Improving LLM memory contention will allow LLMs to use more memory.

> Though the other day I learned there are many technologies for "RAM"

I'm an advocate of sticking a $5 16Gb Optane stick from eBay on a $10 M.2 to PCIe 1x adapter from eBay. Set it up as swap in Windows or Linux. Or pay $200 for a 16GB stick of DDR5.

ECC is actually slower. The hardware to compute every transaction is correct does add a slight delay, but nothing compared to the delay of working on corrupted data.

And waste all those PCI Express lanes?

Somewhere along the line a $10000 GPU has to be equivalent to using a finger to do arithmetic in the dust.