I think this dual issue thing is being misunderstood. And in truth we have AMD and Nvidia to blame for that for deliberately obfuscating the information.
The easiest way to look at dual issue when used with a proper compiler (the compiler is key), is that it makes the GPU so efficient, that it ts as if it's running twice as fast.
How it does this is important to understand. In your typical GPU, your IOPS and flops for says 32bit float are the same. Double the float for a 16 bit workload as you can fit two of those in a 32-bit waveform, and half the floats for a 64-bit task as you need two clock cycles to complete the task.
A dual issue compute task (and this is because we do not have double data address stores in the GPU, unless this has been changed), allows the PU be "faster" by simply doing multiple operations per clock. Note this is not the same thing as twice as many operations per clock. A good way to look at this is something like this...
Typical GPU
1st Cycle - Arithmetic compute (add/multiply/fused add+multiply)....etc
2nd cycle - memory access operations/Reconstruction pass
VOPD GPU
1st Cycle - Arithmetic compute + memory access operations/PSSR pass
2nd Cycle - Rinse and repeat....
Firstly, not all tasks benefit from dual issue due to conflicts... eg, you can't have the GPU doing two arithmetic loads when you have only one place to store the results per pass. ANd secondly, as you can see from above, the dual issue is extremely compiler dependent. Upi have to specifically write for the target shaders. As such, a developer building a game from the ground up to take advantage of say dual issue on the PS5pro... won't even be able to get that game running right on the base PS5. Much less on a PC.