I'm really curious about Sampler Feedback Streaming. So you have a 4k texture (for example) sitting on your ssd. That game image is now directly addressable like a page of virtual memory. The GPU then needs to sample the texture, so that means it decides which page table entry of the texture it needs to sample and loads it from the ssd, somehow, instead of loading the whole texture. I wonder what the page table entry size is, and how fast their new system can stream that table entry into RAM. That's more of a per-frame operation than I was expecting the SSD to be relied upon for.