solo mining on CUDA triggers hardware error warnings

[jolan]

Noticed that while solo mining on testnet, I see a fairly steady stream of hardware error warnings.

I've never seen this while pool/stratum mining. This definitely doesn't seem to be a legitimate hardware error since if I change the pool mining card to stratum mining and vice versa, it only happens on whatever GPU is doing solo mining.

Need to check if this happens on mainnet too.

Not very familiar with the CUDA code but might be a regression from #108 or #116 since I don't remember seeing it when initial CUDA support landed. Could also possibly be related to #121.

[jolan]

Also happens on mainnet but seems less frequent.