Putting aside ETH valuation, rising difficulty, and the specter of PoS somewhere down the road, the Ethereum hash algo will *never* be implemented on an ASIC. It was designed that way. ASICs are great for doing fixed calculations on a (small) block of data. However, the ETH algo is much more than calculations. To be able to perform the hash, you first most load the DAG data, which is current approaching 1.4 GB. That size increases every "epoch", which is approximately 30,000 blocks or, currently, -6 days. It increases because it changes, i.e. it has to be periodically loaded again. For each ETH hash, there are 64 *random* 128 *byte* reads of the DAG, with an intermediate FNV hash between each access. The ETH algo is not compute bound, it's memory access time bound. The limit of current memory cycle times is the limit of the maximum ETH hash rate. An ASIC can do absolutely nothing to improve that, nothing. Not to mention, current mining ASICs don't have any significant memory interface capabilities that could address the needs of the ETH algo. Saying you're at risk mining ETH with GPUs, because ASICs are going to come along, as they did with BTC, etc. is a complete fallacy, it's nonsense, and it reflects a total lack of understanding of what the ETH algo was designed to do, and does very well. The *only* viable way to mine ETH is with GPUs, and that won't change until PoS comes to be.
-Best Care
David
Propably stupid question, but isn't there faster memory than what is used in GPUs? That one, what is used in processors? What if someone builds a dedicated ETH miner using that? Or would that betoo expensive?
