Arstechnica Write Up

You're Xenon number is a wee bit off... Should be 96GFlops (pathological case)... A 3.8GHz Prescott should 15.2GFlops, and a 2.7GHz 970FX would be 32.4GFlops (pathological cases).

 

To bad that it does not have 2 VMX units it would really boost maths engines, but they would need also more cache memory, ergo a big cost per chip?

 

What's with the FLOPS arguments. I mean in the end who cares what the ratings are. I wanna play games, not FLOPS.

 

Shoo then.

 

Me too, but Flops (and such) could give new kinds of games.

I am completely satisfeid with the GPU probably anything more would just add new aesthetics elements, not new gameplay elements, but CPU power probably could add new gameplay elements.

Peace.

 

Do you have a source for this? I've only seen data on the PPEs, which mentioned 21-stage pipeline and a mispredict penalty of 8 cycles. I have yet to see similar data regarding the SPEs.

Personally, I'd rather see more scalar pipes primarily because that's what is used more often. The only thing that really sees a huge boost from vector code is just graphics. If you want to have the CPU help along vertex processing and minimize the GPU load, then okay, more SIMD units will help. But as far as getting the game logic moving, passing messages, jumping around in memory, AI, physics, animation, it's almost all scalar.

 

According to the Cell MPR pdf, the mispredict penalty on the SPE is 18 cycles.

 

Thanks for the info, I thounght that VMX units could boost most of math applications (like physics).