Intel Atom 32 vs 64 bit architectures?
Intel Atom 32 vs 64 bit architectures?
The initial Intel Atom CPUs from Silverthorne only handled IA-32 instructions, as noted by TechPowerUp and Wikipedia. A few months later, the Diamondville architecture supported x86_64. Both Diamondville and Silverthorne chips share 47 million transistors, a 26mm² package, and use the same microarchitecture. Despite having identical transistor counts and sizes, one is 32-bit while another is 64-bit. Some sources claim Silverthorne is a 64-bit CPU, but since it runs on an RISC core, it might only translate IA-32 instructions. Alternatively, the chipset pairing could limit it to 32-bit operations, keeping it aligned with x86. Your confusion is understandable—this technical segmentation seems complex.
there are unusual aspects about early atoms. I don’t have all the specifics, but I remember some early devices could run 64-bit Windows while using only a 32-bit bootloader.
The Wikipedia entry provides detailed information on which CPU models support which ISA and the official explanations behind those choices. For understanding the physical distinctions between CPUs, it's challenging to pinpoint exact differences—often they stem from microcode variations or subtle manufacturing tweaks. Small construction differences can also enable features like x86-64 support in chips that lack native 64-bit registers.
The provided items excluded Silverthorne completely, which seems strange.
Companies design a chip with certain capabilities. Occasionally, some parts fail or fall short of requirements. Rather than discarding the entire unit, they isolate the faulty section and create a reduced-spec version. Missing features might include iGPU, cache, or processing units. This situation could affect a 64-bit component or occur when makers avoid devaluing quality chips by artificially restricting capabilities. For instance, some AMD processors with multiple cores are built from smaller chiplets; if just a couple of cores malfunction, selling a 7-core model at a higher price might still make sense. Over time, as production continues, the rate of failures decreases, so they may disable certain features to keep prices competitive. The same applies to Atom chips—its limitations aren’t necessarily tied to a single defect but to manufacturing realities.
It's possible they hadn't finalized the 64-bit microcode yet, so they launched it with a 32-bit version first.
It would seem like a big improvement if you only clocked half of the buses and registers for the 32-bit sections of the processor—this would mean a lot of dark silicon usage. Observe how the highest power consumption comes from 32-bit units at 3W, while the lowest power 64-bit enabled ones reach 4W. Yes, if the software could run faster in 64-bit mode, it would naturally use less overall power. There was still a significant amount of 32-bit code running at that time.
I remember a couple of random machines that could run 64-bit operating systems but required a BIOS upgrade to actually boot them. These were 64-bit chips running 32-bit BIOSes, which is unusual. I’ve often wondered why they didn’t just support 64-bit from the start if the hardware clearly allowed it.