Quote:
CPU Game Math Performance
- 9 billion dots per second
Huh? 9 billion "dots"? WTH is that supposed to mean?
EDIT : If that's supposed to be dot products, 9 billion wouldn't quite agree with the clock speed numbers. If the cores were running at 3 GHz (as opposed to 3.2), a retire rate of one dot product per cycle on 3 cores would yield 9 billion. At 3.2, it would be 9.6 billion.
Besides which, since when are dot products used as a unit of performance measurement.
Quote:
Memory Bandwidth
- 22.4 GB/s memory interface bus bandwidth
- 256 GB/s memory bandwidth to EDRAM
- 21.6 GB/s frontside bus
How does one have a 21.6 GB/sec FSB if it's sharing the same 700 MHz RAM with the GPU. 22.4 makes sense (128 bit bus * 700 MHz ram * 2 (DDR)). 21.6 doesn't make sense at all no matter how I work it out (serial protocols, ECC on same lines, whatever) -- moreover, the overall stupidity of having the CPU hog all the memory bandwidth is pretty severe. 12.6 works, but only if you include ECC codes as part of the stream, which you really shouldn't. The effective bandwidth would still only be 11.2 GB/sec. Quote:
Custom ATI Graphics Processor
- 500 MNz
- 10 MB embedded DRAM
- 48-way parallel floating-point shader pipelines
- unified shader architecture
I love these abbreviations... "Meganertz"... Unified shader architecture is about the only thing there that surprises me. Frankly, that's the one thing that I still wouldn't believe until I see it.