The biggest thing Monero would benefit from right now, is a GPU miner. CPU-only is a guaranteed way to put the majority of the coins in the hands of criminals/botnet owners.
Will Monero ever have a GPU miner?
CryptoNote is about more than just privacy. It's about freedom. That's the point of egalitarian mining. Live in an oppressive state? Fuck you, government! Anyone with an internet connection can download the miner and receive a fair share of money. What's the point of anonymous transactions if only priviledged individuals can get the coins? What's the point of a fair launch and fair distribution if the rich get disproportionately richer via priviledged mining?
I'm sorry if you have a snazzy GPU farm that you're dieing to take advantage of, but Monero doesn't play favorites. It goes against the entire philosophy of this currency. I'm fairly certain that the developers and other contributing parties agree with me (speak up if you don't). Even if they are unanimously brainwashed into thinking GPU mining is a good idea, I will fight tooth and nail to make sure it never gets implemented. I swear to God, I will haunt this thread for as long as it exists and scare the bejesus out of anyone who dares to come up with some half-brained argument to enable privileged mining.
Actually I don't think a GPU can be faster one-on-one, as the PoW depends heavily on the characteristics of the CPU. If CryptoNight ASICS do become a reality, I believe the GPU mining phase will be skipped entirely.
Highly unlikely. This was said about scrypt (which became rapidly GPU mined) - in fact many things ring very similar. The main 'protections' against GPUs and ASICs that are claimed of CryptoNight are related to L3 cache speed and scratchpad size requirements. Guess what? The same was said about scrypt... I was waiting to discuss this in more detail until the AES-NI optimizations were released.
So let's jump in:
CryptoNight makes three decisions differently than scrypt does:
- Its central bit-mixing function is one round of AES, which is a single instruction on modern Intel CPUs, but requires quite a bit more computation on GPUs.
- It mixes in 128-bit quantities instead of the 512-bit quantities that scrypt does. This favors CPUs a bit because the wider memory bus of the GPU won't provide an advantage.
- It writes back to the mixing scratchpad to prevent LOOKUP_GAP style optimizations.
I said it earlier: I think CN is GPU-able. But these design differences *will* shrink the CPU-to-GPU gap compared to scrypt. My *hunch* is that GPUs will still have a slight advantage in cost-per-hash-per-second, but I'm honestly not sure. The designers of this PoW clearly learned from scrypt and made progress. I look forward to seeing what happens on this one - whomever designed the PoW was clever.
