@DrBombcrater
IBM tells 750GX is up to 18% faster than 750CX with the same clock (in some memory intensive test I would assume). Obviously it depends much on the application. If the specific code snippet ran out of L2 cache before, it will be much faster than with 750CX. 750GX also implements some in-between cache buffers and other tricks, which all help, to a point.
However, 750GX is still pretty much the basic 750 design. It doesn't really go much faster when the code is executing inside the cache, except of course due to the higher clockrate.
Further, to get most out of 7447 one needs to optimize the code for the deeper pipeline in mind to make sure all execution units stay busy, and that there are minimal number of pipeline flushes. 750CX and 750GX handle unoptimized "generic" ppc code very efficiently, due to small depth of the pipeline.
I believe this is the source of the misconception that 750GX would be as fast or much faster than 7447. In some specific test cases, 750GX with high clock might get good and even better results than 7447 with smaller L2 cache.
However, in pure number crunching, or when designing the code to fit the available L2, taking advantage of the deeper pipeline (avoiding pipeline flushes specifically) not to mention the 4 AltiVec units, I believe 7447 eats 750GX's alive.
Personally I'd pick the 7447 over 750GX any day.
