TinyML is an incredibly powerful piece of software, and you can easily train your own model and deploy it on an ESP32.
That 36 MB/s has to be sustained from PSRAM to the LCD DMA, while the CPU (and LovyanGFX, used for graphics) are also writing ...