1) Tested the server market. A 32-bit chip has its limitations, but for things like caches that are IO limited it is fine, and
2) Got their 64-bit architecture nailed down. I expect they don't want to give Intel a long lead time to respond.
1) Tested the server market. A 32-bit chip has its limitations, but for things like caches that are IO limited it is fine, and
2) Got their 64-bit architecture nailed down. I expect they don't want to give Intel a long lead time to respond.
Comparing to MIPS, they have had a 64-bit ISA variant for a very long time, yet all the recent cores have been 32-bit as this is all their current target markets require.
A full 64-bit system is really required only for HPC-type applications, a domain currently owned by Intel and IBM. If ARM are interested in eating into this market segment, it is most likely a very long-term goal, and obtaining a foothold in the 32-bit server space would serve as a first step in that direction.
Databases are another place where 64-bit systems shine. Maybe you already assumed that was included under "HPC-type applications", though. When I hear HPC, I think number-crunching, not databases, though.
I'd also guess that limiting the addressing to 40 bits is partly for power consumption, as mentioned elsewhere.
You're probably right about not wanting to start an open fight with Intel, but it's hard to not see this as a shot across the bow, especially compared to Atom. If nothing else, it should make the competition between Atom and Bobcat more interesting :)