Well, they clearly dropped the ball on this issue. I have tons and tons of invalids.
Does not bode well, for the upcoming move to Krembil.
Edit: Where is Uplinger in all this? Haven't heard a word from him, since the big OPNG stress test ended, back in April/May.

Edit2: OK someone woke up, and shut down GPU crunching it seems. Now, my computer only asks for CPU work.

Thanks for the explanation knreed. Looking forward to GPU work coming back, at a faster pace. Even if it's just for a short time.

Is there any update on changing the GPU client to write less frequently to solid state drives?

As I understand it, the [full] output produced is necessary both for validation and credit calculation, so it isn't going away. The trouble is that it appears to be extremely verbose (or, at least, it was when I last looked!)

When this topic came up in the past, I had a look at a running OPNG task to see what all the I/O was about, and the file(s) seem to be designed to be human-readable without any need for post-processing; lots and lots of text and white space as well as the actual data...

One possible mitigation might be to borrow a technique I first came across in the days when mainframe computers wrote error logs to magnetic tape :-) In order to condense them as much as possible, the output consisted of the error number and a list of parameters, and the programs to display or print the errors would look up the error number to find a format string which would then be used to output the full text. (It might, however, make more sense here to use code words (acronyms? abbreviations?) instead of numbers in this case.)

Unfortunately, at the moment there's no data to look at, so I can't tell how much of a reduction in size might be achieved against the current file format by doing something like that to the OPNG output. In an ideal world, the files would not be expanded out unless/until the human-readable output was required, thus reducing file sizes throughout the process!

And, presumably, the programs that currently analyse the dialogue file (and others?) at the WCG end might not need major overhaul to process the condensed format!

@knreed - any chance something like this could be done as a side-effect of the OPNG down-time? I appreciate that it would require an application version change (so the validator [et cetera] would know what format the file was in!) and a beta test, so perhaps not...

Just a thought (and probably offered in completely the wrong place...)

Cheers - Al.

[Edit: added question to knreed]

The output file doesn't have to be any smaller, it just shouldn't write to the disk faster than the user allows. It can be stored in RAM.