Ingeneer (a@b.c) on 1/25/12 wrote:
---------------------------
>Foo_ (foo@nomail.com) on 1/25/12 wrote:
>---------------------------
>>Atom has had x86-64 support for years (when not fused off), it's nothing new.
>
>What makes you certain the 64-bit and 32-bit parts are identical designs?

It's well known that from the very first, 64-bit capability was a fuse option for Atoms. If you search hard enough I'm sure you can find direct statements from Intel employees to that effect. You can also look up die size data and make inferences:

http://ark.intel.com/products/codename/32202/Diamondville

The N270 (32-bit) and 230 (64-bit) have exactly the same die size, 26mm^2. The odds of these being different designs rather than fuse options are small; the only important reason to make 32-bit a different design would be to reduce die size to cut costs.

The thing is, 32-bit doesn't save enough area to justify that. It's been about 20 years since MIPS designed the 64-bit R4x00 series core, and (if I recall correctly) disclosed that the overhead of 64-bit was about 5% die area. It was cheap enough that the Nintendo 64 used a 64-bit R4300 even though it had no real need for a 64-bit CPU.

It wasn't a big deal back then, when the CPU itself was basically the whole chip (that was when integrating the FPU, MMU, and L1 cache was the state of the art). Today, 64-bit is a trivial amount of overhead, what with uncore (L2/L3 cache and SoC peripherals) taking up increasing fractions of the total die area in any given design.