot of old tasks to abort, the best plan is to abort the majority, leaving a small number - 4 would be more than sufficient. Completing those 4 after a big abort (but before swapping the GPU) will restore more than enough of your daily allowance so that there will be no risk of not being able to get sufficient tasks for the new GPU immediately after it is installed, recognised and the new app version downloaded.
I thought you could:
1) Install the new card
2) Start the system - old GPU tasks will be stuck but new GPU ones download
3) Abort the old GPU tasks
4) Recover normal operation once the first new GPU tasks get returned.
It's been a year or so but I'm ~95% sure that's what I did when replacing an amd card with one from nvidia.
I want to thank everyone for their good ideas and suggestions. The change-over went fairly smoothly. One additional step I needed to perform was to do a "Project Reset" on the box that got the AMD cards to clear error messages I was getting in the "Event Log". It seems I needed to flush out the no longer used NVIDIA executables from my system.
yes, set NNT, finish your
)
yes, set NNT, finish your work. then close BOINC, purge your nvidia drivers, install your new card, install AMD drivers, and re-open BOINC. you'll have the smoothest transition that way.
_________________________________________________________________________
One additional suggestion I'd
)
One additional suggestion I'd add to the advice given by Ian&Steve C. is greatly to reduce your requested work fetch queue depth. I'd suggest 0.1 + 0.1 days. There is plenty of time to raise it once things have settled down after your transition.
If you want to install the
)
If you want to install the new card before your old tasks run out, you can manually abort the old GPU ones. Just make sure you've downloaded at least a few batches of tasks for the new GPU. All the aborted tasks will count as failures; aborting them in large numbers will cause the server to temporarily stop issuing you more work until you start returning successful tasks again. If you've got some new GPU tasks on hand the recovery will go much faster than if you're stuck until a few CPU tasks finish.
archae86 wrote: One
)
+1
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
DanNeely wrote:If you want to
)
A very good suggestion but there's just one small thing to consider. You won't be able to download tasks for the different flavour of new GPU until you have actually installed that GPU.
So if you need to do things quickly and have a lot of old tasks to abort, the best plan is to abort the majority, leaving a small number - 4 would be more than sufficient. Completing those 4 after a big abort (but before swapping the GPU) will restore more than enough of your daily allowance so that there will be no risk of not being able to get sufficient tasks for the new GPU immediately after it is installed, recognised and the new app version downloaded.
Cheers,
Gary.
Gary Roberts wrote: DanNeely
)
I thought you could:
1) Install the new card
2) Start the system - old GPU tasks will be stuck but new GPU ones download
3) Abort the old GPU tasks
4) Recover normal operation once the first new GPU tasks get returned.
It's been a year or so but I'm ~95% sure that's what I did when replacing an amd card with one from nvidia.
I want to thank everyone for
)
I want to thank everyone for their good ideas and suggestions. The change-over went fairly smoothly. One additional step I needed to perform was to do a "Project Reset" on the box that got the AMD cards to clear error messages I was getting in the "Event Log". It seems I needed to flush out the no longer used NVIDIA executables from my system.