Matrix size is block height * 24576 bytes per thread. Currently that's 60 MB per thread or 560 MB for an i7 with 8 threads.
At block 1 million its 24.576 GB, at 10 million 240GB. With big Xeons it would be close to 1 TB at 1 million blocks.
I actually like the concept of increasing mining resource requirenents as block height increases because it protects against tech advancements,
faster CPU clocks, more threads, more mem, faster mem, etc.
The problem with the current algo is the curve is too steep. Memory requirements of the algo would probably increase faster than memory tech.
So what we have is an algo that is impeded from graduating to GPU* and ASIC while it's ability to mine on low end systems diminishes over time.
It's squeezed at both ends.
* I'm not familiar with GPU mining specifics but typically they have many more threads than CPUs so I assume greater memory usage.
Someone who knows please correct me.
So its turns out the RAM requirements could be more restricting, than CPU requirements. But from the other hand, with 10 min block target, block 1 million will be after 19 years!! I suppose the RAM size of 24GB+ or even 1TB will not be any problem by then...