Intel finally puts the mighty Itanium out of its missery

[DuckDodgers]

 

 

Once thought as the Nemesis for both RISC and x86 "establishment", the first native 64-bit architecture from the Intel/HP alliance is heading on its way out of the market and the industry for good, completely superseded by the humble x86-64 ISA. 

 

The first nail in Itanium's coffin came in 2003, when AMD released its first 64-bit processor. AMD's hybrid approach was so successful that Intel actually adopted AMD's 64-bit extensions, as the dominant industry standard. This left Itanium to die a long, slow death.

 

Quote

Remember Itanium? Intel's first crack at 64-bit server processors from circa the turn of the millennium? Well, two things: Intel is releasing four new 9700-series Itanium CPUs based on the "Kittson" architecture, and also they're the last new Itanium processors that the company plans to ship.

 

An Intel spokesperson confirmed to PC World that this was the end of the line for Itanium, which jibes with an agreement under which HP would pay Intel to continue developing Itanium chips through the end of 2017. Absent some kind of extension to that agreement, it's unlikely that anyone else is interested in keeping Itanium alive at this point. IT World reports that Hewlett-Packard Enterprise is the only vendor expected to ship servers that use the new CPUs.

 

Kittson chips offer either four or eight CPU cores manufactured on Intel's circa-2010 32nm process, the same one used for the very first Core i3, i5, and i7 CPUs. All the Kittson chips support Hyperthreading and ECC DDR3. The quad-core chips sport a 130W TDP, while the octo-core versions have 170W TDPs.

 

Source: http://www.pcworld.com/article/3196080/data-center/intels-itanium-once-destined-to-replace-x86-in-pcs-hits-end-of-line.html

[Kherm]

Good. It deserved to die. 

[vanished]

wait, they were still making those?

[mariushm]

They had some contract with HP , some collaboration where  Intel was supposed to keep making these for some period of time, Intel gave up on it and HP sued them and won, so now just to not pay huge fines to HP, Intel's doing the minimum effort required to respect that contract .. by launching these Itanium processors in very limited quantity.

 

At least that's how i read it...

 

[2FA]

3 minutes ago, Ryan_Vickers said:

wait, they were still making those?

They had a few customers that had mission critical systems that relied upon them, including the US government. They literally couldn't afford any downtime.

[Remixt]

Just another reason to hate HP.

[flibberdipper]

RIP Itanium. It needs a cookie for trying.

[porina]

7 minutes ago, mariushm said:

They had some contract with HP , some collaboration where  Intel was supposed to keep making these for some period of time, Intel gave up on it and HP sued them and won, so now just to not pay huge fines to HP, Intel's doing the minimum effort required to respect that contract .. by launching these Itanium processors in very limited quantity.

 

At least that's how i read it...

This could use fact checking. My memory involves software vendors claiming discontinuation, which I think was what HP sued over, not Intel. But I could be very wrong on that...

 

As for the CPUs, I do wonder if someone were to do an optimised software test of it vs AMD64, how would it do? AMD64 was a bodge onto 32-bit onto 16-bit onto 8-bit Intel. While much old baggage has been kicked, it would still be different from something designed ground up to be 64-bit. I really do wonder where we would be today if AMD hadn't played the lowest common denominator card, like they're doing to a lesser extent with FreeSync now.

[Jito463]

3 hours ago, Ryan_Vickers said:

wait, they were still making those?

Surprises me, too.  I thought the Itanic was already dead and buried.

 

3 hours ago, porina said:

As for the CPUs, I do wonder if someone were to do an optimised software test of it vs AMD64, how would it do? AMD64 was a bodge onto 32-bit onto 16-bit onto 8-bit Intel. While much old baggage has been kicked, it would still be different from something designed ground up to be 64-bit. I really do wonder where we would be today if AMD hadn't played the lowest common denominator card, like they're doing to a lesser extent with FreeSync now.

It's probable that IA64 would run faster than AMD64, just purely based on the fact that it's strictly a 64-bit chip.  However, that's also what lead to it's demise.  The Itanic would have caused a major upheaval in software development, either forcing everyone to switch to IA64, or fragmenting the market into x86 and IA64.  At least what AMD did allowed for software developers to slowly transition to 64-bit.

