Of course not, finding shares is proof that you did the work needed to find solutions too, there's just a bug in the miner.


So maybe the GPU miner does work properly on some hardware. Or maybe it's just bad at finding solutions, not completely unable. Still, something is wrong.

As I've already said once, apparently without anyone listening, the GPU-miner is finding only shares, not solutions. I'll use more words:

Finding a share is useless. Competely useless. It always was. It's just that you can't do it without also having some probability of finding a solution (which creates a block). But that's "can't" in the math sense, obviously it's trivial to make code that doesn't actually report the solutions.

So the GPU miner will find shares, making the pool think it is contributing to finding solutions, when it really isn't. Look at the block overview on mine-pool. The percentage is going way up.

This makes you get basically no coins for your useful CPU-mining on the pool, and very little for your useless GPU-mining. Solo-mining still works fine (with CPU), since the difficulty doesn't rise from this. (Unlike the pool payouts you can't fool the difficulty since it's only based on solutions found.)

I'm pretty sure there's still something wrong with the GPU miner. It's finding only shares, not solutions. Or at least far more shares than solutions. Look at the statistics for recent blocks on a pool - there's way too many shares. And network difficulty isn't going up, which means more solutions aren't being found.

(So in practice running the GPU miner is stealing the coins people with working miners make on a pool.)

I would hope quite a lot. I'm getting several blocks per day on my desktop (AMD FX8350) using qubit-cpuminer (as found in the first post). (The built in miner is slower.)

In case anyone is unclear on how to do that:
./minerd -a qubit -o http://127.0.0.1:7799 -u QubitCoinrpc -p <from your conf>

Obviously if you changed the port or user in your conf you need to use what you set them to, but those are the defaults. (Well, they were the defaults for me, but I probably managed to do something unusual. minerd will give an error if you get it wrong at least.)

Is there some reason to think it is? I don't know much about quantum computers, but Merkle-Damgård is used all over the place, including sha-2 (bitcoin).

Not that I know of. If you compiled the miner yourself playing with compilation flags can help a little. If you have an AMD CPU you're probably limited by L1 cache, so I would suggest something like CFLAGS="-march=native -Ofast -msse4 -DSPH_CUBEHASH_NOCOPY -DSPH_SIMD_NOCOPY -DSPH_CUBEHASH_UNROLL=2". (Cubehash defaults to unroll 16, doesn't support none since even and odd laps are not the same.)