Wilco (Wilco.Dijkstra@ntlworld.com) on 9/18/09 wrote:
---------------------------
>Cortex-A8 and Atom are similar 2-way in-order cores to compare with. You're not
>going to get a better comparison than this. Cortex-A8 is pretty much the same as
>Cortex-A9, with the key difference that it is in-order. In fact they are so similar
>that you could argue that ARM should have skipped the A8 and immediately go OoO.
>

I don't agree that Cortex-A9 is equalt of Silverhorne when it comes to execution resuorces:
1. Silverthorne is sort-of 2.5-way core in 2-way closes. With right input stream it could sustain load+two ALU operations.
2. Cortex-A9 claims "double the performance of previous ARM FPU", which probably means peak scalar double-precision throughput between 1/3 to 1/2 of Silverthorne. Of course, on real-world compiled codes Cortex-A9 would be MUCH closer to his peak DFLOPs peak than Silverthorne would be to his, but that's another story.
3. Another issue is L2 cache. It seems, Silverthorne's is much higher performance than Cortex-A9, esp. when it comes to latency. For bigger applications that could be a single most significant difference between the two, giving to inferior in most other aspects Silverthorne core a significant edge.

As to Cortex-A8, wasn't it just rushed out to the market half-baked? I can't believe that it is the based possible general-purpose-performance-oriented 2-way in-order ARM-Cortex+Neon design.