How to make RandomX ASIC resistant forever: memory scaling schedule

25D Ago
So we all know things like Ethash and Progpow and other reasonably successful, asic resistant algorithms, exist for the GPU that work primarily by increasing memory requirement over time. But the problem is they are not perfectly asic resistant. Why? Well I think it is a very very somewhat stupidly simple reason. The memory requirement does not scale with moore's law. Memory has scaled with moores law []( (btw that 2007 article was worried we would hit a 25 nm barrier, which we surpassed no problem). So why not utilize that? The biggest risk is that we fall significantly below moores law for memory and thus emission of coins decreases slowly over time. Isn't this ok for a tail emission coin like monero? I think it would be. Actually come to think of it, the difficulty adjustment would make sure we are emitting the same amount of coins at all times, so I don't think there is a risk coin production slows down even if we don't keep up with moores law, we would just risk loosing "too much" hashrate. Anyway the proposal is simply add a built in doubling of memory requirement every 2-3 years, perhaps 2.6 years (683,280 blocks) to match for a more conservative koomey's law (moores law doubles every 2 years which is more aggressive). This should keep up with consumer chips and leave behind asic development. This increase in memory req can be done slowly over time or in jumps every 2.6 years. Am I wrong? Is this just something algorithm designers overlooked until now? Is it too risky? Let me know your thoughts! Links: Audit that shows moores law is an immediate risk to asic resistant algos: [](