Reactions from ranked influencers
2 posts
this would essentially be the consumer version of the grace blackwell platform with universal memory and storage bypassing the southbridge and the pcie lane, but i doubt nvidia would do this without pressure (i hope someone else does and that its cuda native)
@jun_song Only because M.2 SSD bandwidth sucks. If a GeForce 5090 had 4TB of Flash with sufficiently many pins in parallel, that would run any leading edge model awesomely. If NVIDIA doesn’t build this architecture, someone else will, and soon.
@jun_song Only because M.2 SSD bandwidth sucks. If a GeForce 5090 had 4TB of Flash with sufficiently many pins in parallel, that would run any leading edge model awesomely. If NVIDIA doesn’t build this architecture, someone else will, and soon.