Thought I’d add some comments here, and give a little background on myself… I’ve been a GPU/CPU Miner since back in early '17, starting with ETH and then on to almost everything else that’s been available at one point or another. I remember mining XZC on linux when Todxx first released his AMD implementation. Part of what made Lyra2z such an awesome algo was how power friendly and cool it was on GPUs, definitely lost that with the move to MTP.
It really depends on who you’re appealing to. You have 4 options: ASIC, FPGA, GPU, and CPU.
Concerning ASICs, it’s fairly obvious they become a point of centralization when their manufacturers can corner a market on a new algo and have first access to hardware the average person doesn’t.
You have similar things with FPGAs. The amount of capital, technical sophistication, high devfees, all contribute to making it unlikely that the average person could reasonably and profitably support the network with an FPGA.
Which leaves you with GPUs and CPUs. Concerning CPUs, I’ll admit I was initially supportive of XMR’s move to RandomX as it seemed a way to democratize and decentralize the hashrate, and I have several ryzen CPUs. The experience so far has been poor for me across the board for RandomX. Whether RandomX for XMR or LOKI or Wownero, the profit margins on RandomX are so small, razor thin, and impractical that even on my Threadripper CPU I don’t even bother. I can speculate why, but no one really knows; typical answers are “server farms” or “botnets” or “AWS instances” for the cause of RandomX’s low profitability. So sure, anybody CAN do it and everybody HAS a CPU, but why would anyone do it? Unless they like being hot and sweaty and throwing money down the drain every month and burning out their computer faster. And it’s not scalable enough for an average user to support the network profitably… existing server farms which can quickly be switched between purposes/uses will maintain an incredible advantage and scale over the average user.
Which leaves GPUs for last I suppose. GPUs are obtainable for practically anyone who wants one, and for a person willing to do more and scale up, it’s not difficult to add GPUs to an existing mining rig (up to a point, obviously). However, in my opinion a GPU farm is a much more risky investment than a Server Farm, as a GPU Farm can’t be shifted to a different use like a Server Farm can if mining profitability tanks. There also is the aspect of constant depreciation of GPU hardware as newer generations are released. As long as a larger entity doesn’t find a way to game the system with an ASIC or FPGA, in my opinion they don’t have a significant enough advantage over a person who can run a rig at home. Sure, they may have better electrical rates, but they also have warehouse overhead and concerns of hardware depreciation, while the home miner has nearly zero “margin”, to the extent they need a home to live in anyways. When ETH Profitability came back I flipped my rigs back on and continued life as normal, and counted my blessings I didn’t have warehouse or employee overhead to worry about.
To the extent that XZC is trying to appeal to users like me to support the network, my vote will be GPU friendly algorithms. To the extent possible, algorithms that are memory-bound as opposed to compute-bound will not only be more power-friendly, but will also provide less advantageous implementations on FPGA’s/ASICs. This is what I think has made dagger-hashimoto/ethash so successful, and to a large extent even Cryptonight algorithms for quite a while. CN algorithms that require larger memory scratchpads (CN-Heavy) can and still are being successfully mined on GPUs. To that extent I think MTP was successful in how it had large memory requirements. On ProgPow this could also be achieved depending on the size of DAG chosen for the implementation, although, there is potentially a lot of orphaned 4GB hashrate that may be looking for a home come later this year when the ETH DAG grows too big. In terms of available options as of today, I’d say ProgPow is likely the easiest options as both AMD and Nvidia GPUs are competitive. One other comment I’d make is that smaller networks should be careful adopting the algorithms of other existing projects; ETC’s 51% attacks over the last month provide an adequate example. To that end, MTP has actually been pretty successful, not only because it’s unique, but the large proof requirements actually made Nicehash delist it; I know that many have complained in the past for how Nicehash provided an easy attack vector against smaller projects because of nefarious actors ability to purchase whatever extra hashrate they needed.
Think I’ve rambled on enough… but hope my commentary was helpful.