Not sure how practical it is but would it be possible for neuro to be distributed on viewers computers so reduce to upkeep cost and allow for neuro and evil to be on 24/7 and what would the viability of something like this be and would it increase performance at all?
#Distributed Neuro(l) Net
molten vessel
snow storm
I think it would require far too much data transfer between devices
tall plover
I think it would require far too much data transfer between devices
I dont think its the data transfer, but the latency. the RAM and vram access is measured in nanoseconds, whereas network latency is measured in milliseconds - literally millions of times slower. AI datacenters need optical connections to transfer huge amounts of data just a couple inches or feet. Folding @ Home is a distributed network thats very very powerful, but doesnt require data to be sent and recieved with the responsiveness of system RAM (which is considered stupid slow compared to vRAM).
Vedal would be best off with something like a GB100, if a PCIe version ever becomes available. I doubt he'd be able to afford to purchase an NVL72 (its a whole server rack of AI specific Blackwell [RTX 5000] GPUs). While Blackwell is... abysmal for gaming, its apparently quite the uplift for AI.
snow storm
I dont think its the data transfer, but the latency. the RAM and vram access is ...
That’s what I meant
oak sandal
Neuro and evil but it's a virus