even 8x rtx pro 6000 is only 768GB of VRAM. IDK how anyone is going to run k3
Usually offloading experts to system RAM. DDR4 has gone up a lot, but on a 8-channel used Xeon motherboard or whatever, you can get tolerable mem bandwidth out of it.
This is why God gave us 1.58-bit ternary quants?
on the 1.5 tB macStudio that is going to be released next quarter
Free server racks for everyone when the bubble bursts!
Usually offloading experts to system RAM. DDR4 has gone up a lot, but on a 8-channel used Xeon motherboard or whatever, you can get tolerable mem bandwidth out of it.