THERE HAVE LONG been rumors of a major player moving their data centers from x86 based PCs to the ARM architecture. It looks like the first big player to jump in to the market is going to be none other than Facebook.
If the rumors are true, Facebook will be putting ARM based servers in their upcoming Oregon data center, dumping x86 in a nearby river. In an environmentally friendly fashion, it is Oregon after all. In any case, there won’t be any x86 parts needed at the privacy specialists.
What ARM chip are they going to be using? This is where the game of connect the dots begins, and all the rumors point to Smooth-Stone, an Austin, Tx chip design startup. A few more rumors link the boards to SuperMicro, but that link is a bit more tenuous. In any case, it won’t be long before all the details are revealed.
It looks like ARM has grown up enough in raw CPU power to hit the proverbial ‘fast enough’ point for a single Facebook thread. Now it becomes a question of wattage vs threads. An Intel/AMD x86 CPU runs more threads but consumes more watts. The wattage per thread appears to have come down on the side of ARM, at least for Facebooks PHP-ish code.
The thought of this will scare Intel silly, not only did they lose a contract potentially for tens of thousands of Xeons a month because of power, but they didn’t win it with Atom. In fact, they lost it to the one rival that they don’t have a direct answer to, the ARM ISA.
If Facebook’s experiment pans out, it may change how things are done in the data center, and collapse prices. Actually, since Facebook is moving to ARM, it isn’t an experiment, so you can substitute all of the forward looking statements with more definitive ones. Things have changed in the data center, you just haven’t seen the results yet.S|A
Latest posts by Charlie Demerjian (see all)
- Intel shows off 10nm 112Gbps SerDes - Mar 12, 2019
- Intel releases Compute Express Link spec - Mar 11, 2019
- Qualcomm rolls out a second gen 5G modem called X55 - Feb 19, 2019
- What is Intel’s Foveros tech and what isn’t it? - Feb 11, 2019
- Why SemiAccurate called 10nm wrong - Jan 25, 2019