67 and 33.5Just to people understand the Dual Issue.
It means you can issue TWO instuctions at the same time to a SP when you meet some rules (sending the same instruction to all SPs in a cluster).
That doesn't mean it will execute both instructions at same time.
It will still execute one instruction after the other.
The gain lies in saving the overhead to make two separably calls.
The gain is minimal to be fair.
And due that AMD used the marketing bullshit to say they GPU are now can do double the TFs they actually do.
And they don't.
A 30TFs RDNA3 will perform similar to 15TFs RDNA/RDNA2.