Intel has announced that it will discontinue the Itanium series of CPUs in 2021. The Itanium 9700 family, codenamed Kittson, is the last generation of Itanium to ship, with quad-core and eight-core models at clocks ranging from 1.73GHz to 2.66GHz and a TDP of 170W.
Kittson was originally supposed to be a major architectural overhaul for Itanium, introducing new features and capabilities, but after the infamous HP versus Oracle lawsuit, Intel basically threw in the towel and dropped an update that included clock speed tweaks and little else.
Intel customers must order their final Itanium CPUs and associated scalable memory buffers by January 30, 2020. Intel will then ship the CPUs until July 29, 2021. HPE has promised to keep hardware in-stock through December 31, 2025, but will presumably stop selling servers sooner than that to avoid running out of inventory.
What a Long Strange Trip It’s Been
Once upon a time, Itanium was the future. The chip was designed to solve the performance bottlenecks that Intel believed would handicap x86 and prevent it from scaling in the future. Intel threw so much weight behind it, companies like HP flat-out canceled their competing RISC designs, while Compaq chucked Alpha.
Unfortunately, Itanium wasn’t capable of doing the job its creators envisioned. The original idea behind its EPIC (Explicitly Parallel Instruction Computing) was to push the complexity of instruction scheduler from the CPU to the compiler. At the time, performing out-of-order execution on the CPU consumed significant amounts of power and it wasn’t clear how well the capability would scale. Itanium took an altogether different approach to improve computing performance by essentially dedicating the entire chip to cache and execution units, with all of the difficulty of scheduling workloads offloaded to the compiler.
The problem with offloading parallelization to the compiler at runtime is that the delay incurred when loading data from CPU caches or DRAM isn’t deterministic, which means it can’t be anticipated in advance. Not being able to determine when memory accesses would occur means the compiler’s load instruction schedule will inevitably leave the CPU stalled for long periods of time. Put simply, Itanium failed in part because Intel pushed a task into software that software compilers aren’t capable of addressing all that effectively. More details on this issue are available here.
Because Itanium was such a radical departure from anything else on the market, it ran ported software very poorly unless said software was hand-optimized for IA-64 (the Itanium 64-bit architecture) from the ground up. Squeezing acceptable performance from the silicon was a challenging endeavor that ultimately limited the CPU to a few niche markets.
I genuinely don’t know if Intel could’ve transformed Itanium into a suitable consumer desktop product if they’d had no 64-bit competition. The challenges Itanium faced were formidable and it consumed a great deal of power. It’s quite difficult to imagine a laptop running an IA-64 CPU that looked anything like the chips Intel actually manufactured. One reason we never found out is that AMD’s Opteron introduced x86-64, with full backward compatibility with IA-32 software. Intel held off announcing any kind of x86 64-bit project themselves for several years, but eventually caved and acknowledged they would also build a 64-bit CPU around the x86 architecture. Once they did, Itanium’s trajectory was set.
I’ll cop to a bit of affection for Itanium, odd as it was. Like Sony’s Cell Broadband Engine, it’s an example of a radically different CPU architecture that didn’t manage to achieve its own goals, but still represents a great deal of work by some incredibly intelligent people. Sometimes the chips that don’t work out as intended are the most interesting ones to read about.
Itanium’s Last Hurrah: Intel Releases the 9700 SeriesShadows of Itanium: Russian firm debuts VLIW Elbrus 4 CPU with onboard x86 emulationThe Worst CPUs Ever Made