1. I’ve been experimenting with a memory-hard PoW design for a cryptocurrency protocol and I’m trying to figure out whether the core assumptions are fundamentally flawed before investing more time into it.
2.  
3. The broader protocol idea is a CPU-oriented PoW + staking system intended to reduce industrial mining concentration while avoiding fixed supply schedules and governance-controlled monetary policy.
4.  
5. But the PoW assumptions are the part I distrust most.
6.  
7. Very simplified overview of the current prototype (“Evo-OMAP”):
8.  
9. * miners generate a large memory dataset derived from previous block state
10. * mining traverses pseudo-random memory locations
11. * each lookup affects future traversal paths
12. * execution is intentionally branch-heavy and data-dependent
13. * performance becomes dominated more by memory latency/cache behavior than raw arithmetic throughput
14.  
15. Current prototype uses ~256 MiB working memory per mining instance, though that parameter is still experimental.
16.  
17. The hypothesis being tested:
18. forcing irregular memory access patterns + branch divergence may reduce the optimization advantage of ASIC pipelines and massively parallel GPU execution relative to general-purpose CPUs with strong cache hierarchy and DRAM bandwidth.
19.  
20. I’m aware most “ASIC-resistant” systems eventually admit specialized hardware anyway, so I’m trying to understand whether the assumptions here are meaningful or mostly wishful thinking.
21.  
22. Questions I’m struggling with:
23.  
24. * How meaningful is branch divergence against modern GPUs today?
25. * What hardware optimization paths am I most likely underestimating?
26. * Do memory-hard traversal systems inevitably converge toward efficient ASIC implementations anyway?
27. * Is “ASIC resistance” even the right framing, or is minimizing specialization gradient more realistic?
28. * What failure modes usually blindside custom PoW designers?
29.  
30. I’ve been studying RandomX, CryptoNight, Ethash, Argon2-related work, and memory-hard function literature, but I assume there are important prior results I’m missing.
31.  
32. Mostly looking for criticism from people who understand hardware, cryptography, distributed systems, or adversarial compute design.
33.  
34. Prototype repo:
35. https://github.com/AdamBlocksmith/evo-omap
36.