The Intel Itanium was an ambitious project launched by Intel, also known as “Itanic,” aimed at redefining microprocessor architecture. Its history began in the 90s, when Intel introduced the concept of EPIC (Explicitly Parallel Instruction Computing). This revolutionary idea was based on out-of-order execution (OoOE) in microprocessors, which allowed complex x86 instructions to be broken down into smaller, RISC-like pieces, rearranged on the fly to execute them as quickly as possible, and then reassembled in the order intended by the software.
In 1994, it was unclear whether future x86 processors would be able to effectively exploit Instruction Level Parallelism (ILP). Intel, therefore, opted for a different approach: rather than relying on significant improvements in CPU design to execute existing code, it chose to have the compiler do this work ahead of time, designing a new instruction set with Words of Very Long Instruction Words (VLIW) to make this operation possible.
However, despite its innovative premises, Itanium has encountered numerous challenges. One of the main criticisms of the Itanium architecture was its complexity and inefficiency in handling legacy x86 code. Furthermore, the market has moved in other directions, with the consolidation and evolution of the x86 architecture, which has continued to dominate the sector.
As the years went by, support for Itanium declined significantly, as did sales.
The fate of Linux on Itanium now seems sealed, and there seems to be no hope of rescue. This conclusion, however inevitable, hides a fascinating and nuanced story behind it.
Support for Intel's IA64 architecture was recently removed from the Linux kernel, sparking a flurry of complaints and heated debate. Although an initial proposal to remove support for Intel's infamous Itanium architecture, nicknamed "Itanic," was rejected in February, just a few weeks ago, in October, this decision was approved for kernel 6.7. As expected, this sparked ire among some, and was considered by many to be a bad move. LWN published an excellent analysis on this topic, titled "The Push to Save Itanium." https://lwn.net/SubscriberLink/950466/0999d63fdf270781/
Linus Torvalds, the creator of Linux, responded in turn, saying that he was willing to resurrect support for Itanium, despite personally detesting the architecture, criticizing it for being based on flawed design premises and decisions influenced more by politics than by sound technical foundations.
His rather modest and reasonable proposal was summed up by @mewse on Lobsters: " You who still want this architecture, keep a patch set off the main branch for the architecture for a year, and we'll consider making it main again. " In other words, if the people complaining about this change taking longer suddenly find time to modernize the code they don't want removed, then it could be added back. However, this seems unlikely to happen, and many know it.
Intel's EPIC project – which stands for Explicitly Parallel Instruction Computing – was a revolutionary idea in the world of microprocessors. Out-of-order execution (OoOE) was a new concept for x86 processors at the time, allowing complex x86 instructions to be broken down into smaller, RISC-like segments, rearranged in real time to execute as fast as possible, and then reassembled the results in the order the software originally expected them. In 1994, it was unclear whether future x86 processors would be capable of effectively exploiting instruction-level parallelism (ILP) in machine codes. As a result, rather than relying on significant improvements in CPU design to run existing code, it seemed safer to have the compiler do this work up front, by designing a whole new set of instructions with Very Long Instruction Words (VLIW) to make this is possible.
The history of Itanium is that of an ambitious project which, despite its innovative premises, encountered numerous challenges and failed to keep pace with market and technology developments, leading to its gradual obsolescence and the recent decision to remove its supported by the Linux kernel.
It really seems over.
