2) Strict FIFO does not allow long queues to be used or for Resource Share to be respected if more than two GPU projects are run at the same time.

How about allowing specific GPUs to be enabled/disabled ie disable GPU1 on a gtx295 while leaving GPU0 enabled

How about being able to disable screensavers on some GPUs (such as those used by GPUGRID) but not others (such as those actually connected to a monitor, if there is no overlap) in multi-GPU situations?

Also, if we can first persuade Nvidia to add a function to their drivers to detect whether the monitor is connected and turned on, an option to automatically disable the screensaver on a GPU if the monitor connected to it is turned off or none is connected. (Leave it an option in case some monitors are unable to tell the driver that they are turned on.)

How about a function to detect whether the BOINC screensaver is about to run or currently running, so that GPUGRID has the option to pause the workunit, wait until the screensaver has either blanked the screen or stopped running, then restart from a checkpoint?

How about an option, applicable mainly to laptops, to set a different amount of time BOINC can continue running without user actions when on batteries, and when on wall power instead? This should include the option to allow running indefinitely when on wall power.

How about an option for GPU workunits for the CPU part to include a parameter for telling BOINC that it wants to be responsible for deciding how long to wait before deciding whether to abort if the GPU is unavailable at present, instead of having BOINC immediately abort the workunit automatically?

This is the king of things that should be reported to BOINC directly.
Here is for project specific issues.

An awful lot of BOINC's behaviour is still inherited from the earliest days: implicit assumptions that there is only one computing resource (the CPU, even if there are multiple copies of it), and that each project only has one kind of application to run on that CPU.

I'd like to see a lot more of BOINC's control functions (both user controls, and internal automatic controls) reviewed and relocated to take account of the new multi-resource, multi-application environment.

We had a big step forward recently, when 'use CPU', 'use NVIDIA', 'use ATI' were made into preferences which actually controlled work request behaviour. Here's what it looks like, for those who haven't seen it yet:

11/10/2009 11:41:39 GPUGRID [wfd] CPU: fetch share 0.00 debt -404704.37 backoff dt 0.00 int 480.00 (overworked) (blocked by prefs)
11/10/2009 11:41:39 AQUA@home [wfd] NVIDIA GPU: fetch share 0.00 debt 0.00 backoff dt 0.00 int 0.00 (blocked by prefs)

This is the king of things that should be reported to BOINC directly.
Here is for project specific issues.

It has been, and it is a project specific issue in that you, like MW, could lose significant capability because Resource Share is over-ridden by strict FIFO ... but, up to you if you are not interested in fighting for your proper Resource Share ...

Hi!

I am switching right now from nVidia to ATI, so it is naturall that if I want support GPUGRID I, and other users with ATI needs app.... for ATI....
____________
POLISH NATIONAL TEAM - Join! Crunch! Win!

I would like to see some manual controls, such as processor affinity and the ability to limit the number of CPU cores a given project can use at once. Some projects use huge amounts of ram and only work effectively if limited to one or two cores of a quad. Also the ability to assign a backup project(s) that will DL and run one WU at a time only as needed and only if the main project runs out of work.