AMD 7970M crippled/limited double precision?

Something seems wrong with the double precision speeds of new 28nm mobile cards released by AMD:

7970M:

2176 GFLOPS Single Precision compute power
136 GFLOPS Double Precision compute power

7870M:

1024 GFLOPS Single Precision compute power
64 GFLOPS Double Precision compute power

7770M:

691 GFLOPS Single Precision compute power
43.2 GFLOPS Double Precision compute power


Now look at the desktop cards:

7970:

3.79 TFLOPS Single Precision compute power
947 GFLOPS Double Precision compute power

7950:

2.87 TFLOPS Single Precision compute power
717 GFLOPs Double Precision compute power


As you can clearly see, the ratio between single to double precision on the desktop models is 1:4 but is 1:16 on the notebook models.

Seems they're working at only 25% their full potential. I'm disappointed. Anyone know why they're doing this?

I'm sure if double precision were to be simulated using two floats, I've seen some papers around regarding this ( http://stackoverflow.com/questions/6769881/emulate-double-using-2-floats), it should be able to perform better than 1:16.

 

7970M is based on Pitcairn (desktop 7870) not Tahiti (desktop 7970). Pitcairn at 1GHz has 2560 GFLOPS single precision and 160 GFLOPS double precision. The 7970M is in line with what you'd expect from a lower clock Pitcairn part. The miserly 1/16th rate is likely a side effect of Pitcairn's extremely high performance per watt (compared to Tahiti).

 

Yeah, AMD pretty much did what Nvidia did with gk104 and reduced the transistor count and die size by sacrificing DP perf.