> You can easily fit a 1 TB 3 Gbps NVMe SSD into a $1k PC budget, and you aren't going to notice the delta between that and the 5.5 claimed Gbps
Bearing in mind here, that it’s memory+storage architecture has a theoretical top of 22 Gb/ps, and they demonstrated recently, when assets are compressed using the “Octo” (naming?) lib, they were getting throughput of 12-18 Gb/ps. That’s written by the storage controller directly into memory the GPU can access (without the CPU being involved) so the normal latency to load data into a GPU is effectively gone.
That speed is significantly above what drives can do on a normal consumer computer (to the best of my knowledge)
Bearing in mind here, that it’s memory+storage architecture has a theoretical top of 22 Gb/ps, and they demonstrated recently, when assets are compressed using the “Octo” (naming?) lib, they were getting throughput of 12-18 Gb/ps. That’s written by the storage controller directly into memory the GPU can access (without the CPU being involved) so the normal latency to load data into a GPU is effectively gone.
That speed is significantly above what drives can do on a normal consumer computer (to the best of my knowledge)