GPU to 72 status...

[Graff]

Down to 36 P-1 assignments this morning. Go to grab another 80 (# CPUs = 2,
# per CPU = 40). Download the worktodo.txt file, find there are 140 entries
in the file. In addition, the count of my assignments only goes up to 76.

Gareth

[chalsall]

Originally Posted by Graff

Down to 36 P-1 assignments this morning. Go to grab another 80 (# CPUs = 2, # per CPU = 40). Download the worktodo.txt file, find there are 140 entries
in the file. In addition, the count of my assignments only goes up to 76.

Sorry Gareth... I've obviously got a bug in the P-1 assignment code (TheJudger reported similar).

I've been away from the system for the last few days (deep in a complex computer vision project). I'll figure out what's going on today.

[Dubslow]

Originally Posted by chalsall

(deep in a complex computer vision project)

[OT] I'm really curious. What does that mean? [/OT]

[chalsall]

Originally Posted by Dubslow

[OT] I'm really curious. What does that mean? [/OT]

Without going into too much detail, a distributed SfM system.

Edit: And, as an aside, if anyone here has a Canon digital camera you should definitely check out the Canon Hack Development Kit. RAW files from cheap kit!!!

[kladner]

Originally Posted by chalsall

Edit: And, as an aside, if anyone here has a Canon digital camera you should definitely check out the Canon Hack Development Kit. RAW files from cheap kit!!!

That's really interesting! It does not apply to my camera (50D DSLR), but my partner has an SX20IS, which is listed. Getting RAW capability on that one would be cool.

[chalsall]

Originally Posted by kladner

That's really interesting! It does not apply to my camera (50D DSLR), but my partner has an SX20IS, which is listed. Getting RAW capability on that one would be cool.

Yeah. And RAW is just the beginning. Scripts, automatic bracketing, ultra-fast (and ultra-slow) exposure times, live histograms... The list goes on and on!

It's a little scary the first time you use it (Do I really want to update my firmware with this code (even though it doesn't actually touch the firmware)?) but I have not had any problems with it, and I haven't heard of anyone else having any problems either. And by setting the SD card(s) to be "bootable" (from within the system after the first manual boot) the code will run automatically every time you turn on the camera.

You do need a USB SD Card Reader (you can't upload the software to the SD card via the camera), but they're dirt cheap and handy to have anyway.

[chalsall]

Originally Posted by chalsall

I'll figure out what's going on today.

OK, just so everyone knows, this bug has been squashed. Stupid Programmer Error (as usual) -- I missed initiating (clearing) the string variable used to build the WorkTodo data which is passed to the script which downloads the WORKTODO.TXT file when asked. ModPerl is unforgiving that way...

The reason this bug has been lurking undetected for so long is I only missed it in the P-1 assignment script, it only manifested when someone used the "Download WORKTODO.TXT" function, and then only when another user had previously and recently used the same server context to get P-1 work.

See Dubslow -- Once A Month Bugs aren't limited to C...

As a side effect of debugging this, I added the ability to toggle assignments by work type on the View Assignments page (Under the "Select..." drop down at the bottom of the page). Needless to say, I was doing a lot of unreserving of P-1 assignments in my own account today....

[Graff]

Originally Posted by chalsall

OK, just so everyone knows, this bug has been squashed. Stupid Programmer Error (as usual) -- I missed initiating (clearing) the string variable used to build the WorkTodo data which is passed to the script which downloads the WORKTODO.TXT file when asked. ModPerl is unforgiving that way...

Just requested 56 P-1 exponents (2 CPUs, 28 per CPU). Downloaded
WORKTODO.TXT file. Worker 1 had 28 exponents, worker 2 only 12.
On the plus side, my count of assigned exponents went up by 40.
[Never mind, just saw the new limit of 40 per request...]

Gareth

[James Heinrich]

I just asked for TF assignments and came up with nothing on default settings (except number assignments requested)

Number of assignments: [100]
Optional range: [0] to [100000000]
Will factor to: [72]
Options: [Whatever makes sense]

However, changing "Whatever makes sense" to "Lowest TF level" happily assigned me some 60.1M assignments 70-72. I could guess that nothing makes sense at the moment, but probably more specifically, WMS is looking for <60M and/or 71-72, of which there are none (wild guessing) and has no fallback plan if that fails to match any (or sufficient) assignments?

[petrw1]

Originally Posted by James Heinrich

I just asked for TF assignments and came up with nothing on default settings (except number assignments requested)
However, changing "Whatever makes sense" to "Lowest TF level" happily assigned me some 60.1M assignments 70-72. I could guess that nothing makes sense at the moment, but probably more specifically, WMS is looking for <60M and/or 71-72, of which there are none (wild guessing) and has no fallback plan if that fails to match any (or sufficient) assignments?

There are at least 11,000 available in the 58-59M range at 71 bits.
http://www.gpu72.com/reports/available/

[chalsall]

Originally Posted by James Heinrich

I could guess that nothing makes sense at the moment, but probably more specifically, WMS is looking for <60M and/or 71-72, of which there are none (wild guessing) and has no fallback plan if that fails to match any (or sufficient) assignments?

Hmmmm... I have no idea why it would do that (not give you assignments for WMS), but it giving you 60Ms when asking for lowest TF level does since that is the lowest range with the lowest TFed candidates available.

I did make a change to that page a few days ago for >72 assignments; possibly ModPerl cached a version of the intermediate code during that process. I've rebooted Apache to flush all of ModPerl's caches; WMS makes sense again....