He explains two strategies for ensuring decentralization based on hash algorithm: keep changing the hash algorithm to prevent development of ASICs and ensure commodity generic computation devices (GPUs) are the only practical target or do not change the algorithm, to ensure that knowledge of how best to implement an ASIC for the algorithm becomes spread out and ASICs for the algorithm are as commoditized as GPUs. ZmnSCPxj argues that the former strategy has practical disadvantages such as developing new hash algorithms is not cheap, requiring coordinated hardforks over the entire network at an alarmingly high rate, and it puts too much power to the developers of the code. On the other hand, the latter strategy requires only survival during an intermediate period where ASICs are developed, but not yet commoditized, and during this intermediate period, the centralization pressure of ASICs might not be more powerful than other centralization pressures. He also emphasizes that non-ASIC-resistance is a non-issue because miner earnings are determined by cost of power supply regardless of whether the most efficient available computing substrate for the hashcash algorithm is CPU, GPU, or ASIC. ZmnSCPxj believes that improving the efficiency of energy transfer to increase the areas where cheap energy is available can solve the issue of miner centralization rather than stopgap change-the-algorithm-every-6-months. He suggests that the impact on existing miners could be made pleasant by gradually moving the block reward from the previous hash to the next (such that both are accepted with different rewards) and developing the frequency of introduction of new hashes such that once present-day ASICs are effectively obsolete anyway due to competition, new ones do not have time to develop.