It WILL run better in 32-bit per-clock, there's no denying that, it has the memory controller and we have seen what it can do per clock. Some will go unaffected or be slightly better but overall IT HAS to be better per clock.
It 'HAS' to be because you want it to, or it 'HAS' to be because of hardware, or what?
Theoretically, if we can believe AMD that the ondie memory controller helps out soooo much, it <i>might</i> perform faster than an AXP that's decked-out.
Realisticly, just how much would the slice off of the memory latency of a <i>single channel</i> memory controller help? The A64 has a slightly longer pipeline. AMD's CPUs are not exactly known for stunning prediction and prefetching (which compounds the added pipeline length because you'll have to fetch the hard way more often). The nForce2 can already run XMS, LL, HyperX, etc. at rather low latencies already and further trims memory latencies in dual-channel over anyone else's single-channel so far.
Will an AXP with an nForce2 and LL RAM <i>really</i> be slower than an A64 in pure 32-bit? Or will it run at exactly the same speed if not a teeny-tiny bit faster?
Unless AMD pulls a really nifty rabbit out of their hat (like better prediction algorithms or an ability to process more ops per cycle) for the release of the A64, I just don't see it beating an AXP at 32-bit. Short of something up a sleave, SSE2 so far is A64's only possible hope to impress. But depending on how AMD implemented SSE2 it might not actually boost performance much if any.
We'll have to wait and see, but I'm not expecting much out of AMD.
"<i>Yeah, if you treat them like equals, it'll only encourage them to think they <b>ARE</b> your equals.</i>" - Thief from <A HREF="http://www.nuklearpower.com/daily.php?date=030603" target="_new">8-Bit Theater</A>