CKeichel said:
Your whole argumentation is based on the assumption, that the bottleneck in Wiz/Caanoo emulators is the CL, in many cases it may be the speed of the CPU and in other cases it may be the ammount of memory, e.g. not all NeoGeo games are Fighters, games like Metal Slug or Strikers Plus are constantly accessing different parts of the Rom data, I am sure, GnGeo is accessing the SD card more often with fewer Ram.
(And apart from that, I benchmarked my systems quite regulary for a decade now, I think my first PC with DDR-SDRam was around 2003, no prefetching, no L3 Cache, no hyperthreading, etc. on an Athlon 1000).
Fact remains, unless you can show benchmarks that proof such a difference in speed between Wiz and Caanoo, I doubt your numbers.
No, my "whole argument" is based on someone else running benchmarks. YOUR argument is based on assumptions, one of them being that I'm making this up, I guess. Look, you can think what you want, but your Athlon 1000 has very very different memory characteristics from the ARM9 in Wiz. I listed a lot of things, but L2 cache is far more critical than L3 cache, and OoOE is more critical than prefetching. I didn't respond to you the way I did in order to justify the 10% claim I made, because for that you either think I'm lying or you don't. I responded that way because you don't even think it's possible.
It's not a correct mindset to look at this as having "bottlenecks" because for the most part, the latency portion of memory accesses and CPU instructions don't run in parallel. Of course, the proportions can be different, changing the amount that it's sensitive to, and it depends on the access patterns. I'm just saying that it was true for several emulators, I have no doubt you can make a program that doesn't care at all what the memory latency is, you can probably also make a program that runs at the same speed even if you disable L1 dcache.
You already mentioned NeoGeo not being able to load entire ROMs, I already responded to it, I'm not going to again. As far as I'm concerned these are corner cases at best. Since you're so eager to make the point maybe you should benchmark THAT for the two platforms.
And fine, since you're being so difficult about this I'll ask the person in question if I can post his results.