Backend upgrade scheduled for Wed Nov 18

And task status counts on the summary lists - thanks for that too.

Everything is in German now (since I am from Germany and have set so in the preferences), which is obviously good for native German speakers. However: I don't really like that since it can lead to confusion if people have to translate Boinc-specific words (e.g. RAC).

Is there any possibility to switch back to English without having to change my country in the preferences?

By looking at some numbers, there seem to be lots of anomalies. I suspect that the stats are just for the elapsed times without any consideration for the extended times when crunching multiple tasks.

As a single example, take a look at the AMD Tahiti series GPUs. Under Linux, they come in at No.2, and less than half the performance of Hawaii. Under Windows, they have a performance metric about 50% greater than Hawaii.

I would surmise that people who run Linux are more likely to fiddle with settings and be aware of how to extract performance from their GPUs by running multiple concurrent tasks. This would have the effect of blowing out the elapsed time of a single task, whilst producing multiple results in that blown-out time. I don't know if this is the main factor (or even a factor at all) but maybe Bernd may be able to enlighten us :-).

Wouldn't it be really useful if the recorded elapsed time could be 'corrected' for concurrency? I have no idea if that would even be possible, let alone the work involved if it were. For it to be possible, a result being returned would have to have the value under which the task was crunched returned by some mechanism as well. Once again, it will be interesting to see if Bernd has any ideas about this :-).

EDIT: A bit more evidence for concurrency skewing the results. I have a large number of 7850 (Pitcairn series) GPUs all crunching 4x, all under Linux. Look how low the performance metric for these is, compared to the Windows metric.

EDIT2: If there indeed is no accounting for the possibility of multiple concurrent tasks per GPU, there needs to be a warning on the Top GPU Models page to remind people that when it mentions "average elapsed time of tasks" it's not accounting for the fact that multiple tasks may have been completed in that average time. It should be spelt out that high end / high performing cards are going to appear to have less performance than they actually have because of this.