What is going onThe AIDA64 read speed (important for AI / LLM self-hosting) of
Quote63260 MB/s
is too low. In the same table the Yoga Pro 7 14 G9 reaches
Quote99376 MB/s
. The G9 should actually be slower, as its memory bandwidth is lower: LPDDR5x-7500 (vs 8000), dual channel.
The theoretical RAM memory bandwidth of LPDDR5X-8000 is:
8000 * 128-bit (2 * 64-bit per channel) / 8 / 1000 = 128 GB/s.
The practical is something like 70% to 80% of it:
128 GB/s * 0.7 = 89.6 GB/s, but only 63.26 GB/s is measured.
Can anyone explain this?