Mali T860 (MP2?)

- 2 arithmetic* bundle/cycle
- 1 ldst* bundle/cycle
- 1 tex*,** bundle/cycle

Mali G76 (MP8?)

- 24 arith tuples/cycle
- 2 texture*,** messages/cycle
- 8 x 32-bit varying channels* interpolated/cycle
- 1 load/store* message/cycle

Mali G78 (MP??)

- 64 FMA instructions/cycle
- 64 CVT instructions/cycle
- 16 SFU instructions/cycle
- 8 x 32-bit varying channels interpolated/cycle
- 4 texture*,** instructions/cycle 
- 1 global or indirect ldst* instructions/cycle

* may be slower due to cache stalls
** in ideal conditions, in practice may be slower depending on sampler state