40 TOPS (Dense, INT8) of AI acceleration through an M.2 slot.
The MLA10 M.2 brings 40 TOPS of dense INT8 inference to systems with an M.2 slot. A Hat Board carries the NPU and memory. It connects by flex cable to an FPCB that plugs into the host’s M.2 slot and draws power from it.
AI performance
40 TOPS (Dense, INT8)
NPU cores
4
Memory
LPDDR4x, 4266 MT/s
Memory capacity
8 GB, 16 GB or 4 GB
Host interface
PCIe Gen4 x4
Power
~10 W, from the M.2 slot (DC +3.3 V)
Reliability
ECC on all cache and memory units
Hat Board
85.6 × 61.5 mm
FPCB
22 mm (L) × 80 mm or 42 mm (W)
