I also wonder what this moment will mean for chip design going forward. The fact that we are so constrained on the GPU side along with the success of unified memory with apple silicon does make me wonder what chips in 3-5 years will look like.
I also wonder what this moment will mean for chip design going forward. The fact that we are so constrained on the GPU side along with the success of unified memory with apple silicon does make me wonder what chips in 3-5 years will look like.
Those that can will incorporate memory into their designs, and the Chinese will be a bigger part of the memory market worldwide going into the future, that alone will drive more tech companies to look at memory/ssd design differently than what they did before. Market conditions have changed you either change/adapt with it or you die.
I think memory and ssd design will end up being incorporated into the overall design of the SOC chip. And several companies that can probably will sponsor a existing company or build a foundry going forward. Once again, change or die.
Unified Memory simply isn't viable at the scale that a full Nvidia cluster operates at, and it would hamstring the parallelism that gives these clusters an edge. The current solution of Mellanox-style interconnect is probably still going to be the status quo in 5 or 10 years.