All platforms...

Two major changes of significance. There was a memory leak which means that the client grows in size over time and in some cases crashes ... this SEEMS to have been fixed. At least the reports we have so far indicate that there is grow and shrinking going on that is more reasonable ... second the GPU memory test that eventually caused BOINC to stop launching tasks on the GPUs was removed (commented out, it may be restored later) ...

6.10.55 Change Log:

- client: make GPU available RAM measurement #ifdef-selectable,
and default it to off
- client: if we print available GPU RAM (which we now don't)
have a separate timer per GPU type

6.10.54 Change Log:

- client: enable bandwidth quota limit only if both
#MB and #days are nonzero.
- client: fix memory leak
- client: on resume from net bandwidth quota suspend,
apply same random backoff to all transfers.
- client: when changing ncpus via config file,
don't modify host_info.p_ncpus
- client: show effective #CPUs separately from physical #
- client: clear config before rereading config file.
This means that, e.g. if you remove <ncpus>
it will revert to the actual number.
However, it also means that any config params
set via command line will be overridden.
- client: fix crashing bug.
Deleting records in ~COPROCS() was bad,
since we copy HOST_INFO to APP_INIT_DATA,
leading to a double deletion.

Actually its already superceeded. See this message thread in Number Crunching.

It seems they may have cracked the GPU-starving bug. Operative words may have :-)

---

BOINC blog

It is a minor tweak, in .56, the major change is in 6.10.55 where they commented out the memory size detect code which was causing something bad to happen when called repeatedly ... the more it was called, the sooner it quit returning a value ... no value, BOINC stopped using the GPU ...

When they commented out the code they added a small section to just fill the arrays with fixed values and forgot to set a boolean flag ... anyway, that is the change in .56 that amounts to anything ...

I have been up for more than a day and a half with .55 and now .56 on 4 systems and no lost GPU yet ... so, I think they may have it ... won't say for sure because I was getting random failures of the same ilk with 6.10.45 but it would take a week or more of uptime before you would see it ... so ... 6 more days to go ... :)

Oh, don't forget the memory leak ...