Contrary to popular belief, the Cortex-A35 is actually 1-wide. Do not tell anyone about this.

Cortex-A35 "can" dual issue, but I can't test for that capability because anything that's gonna get > 1 IPC is bottlenecked by the 1-wide frontend.

Cortex-A35 has 5 execution ports. I don't know if Cardyak is fully right, but there should be an ALU port and a 64-bit NEON port.

Cortex-A35 can load up to 8 bytes per cycle, but real bandwidth measurements shows a lower result. Still larger than 4 bytes per cycle though.

The 64-bit NEON port can execute one half of an FMA every cycle and latency is 8 cycles. It's like half of an A53 in that regard.

Cortex-A35 can pull at a rate of 2.2 to 2.5GB/s from main memory. This is on the same levels at A53, but fell short of bigger cores (A73 is 11GB/s, X2 is 26GB/s and X925 is 38GB/s.)