Ok just had a thought. Could be totally wrong here.
The chips are rated at 5gh.
So the hw errors we are receiving from 50-60 percent is the reduction of hash power.
That's why we are getting these hash speeds of 2.2gh
It's just a thought
What does everyone else think
Haven't looked into it, but I assumed the HW errors were just timing out or not returning valid data when no nonce was found. As for the chips being rated for 5GH, not under the 2.5W you can get out of a USB port (and don't forget the power for the supporting hardware besides the bitfury chip).
Interesting. So basically, these chips are running in a constant electricity-starved state. And if they run too fast for their supplied power, they're supposed to scale back to the most stable speed possible. Great in theory, but in practice, starving ICs doesn't seem to produce great or even legible results. In my opinion, there probably isn't any hope in cgminer or any other miner working around this problem. A firmware change MAY help if it addresses the garbled communications that happen when it chokes itself out of power.