6 ms·

I'm still puzzled why Itanium didn't go any further. A precursor of the ARM revolution. And Intel just wrongly killed the project.

tux3 5y ago
Well, have you seen the famous Itanium Sales Forecast graph? https://en.wikipedia.org/wiki/Itanic#Itanium_(Merced):_2001 https://en.wikipedia.org/wiki/Itanic#Itanium_(Merced):_2001
tyingq 5y ago
It depended on compilers smart enough to make not-parallel-friendly code run well in parallel. In a way that would outpace the price/performance growth curve of the (somewhat new at the time) x86-64 CPUs. The compilers never fully delivered on that. X86-64 just beat them with process size and clock speed on a less elegant architecture.
- marcosdumay 5y ago
  Well, I'd say that whether VLIW is more elegant than OoO is a matter of subjective opinion.
  - guenthert 5y ago
    Elegance might be in the eye of the beholder, security is not.
blacktriangle 5y ago
I worked on a project mid-aughts to port a scientific package to Itanium. The issue that we observed was that Itanium really pushed the hard work out to the compiler and by association the devs to make everything work with their compilers in order to get the full benefits. And since very few people had Itaniums and the compilers weren't quite there yet, the overall dev effort was not worth it. Somewhat of a chicken and egg problem. ARM managed to be worth the trouble since it solved a real problem of power useage which got it the installed base, at which point it was worth the trouble of supporting.
chrisseaton 5y ago
> I'm still puzzled why Itanium didn't go any further. It turned out nobody knew how to build effective compilers for them, so nobody bought them. > And Intel just wrongly killed the project. What's the point of a processor you can't program efficiently? What do you think Intel should have done? Just kept making them regardless of the fact nobody bought them?
- jeffbee 5y ago
  > It turned out nobody knew how to build effective compilers for them, so nobody bought them. It's a funny sign of those times. People really thought at one time that we were leaving a huge amount of potential instruction-level parallelism on the table. There was a DEC-WRL paper in the early 90s that said in the future (i.e., now) we'd be disappointed with 15 instructions retired per clock. But in the future we presently occupy we're all glad to have IPC of 0.8.
psanford 5y ago
Ha. Intel pushed Itanium very hard. AMD released AMD64 and that was the end of Itanium. No one wanted it. Intel killed it because it was a failure in the market place. I love this graph of Intel's projected vs actual Itanium sales. They certainly believed in Itanium for a long time: https://upload.wikimedia.org/wikipedia/commons/8/88/Itanium_Sales_Forecasts_edit.png https://upload.wikimedia.org/wikipedia/commons/8/88/Itanium_...
- tyingq 5y ago
  I think those projections weren't really all that crazy. AMD surprised everyone with AMD64, and they were the only ones that could have done 64-bit x86 other than Intel. Intel should have had better spies at AMD :)
  - chasil 5y ago
    They did. There was a skunkworks project to copy it, and the copy was not entirely correct. https://www.informit.com/articles/article.aspx?p=366538 https://www.informit.com/articles/article.aspx?p=366538
    - Grazester 5y ago
      Did you post the wrong link or something? That link only talks about Intel implementation(EM64T) of AMD64
      - chasil 5y ago
        This is likely more appropriate. https://www.extremetech.com/extreme/52858-picking-up-the-chips-for-2003 https://www.extremetech.com/extreme/52858-picking-up-the-chi... Also related to the 3.06GHz P4 launch were rumors that Intel has secretly included x86-64 support in the new P4 in the form of disabled circuitry on the chip’s die. Such circuitry is allegedly the product of Intel’s skunkworks Yamhill project, which is intended to give Intel some flexibility in responding to the potential threat posed by AMD’s upcoming x86-64 “Hammer” line of processors. Intel has repeatedly denied the existence of Yamhill, insisting that the only 64-bit instruction set they plan to support is IA-64. So will we see Yamhill publicly unveiled in 2003? It’s possible, but not likely. https://www.theregister.com/2002/01/25/intels_yamhill_itanic_on_ice/ https://www.theregister.com/2002/01/25/intels_yamhill_itanic...
soco 5y ago
Thinking that Compaq killed Alpha to bet everything on Itanium makes the history even more frustrating.
hestefisk 5y ago
Answer is IA-64 was good enough and industry did not have a good incentive to rewrite working software for another architecture.
- Denvercoder9 5y ago
  IA-64 usually means Itanium. You probably mean x86-64 (also known as x64, amd64, Intel 64 or EM64T).
  - hestefisk 5y ago
    Yes you are right. I meant amd64. Thanks.
- veltas 5y ago
  IA-64 is Itanium
pjc50 5y ago
ARM pre-dates Itanium by quite some way as well as outliving it. Design started in 1983 (!) https://www.cs.umd.edu/~meesh/cmsc411/website/proj01/arm/history.html https://www.cs.umd.edu/~meesh/cmsc411/website/proj01/arm/his... and by 1987 you could buy ARM home computers. The ARM revolution has been a long time coming, but it was given a big boost by the use of the chip in phones, a market that Intel failed to break into (I have one of the rare x86-Android phones, a Zenfone 2). I can't imagine Itanium would have been any better at scaling down to fit into that market. Intel had no choice but to kill it; it was outclassed by ARM64 and had a slowly shrinking customer base.
- dangerbird2 5y ago
  I love how ARM's killer feature of low power consumption came completely by accident. [1] ACORN designed ARM as a desktop processor with good memory bandwidth, and only discovered its energy efficiency when a prototype continued running without a power supply from signal input voltages. [1] https://www.youtube.com/watch?v=1jOJl8gRPyQ https://www.youtube.com/watch?v=1jOJl8gRPyQ
wbl 5y ago
Most code cannot be statically scheduled and code that can be is amenable to the big vector units we have now.

