• Please review our updated Terms and Rules here

Just curious--anyone use any of the IA-k64/Itanium gear?

Chuck(G)

Chuck(G)

25k Member
Joined
Jan 11, 2007
Messages
44,555
Location
Pacific Northwest, USA
All this debate over whether or not to include P4 in these discussions has me wondering if anyone collects the Itanium gear? After all, it was introduced over 30 years ago and is no longer produced.

Call it idle curiosity. :)
 
There's a few people in the Silicon Graphics groups that have some of the Itanium offerings like the Prism and I'm sure I've seen one of HP's Itanium offerings in desktop form at least once (probably a ZX series), but that's it.

I think 30 years on much like it was when new we all know better to touch that stinker with a 20 foot pole.
 
What is "IA-k64" (vs the normal "IA-64")? A search for that literal term comes up with nothing.

For a while I had an HP Integrity rx2660 Server with an AH236AX AH236-2100A 1.6 GHz / 6 MB single-core CPU.

I think I got as far as installing the IA-64 version of OpenVMS 8.4 on it before quickly loosing interest and deciding to get rid of it 3 years ago.
 
NeXT said:
I think 30 years on much like it was when new we all know better to touch that stinker with a 20 foot pole
Click to expand...
I understand that the thing lasted as long as it did mostly because of HP, who had invested quite a bit in their server business. I suspect that Intel may have dropped it much earlier otherwise; witness Quark.
 
stepleton said:
I'd love to come across a machine for the preposterous challenge of trying to do some assembly coding for the architecture. I enjoyed reading Raymond Chen's articles on the topic.
Click to expand...

As Windows Server 2008 R2 was apparently the last version of Windows Server to support the Intel Itanium architecture, it must have been over 15 years ago that I had a few occasions to look at IA-64 code in a debugger when something bug checked or did something else wrong. Fortunately I didn't have to do it often enough that I really needed to understand the architecture as long as the debugger symbols matched everything up to the source code and data structures.

There were extremely few people around at the time that had any need to understand the IA-64 architecture. Raymond is one of those guys that could understand anything software related that he was interested in or had a need to understand.
 
NeXT said:
I think 30 years on much like it was when new we all know better to touch that stinker with a 20 foot pole.
Click to expand...

I mean, in all fairness "Merced" sucked pretty hard, but there was this window, I dunno, maybe 2003 to 2007?, where Itanium actually could make a pretty good account of itself. Being a VLIW architecture how well it performed did depend a lot on how good the compiler was(*), but when it worked it broke the meter with some crazy high instructions-per-clock rankings. Had a friend that worked at SGI when they were building the Altix and he swore by it, and Itanium supercomputers were in the TOP500 top-ten for a good three years running. (Peaking at number 2.)

(*I don't know if Intel intentionally set out to make a CPU that in retrospect would remind people of the i860, and it's maybe a little unfair to compare Itanium to that, but...)

From a future collector's standpoint I think the main problem with Itanium is the number of "personal" workstations made using them is miniscule; collecting rack servers is a pretty niche hobby. Also, honestly, high powered/high density systems like that just don't age well. I suspect any survivors are already prone to just nuking their power regulators and spontaneously cracking their RoS solder joints at the slightest provocation.
 
Last edited:
Chuck(G) said:
After all, it was introduced over 30 years ago and is no longer produced.
Click to expand...

... Wait, 30 years ago? I mean, I guess so, if you count from when HP and Intel put their heads together (with a sound like a bowling alley?) to develop the "P7" based on HP's skunkwork VLIW tech in 1994, and I guess they were showing off Merced prototypes as early as 1998, but a mortal human couldn't actually *buy* one until 2001.

(I did witness a working Itanium prototype at a LinuxWorld expo in... August 2000? I know it's a terribly sexist joke to make, but, yeah, it was a lot less interesting than the girl in the demon costume that was hanging out at the FreeBSD booth.)
 
gslick said:
For a while I had an HP Integrity rx2660 Server with an AH236AX AH236-2100A 1.6 GHz / 6 MB single-core CPU.
Click to expand...
Me too! Well mine was some form of dual CPU IA 2600 that I bought on eBay after they carved up some supercomputer it was part of. Unfortunately mine was lost about five years ago in my big move. I had it set up to triple-boot at the time to Windows Server 2008, RHEL 4 or 5 (I would have to look up which) and of course HP-UX.

I guess that I don't collect IA64 anymore since its gone (and I only had one any way) :(, but sort of, since I do still keep backups of all the software and documentation that I used to put it together.
 
Last edited:
Eudimorphodon said:
... Wait, 30 years ago? I mean, I guess so, if you count from when HP and Intel put their heads together (with a sound like a bowling alley?) to develop the "P7" based on HP's skunkwork VLIW tech in 1994,
Click to expand...
Yeah, that's when I first learned of the effort, so I'm counting backward from there. Reportedly, Intel shipped about 4G units, which isn't terribly shabby.

Strictly speaking, the Itanium wasn't VLIW, but EPIC. I know, that's splitting hairs, but it is what it is.

I confess that I was enamored of the i860; still have the Intel book. I thought it had a lot of potential.

There's an HP ZX6000 workstation on eBay. Stupid high asking price, considering its capabilities. Still, that might be reasonable if you're replacing an existing unit requiring the IA64 architecture.
 
Last edited:
On the 1969 CDC STAR, there were 256 64 bit general registers as well as various status registers to keep track of in context switches. Memory was core, so the swap was achieved via the normal read-rewrite cycle. Bus width was 512 bits. Not as bad as it would appear. I wonder if the IA64 didn't simply have extra register sets and switched among them.
 
I wouldn't say I collect them, but I have an HP zx2000 "workstation" and the IBM version of the Itanium Development Vehicle ("Intellistation Z Pro"). I don't actually like them much, even after you disregard my feelings about how the aggressive way they were marketed impacted the industry at large.

In my limited/silly benchmarking, the 800 MHz IDV gives about the same results as a 200 MHz Pentium Pro, in a machine from about 4 years earlier.


The 900 MHz zx2000 (Itanium2) does a little better, but still embarrassing compared to, say, the 375 MHz PPC604e3, again in a machine from about 4 years earlier.

 
bear said:
I don't actually like them much, even after you disregard my feelings about how the aggressive way they were marketed impacted the industry at large.
Click to expand...
That seems to be a repeating opinion. The are undeniably fast but if you knew them, you hated them for one given reason or another.
 
Chuck(G) said:
What I've been given to understand is that Itanium can be spectacular on some types of tasks and dreadful at others.
Click to expand...
My cranky opinion is that it was fintech bozos with monstrously huge monte carlo simulations coded in excel who ended up being the actual primary market for them.

It's at least a little true. My bad attitude makes me jump to the conclusion it's all the way true.
 
bear said:
My cranky opinion is that it was fintech bozos with monstrously huge monte carlo simulations coded in excel who ended up being the actual primary market for them.
Click to expand...

FWIW, Itanium supercomputers like NASA's Columbia were extremely good at weather simulation and physics modeling.

bear said:
In my limited/silly benchmarking, the 800 MHz IDV gives about the same results as a 200 MHz Pentium Pro, in a machine from about 4 years earlier.
Click to expand...

The only benchmark that ever mattered for Itanium was LINPACK, preferably a binary tuned with *just* the right compiler to maximize parallelism. Integer benchmarks are not remotely its thing.
 
Yup, it was LINPACK, back in the day that sold supercomputers. You tweaked the (Fortran) compiler to a fare-thee-well to get good numbers. Shades of Saxpy...
 
You must log in or register to reply here.
Back
Top