• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Itanium: Intel’s Great Successor

erek

erek

Fully [H]
2FA
Joined
Dec 19, 2005
Messages
17,421

View: https://www.youtube.com/watch?v=-K-IfiDmp_w

This video explores the history and ultimate commercial failure of Itanium, a joint 64-bit microprocessor project between Intel and Hewlett-Packard (HP) announced in June 1994 (0:02-0:31). Initially intended to unify computing architectures, the project represented one of the most ambitious and expensive endeavors in Intel's history.

Key phases and milestones:

  • Motivation: Intel sought to enter the high-end workstation and server market dominated by RISC-based competitors (1:49-2:40). They believed a "clean sheet" architecture using Explicitly Parallel Instruction Computing (EPIC)—a philosophy derived from Very Large Instruction Word (VLIW) concepts—would outperform standard x86 and out-of-order superscalar designs (7:14-12:03).
  • Development Challenges: The project faced immense complexity, multiple delays, and intense internal friction between Intel teams in Santa Clara (focusing on Itanium) and Oregon (focusing on the x86 Pentium Pro / Xeon lines) (22:42-23:16, 25:22-28:06).
  • Market Realities: While Itanium struggled with delays (28:39-31:05), the x86 architecture evolved unexpectedly through the Pentium Pro and Xeon, and AMD countered with their own 64-bit extension (AMD64), which became the industry standard due to its backward compatibility (38:18-38:44, 43:45-44:09).
  • Legacy: By the time Itanium launched in 2001, the shift toward cheap, commodity-based "compute clusters" using x86 servers made proprietary, expensive architectures less relevant for most market segments (41:45-42:52). Despite HP's continued support and legal efforts to keep the ecosystem alive, Intel officially ended the line with the 9700 series, with the last chips shipping in 2021 (45:53-47:23).
While Itanium ultimately failed to displace x86, the video notes the audacity of the project and how its legacy eventually solidified the market dominance of the x86 architecture for the following decades (47:26-47:49).
 
Guys voice is kind of annoying, but I am going to suffer the voice and watch the video.

winXP-itanium.jpg
 
Could be one of those, ahead of it's time scenarios. Would be pretty interesting if they incorporated that architecture now.
 
  • Like
Reactions: erek
like this
El_Capitan said:
Could be one of those, ahead of it's time scenarios. Would be pretty interesting if they incorporated that architecture now.
Click to expand...

Eh, it was at least the second time Intel tried to remake processing. Turns out, despite the hype, x86 is pretty decent. VLIW makes some computer architecture guys wet, and it can work for DSP, but out of order execution and a generous (but not too generous) pipeline seems to work well for general purpose cpus. The problem is when you have memory accesses that may or may not hit cache and variable cycle counts to service cache misses, the compiler can't schedule instructions anywhere close to what the cpu can do in realtime. If we had low MHz cpus where there's no cache because ram is small and access is always one or two clocks, then maybe
 
I worked at Match.com when we deployed Itaniums for our SQL clusters. We were assured by Microsoft our existing SQL DB's would run just fine on Itanium. Our DBAs told us after a few days of trying told us they would never ever try to use them again. We dragged HP back into the DC to get those things out.
 
The trouble with Itanium was that is expected a magic compiler to do SIMD all the time from plain code. Such compilers do not even exist today.

With a normal compiler, which is all it ever got, it wasn't competitive speed wise.
 
IMHO the idea Intel had when they agreed on this deal with HP was that in rare case it pans out they will have fastest CPU and if it doesn't then maybe they will manage to convince some RISC vendors to stop working on their CPU designs and with it eventually shift them to dark side of the force (read x86).

And what happened is that few RISC CPU vendors did bet on Itanium and then did shift to Xeons.
Also Intel in the end didn't loose money on Itanium - it was used by HP and in some very specific use cases e.g. lockstep is easier on Itanium and it being so different has tighter security... well, maybe its more like security by obscurity with some of the speculative execution exploits maybe not existing but other exploits existing but it doesn't make difference if no one has EPIC exploits because no one has it to begin with. On that note EPIC being simple in its design might lack similar exploits altogether.

Anyways, I remember Intel pushing hard the agenda that ia64 will replace ia32 and that eventually we will all be using ia64 CPUs. Just not immediately and it will take some time.
I had like 256-512MB RAM when Intel was spinning this ia64 fairy tale of theirs. Then Itanium happened... and then amd64 happened and any chances of ia64 becoming mainstream vanished. Especially since AMD pushed Athlons 64 immediately and pushed them hard. Then Intel somehow made Prescott support 64-bit (likely via clever microcode emulation) and we all knew what it meant.

What surprises me is for how long Itanium hanged on.
Would be very interesting to get my hands on latter Itanium CPUs to check how they perform.
 
toast0 said:
Eh, it was at least the second time Intel tried to remake processing. Turns out, despite the hype, x86 is pretty decent.
Click to expand...
Over all it seems like what we've seen with CPUs is that ISA isn't as big a deal as was hyped back in the 80s and 90s. While it isn't irrelevant, it seems like for the most part you can have good performance with any reasonable ISA and all the Erhmehgedciscisdead crap was just that: crap. Even if EPIC had the possibility of being faster, could it be faster enough to make it worth losing binary compatibility?

Seems to keep being the case too. It isn't like anyone has invented a new CPU with a magic ISA that makes it ever so much faster than others.
 
OFaceSIG said:
I worked at Match.com when we deployed Itaniums for our SQL clusters. We were assured by Microsoft our existing SQL DB's would run just fine on Itanium. Our DBAs told us after a few days of trying told us they would never ever try to use them again. We dragged HP back into the DC to get those things out.
Click to expand...
I worked at a large financial company back in the day and they spent a fortune on some Itanium clusters for the highest profile/priority MS SQL stuff. Almost exactly the same thing happened... the DBA's tried to make it work better than it was already doing on the old pentium clusters and failed.... when it even ran at all. SOOOOO many weekend/overnight maintenance windows to patch/reinstall tinker tinker tinker. Everyone hated those things. So the DBAs gave up and kept using the old clusters, someone conjured up some kind of a dummy/fake/demo DB and that is all that the Itaniums ever ran.... just so mgmt could point at something running on the super expensive dumpster fire during executive tours.
 
Burticus said:
I worked at a large financial company back in the day and they spent a fortune on some Itanium clusters for the highest profile/priority MS SQL stuff. Almost exactly the same thing happened... the DBA's tried to make it work better than it was already doing on the old pentium clusters and failed.... when it even ran at all. SOOOOO many weekend/overnight maintenance windows to patch/reinstall tinker tinker tinker. Everyone hated those things. So the DBAs gave up and kept using the old clusters, someone conjured up some kind of a dummy/fake/demo DB and that is all that the Itaniums ever ran.... just so mgmt could point at something running on the super expensive dumpster fire during executive tours.
Click to expand...
We had some at work that a researcher used and they did get good results out of them. Of course that was all custom code, developed for the Itainums. That said, I don't know that they were worth it in any economic sense, it was a federal government project and they chose the hardware. It did start before amd64 though so there weren't as many 64-bit options that weren't completely stupid expensive. Not that Itanium was cheap but compared to IBM POWER...
 
You must log in or register to reply here.
Back
Top