UPDATE:
The ASIC's usually return a nonce if at least the first 32 bits of the found hash are 0 (I think this is called a Work Unit).
But something doesn't fit with 4.10.0 as it finds only a few nonces (ie. the "Proof" lines printed).
I'm testing this on AntMiner S3. Or is the definition of such a WU differrent under AntMiner S3?
Dear author, can you please comment on this important issue as it could be a big bug affecting many people, thx.
The S3 code in master cgminer filters out low diff nonces.
Ah, good to know about that filtering.
What rule gets used in the filtering? Where in the code is it done?
Thx
UPDATE:
ok, I think I found the location in driver-bitmain.c:
diffbits = (int)floor(log2(DATAW(witem)->work->sdiff));
if (diffbits < 0)
diffbits = 0;
// Limit to 4096 so solo mining has reasonable mining stats
if (diffbits > 12)
diffbits = 12;
Replacing this block of code with just "diffbits = 0;" seems to revert it to the default behavior.
But this of course leads to eating up much CPU cycles and the hash rate of the device falls significantly.
Yes, the filtering makes much sense; I'll need to modify my test-code to adapt to the original filtering...