| MikeB said: @ Entroper Of course, but that counts for both CPUs. Each game will have different requirements (resolution, effects, things going on simultaneously, optimizations, etc) it's difficult to use games as a benchmark other than statements from developers regarding that specific game (Resistance less than 10%, Uncharted ~30%). In any case it just shows it is achievable (system bandwidth and general design in the PS3 is efficient enough to get the most out of the Cell). |
I see what you're saying. But in absence of similar tests done on the Xenon, it's not very conclusive as far as drawing a comparison. It's no surprise that MS hasn't published a similar test; the Xenon is not marketed for high performance parallel computing applications.
The Xenon cores do have "their own memory" to work with, it's just called "L2 cache" rather than "local store", and its use is transparent rather than explicit. I realize it has less bandwidth and that the GPU can lock it, but it's not as if all three cores and the GPU will be in contention for the bus on each read. You can optimize for cache utilization on a symmetric multi-core system just like you can on an asymmetric one; the difference is that you aren't explicitly starting DMA transfers to the cache in the symmetric system. Project 2 in that same class I told you about earlier was optimizing the matrix multiply for a consumer-level CPU, using exactly this technique.
