Post

Nick Fisher

@hydroxide.dev

ben3d.caAn M5 Max using Samsung's LPDDR5-PIM would generate 650 tokens/secondReplace the M5 Max's LPDDR5X with eight Samsung LPDDR5X-PIM packages and the memory could deliver 4.9 TB/s where the weights live. The right parallel mapping could push an 8B model toward 650 tokens p...

27/8/26, 3:52 am·1 repost·3 likes