Hacker News Reader

Comment by kstrauser | Hacker News Reader

I had to maintain an IA-64 Linux system at a previous job, and it was such an odd duck. The OS did a decent job of abstracting away most of the weirdness, but at the end of the day it was just a very slow server. The compiler breakthroughs that Intel was counting on to make it competitive never happened, and since its unusual architecture made it bad at running code not specifically tuned to it, the end result was that nothing ran great on it. I’m sure that HP had some highly optimized code that ran like greased lightning, but that never worked its way out to the general public.

I admit that I’m glad Itanium finally died. It killed a lot of other interesting architectures and gave nothing in return.

View on HN

For some periods (including during some of the Opteron era) Itanium 2 actually did okay.

https://www.realworldtech.com/forum/?threadid=27345&curposti...

It had very strong floating point performance and on spec int it was holding pace with Opterons, Athlons and P4s clocked a lot higher. Apparently a lot of the int performance was due to large fast caches (I think it had a bigger die size compared with the others) -- have a look at the 1.5MB version vs the 3MB version. But even there it wasn't doing so bad against the Sun and IBM processors which were out of order I think. So yes you're right some things did run pretty fast.

Unfortunately for Itanium, compiler techniques to make up for in-order execution had just about run out of steam at that point, while OOOE continued to scale up and improve steadily. Which is pretty much the opposite of what HP and Intel had predicted in the 90s, which is why they justified Itanium's approach.

> I’m sure that HP had some highly optimized code that ran like greased lightning,

unwarranted optimism. I recall HP Pentium Pro servers with loads of "built in" stuff and dual PCI buses: all the builtins were hung off the 2nd, chained bus; on one IRQ. The memory was severely limited, too, iirc; NUMA'd through one of 4 CPUs.

HP design seems to be all about locking performance away; it might be in that box in theory, but all their ingenuity was spent ensuring it stayed there.

> It killed a lot of other interesting architectures and gave nothing in return.

What sort of architectures (if any) got killed as a direct result of developing IA-64?

That aside, though odd and impractical, I found Itanium to be at least a technically interesting perspective (both in terms of architecture, ie VLIW, and in terms of the amount of technical work it inspired at HP and Intel, such as the Itanium C++ abi [1])

[1] https://itanium-cxx-abi.github.io/cxx-abi/abi-eh.html

>What sort of architectures (if any) got killed as a direct result of developing IA-64?

Alpha?

And HP-PA and, later, MIPS (as a mainstream computer). MIPS survives in some routers.

MIPS is very, very close to dead now.

To be fair though, I think its death is more attributable to ARM than Itanium. And RISC-V is killing off anything that remained.

MIPS the brand was bought for cheap from a bankruptcy, and resurrected by its new owner… to be an ARM partner building ARM chips.

Arguably it was the nail in the coffin for alpha

Itanium may have been the nail in the coffin, but by the time the hammer fell on that nail, Alpha was already basically dead due to bad business structure and incentives that were inherited from DEC. They had many of the same problems that Intel has been having recently (low yields, failure to keep up with performance increases) without Intel's long-standing business agreements or inertia to hold up the product line through a difficult period.

Alpha, HP-PA.

Reply on news.ycombinator.com