Huh so basicly sidestepping the gpu issue entirly and essentially just using some other special piece of silicon with fast (but conventional ram). I still dont understand why u cant distribute a large llm over many different processors each holding a section of the parameters in memory.
I still dont understand why u cant distribute a large llm over many different processors each holding a section of the parameters in memory.
Because each weight in a layer influences each weight in the next layer, which means the bandwidth requirements are enormous and regular networking solutions are insufficient for that.
Not exactly. Digits still uses a Blackwell GPU, only it uses unified RAM as virtual VRAM instead of actual VRAM. The GPU is probably a down clocked Blackwell. Speculation I’ve seen is that these are defective and repurposed Blackwells; good for us. By defective I mean they can’t run at full speed or are projected to have the cracking die problem, etc.
Huh so basicly sidestepping the gpu issue entirly and essentially just using some other special piece of silicon with fast (but conventional ram). I still dont understand why u cant distribute a large llm over many different processors each holding a section of the parameters in memory.
Because each weight in a layer influences each weight in the next layer, which means the bandwidth requirements are enormous and regular networking solutions are insufficient for that.
Not exactly. Digits still uses a Blackwell GPU, only it uses unified RAM as virtual VRAM instead of actual VRAM. The GPU is probably a down clocked Blackwell. Speculation I’ve seen is that these are defective and repurposed Blackwells; good for us. By defective I mean they can’t run at full speed or are projected to have the cracking die problem, etc.