4 ms·
M1 is a unified memory architecture, so everything goes through the TLB. Although these days most GPUs can access some unified memory and have TLBs.
by rayiner 5y ago
M1 is a unified memory architecture, so everything goes through the TLB. Although these days most GPUs can access some unified memory and have TLBs.
- monocasa 5y agoWhile the AGX does have MMUs and TLBs, unified memory and TLBs are orthogonal. For instance the Raspberry Pi's SoC has unified memory, but no MMU/TLB on the GPU. That IP block can only access physical memory.