tux3 5y ago

Well, have you seen the famous Itanium Sales Forecast graph? https://en.wikipedia.org/wiki/Itanic#Itanium_(Merced):_2001 https://en.wikipedia.org/wiki/Itanic#Itanium_(Merced):_2001

tyingq 5y ago

It depended on compilers smart enough to make not-parallel-friendly code run well in parallel. In a way that would outpace the price/performance growth curve of the (somewhat new at the time) x86-64 CPUs. The compilers never fully delivered on that. X86-64 just beat them with process size and clock speed on a less elegant architecture.

marcosdumay 5y ago
Well, I'd say that whether VLIW is more elegant than OoO is a matter of subjective opinion.
- guenthert 5y ago
  Elegance might be in the eye of the beholder, security is not.

blacktriangle 5y ago

I worked on a project mid-aughts to port a scientific package to Itanium. The issue that we observed was that Itanium really pushed the hard work out to the compiler and by association the devs to make everything work with their compilers in order to get the full benefits. And since very few people had Itaniums and the compilers weren't quite there yet, the overall dev effort was not worth it. Somewhat of a chicken and egg problem. ARM managed to be worth the trouble since it solved a real problem of power useage which got it the installed base, at which point it was worth the trouble of supporting.

chrisseaton 5y ago

> I'm still puzzled why Itanium didn't go any further. It turned out nobody knew how to build effective compilers for them, so nobody bought them. > And Intel just wrongly killed the project. What's the point of a processor you can't program efficiently? What do you think Intel should have done? Just kept making them regardless of the fact nobody bought them?

jeffbee 5y ago
> It turned out nobody knew how to build effective compilers for them, so nobody bought them. It's a funny sign of those times. People really thought at one time that we were leaving a huge amount of potential instruction-level parallelism on the table. There was a DEC-WRL paper in the early 90s that said in the future (i.e., now) we'd be disappointed with 15 instructions retired per clock. But in the future we presently occupy we're all glad to have IPC of 0.8.

psanford 5y ago

Ha. Intel pushed Itanium very hard. AMD released AMD64 and that was the end of Itanium. No one wanted it. Intel killed it because it was a failure in the market place. I love this graph of Intel's projected vs actual Itanium sales. They certainly believed in Itanium for a long time: https://upload.wikimedia.org/wikipedia/commons/8/88/Itanium_Sales_Forecasts_edit.png https://upload.wikimedia.org/wikipedia/commons/8/88/Itanium_...

tyingq 5y ago
I think those projections weren't really all that crazy. AMD surprised everyone with AMD64, and they were the only ones that could have done 64-bit x86 other than Intel. Intel should have had better spies at AMD :)
- chasil 5y ago
  They did. There was a skunkworks project to copy it, and the copy was not entirely correct. https://www.informit.com/articles/article.aspx?p=366538 https://www.informit.com/articles/article.aspx?p=366538
  - Grazester 5y ago
    Did you post the wrong link or something? That link only talks about Intel implementation(EM64T) of AMD64
    - chasil 5y ago
      This is likely more appropriate. https://www.extremetech.com/extreme/52858-picking-up-the-chips-for-2003 https://www.extremetech.com/extreme/52858-picking-up-the-chi... Also related to the 3.06GHz P4 launch were rumors that Intel has secretly included x86-64 support in the new P4 in the form of disabled circuitry on the chip’s die. Such circuitry is allegedly the product of Intel’s skunkworks Yamhill project, which is intended to give Intel some flexibility in responding to the potential threat posed by AMD’s upcoming x86-64 “Hammer” line of processors. Intel has repeatedly denied the existence of Yamhill, insisting that the only 64-bit instruction set they plan to support is IA-64. So will we see Yamhill publicly unveiled in 2003? It’s possible, but not likely. https://www.theregister.com/2002/01/25/intels_yamhill_itanic_on_ice/ https://www.theregister.com/2002/01/25/intels_yamhill_itanic...

soco 5y ago

Thinking that Compaq killed Alpha to bet everything on Itanium makes the history even more frustrating.

hestefisk 5y ago

Answer is IA-64 was good enough and industry did not have a good incentive to rewrite working software for another architecture.

Denvercoder9 5y ago
IA-64 usually means Itanium. You probably mean x86-64 (also known as x64, amd64, Intel 64 or EM64T).
- hestefisk 5y ago
  Yes you are right. I meant amd64. Thanks.
veltas 5y ago
IA-64 is Itanium

pjc50 5y ago

ARM pre-dates Itanium by quite some way as well as outliving it. Design started in 1983 (!) https://www.cs.umd.edu/~meesh/cmsc411/website/proj01/arm/history.html https://www.cs.umd.edu/~meesh/cmsc411/website/proj01/arm/his... and by 1987 you could buy ARM home computers. The ARM revolution has been a long time coming, but it was given a big boost by the use of the chip in phones, a market that Intel failed to break into (I have one of the rare x86-Android phones, a Zenfone 2). I can't imagine Itanium would have been any better at scaling down to fit into that market. Intel had no choice but to kill it; it was outclassed by ARM64 and had a slowly shrinking customer base.

dangerbird2 5y ago
I love how ARM's killer feature of low power consumption came completely by accident. [1] ACORN designed ARM as a desktop processor with good memory bandwidth, and only discovered its energy efficiency when a prototype continued running without a power supply from signal input voltages. [1] https://www.youtube.com/watch?v=1jOJl8gRPyQ https://www.youtube.com/watch?v=1jOJl8gRPyQ

wbl 5y ago

Most code cannot be statically scheduled and code that can be is amenable to the big vector units we have now.