Bitcoin Forum

I've been mining on a 5970/5870 trifire setup, overclocked 950/1200 and 900/1200, respectively.  I have been using poclbm through GUIMiner and with the recent release of the phoenix/phatk kernel for that frontend, I wanted to give it a try.  Here is a screenshot that I'd like explained:

http://img823.imageshack.us/img823/7885/stales.th.png (http://imageshack.us/photo/my-images/823/stales.png/)

You can see that there's a large difference between the stales produced by each kernels.

Here are my custom arguments:


-v -w 128 -f 1 (poclbm)
-k phatk AGGRESSION=19 WORKSIZE=128 BFI_INT VECTORS FASTLOOP=false (phoenix)

Poclbm produces something like 397/395/379 mh/s while phoenix is closer to 404/405/387 mh/s.

So, basically, my question is why does poclbm produces about 0.85% stales (on deepbit) over 200k stales while phoenix is closer to 2-4%?

While we figure this one out, I'll keep using poclbm.  3% higher rate is not worth 3% higher stales.

I encountered the same thing with 5830's

My hash rate increased by about 10Mhash/s, but I increased about 1-1.5% in stale shares.
That didn't happen until I overclocked the card.
Stock clocks using phatk with phoenix yeilds the same amount of stales from the norm.

I encountered the same thing with 5830's

My hash rate increased by about 10Mhash/s, but I increased about 1-1.5% in stale shares.
That didn't happen until I overclocked the card.
Stock clocks using phatk with phoenix yields the same amount of stales from the norm.