debian hardinfo benchmarks

[minator]

I'm still surprised at the fairly low results.  Though it's interesting how close it is to the higher clocked G5 running the same benchmark.

There could be a number of reason for this:

System is set up wrong, but in OS 4 and Linux seems unlikely (unless the same people did both).


The low power has involved compromises that impact its performance - This is quite likely, you don't get low power for free.

The PA6T is "fragile" performance wise - A distinct possibility, this is not exactly rare even at the high end. The G5 was behind the x86s at general purpose stuff but ahead on heavy maths.  Could be something similar but more pronounced - its weak on general purpose but the FFT score is very good.

One of the modes is faster than the other - has anyone compared 32 bit to 64 bit Linux to see if it makes a difference?

Software is compiled for the wrong target - The PA6T was in part designed to have similar characteristics to the G5, to the point that one of the G5's weak points was copied (2 cycle integer instructions).  Code compiled for the G5 should run better than code for the G4.

[minator]

IIRC the key goal for the PA6T was low power (Watts), not high performance. AFAIK they targeted it for Apple to use in laptops post G4, not to compete with the G5 desktops.

Actually, it was designed to provide G5 type performance at much lower power - and if you look at the benchmarks of the X1000 and the G5 2.3GHz is is remarkably close on some of the benchmarks.