 

In my opinion, the biggest problem has been Microsoft refusing to force the switch to x64 with Windows.  If they had abandoned 32-bit builds of Windows long ago, it would have gone a long way towards urging developers to abandon x86 architecture and focus on coding in x86-64 (a.k.a. x64).  After all, even the x64 version of Windows is capable of running almost all of the legacy x86 applications.

[DuckDodgers]

16 minutes ago, porina said:

As for the CPUs, I do wonder if someone were to do an optimised software test of it vs AMD64, how would it do? AMD64 was a bodge onto 32-bit onto 16-bit onto 8-bit Intel. While much old baggage has been kicked, it would still be different from something designed ground up to be 64-bit. I really do wonder where we would be today if AMD hadn't played the lowest common denominator card, like they're doing to a lesser extent with FreeSync now.

Despite the intentions, Itanium wouldn't have the chance in the consumer market. Its very specific architecture and instruction encoding (VLIW-like) required copious amounts of L2 and L3 caches and high-performance and power-hungry logic (the FPU performance is still top-notch, though) made it very questionable whether it can scale down to a low-power/mobile form-factors.

 

The zero compatibility with any existing applications, compilers and development experience was, by far, the heaviest stone on its neck back in the day. Even the notoriously high-level of reliability and resistance to errors (Itanium can self-correct program execution on-the-fly) couldn't hold too much awe.

[spartaman64]

5 hours ago, DeadEyePsycho said:

They had a few customers that had mission critical systems that relied upon them, including the US government. They literally couldn't afford any downtime.

why would intel manufacturing them help them avoid down time? do Itanium system not need to shut down when switching cpus

[vanished]

2 minutes ago, spartaman64 said:

why would intel manufacturing them help them avoid down time? do Itanium system not need to shut down when switching cpus

My guess is that whatever software they were running on these was custom and not compatible with anything else, since it's not your usual x86, and rather than adapt and move to something else, the military just made them continue making the chips, sort of like how Microsoft had to keep supporting XP for them

[Curufinwe_wins]

1 hour ago, spartaman64 said:

why would intel manufacturing them help them avoid down time? do Itanium system not need to shut down when switching cpus

 

1 hour ago, Ryan_Vickers said:

My guess is that whatever software they were running on these was custom and not compatible with anything else, since it's not your usual x86, and rather than adapt and move to something else, the military just made them continue making the chips, sort of like how Microsoft had to keep supporting XP for them

 

Actually... since you asked... You can hotswap Itanium processors. Yes. Seriously.

 

 

 

 

 

 

 

 

 

 

 

 

[8uhbbhu8]

7 minutes ago, Curufinwe_wins said:

 

 

Actually... since you asked... You can hotswap Itanium processors. Yes. Seriously.

 

 

 

 

 

 

 

 

 

 

 

 

OK that's pretty damn cool

[Curufinwe_wins]

8 minutes ago, 8uhbbhu8 said:

OK that's pretty damn cool

It really is. Also 12 executions per thread per cycle. 

[leadeater]

7 hours ago, Jito463 said:

Surprises me, too.  I thought the Itanic was already dead and buried.

 

It's probable that IA64 would run faster than AMD64, just purely based on the fact that it's strictly a 64-bit chip.  However, that's also what lead to it's demise.  The Itanic would have caused a major upheaval in software development, either forcing everyone to switch to IA64, or fragmenting the market into x86 and IA64.  At least what AMD did allowed for software developers to slowly transition to 64-bit.

 

In my opinion, the biggest problem has been Microsoft refusing to force the switch to x64 with Windows.  If they had abandoned 32-bit builds of Windows long ago, it would have gone a long way towards urging developers to abandon x86 architecture and focus on coding in x86-64 (a.k.a. x64).  After all, even the x64 version of Windows is capable of running almost all of the legacy x86 applications.

I take zero credit for the following but this person does seem to know what they are talking about, TL;DR Itanium is extremely dependent on your code and the compiler where as x86-64 is a much more general purpose architecture that can handle it's own instruction scheduling/ordering etc.

 

Quote

MU_Engineer  Sep 4, 2012, 8:21 PM

icebrian said:

Hello everyone, 

I revert to these forums as I am unable to find any specific information.

Currently I have a HP RX7640 (http://h18000.www1.hp.com/products/quickspecs/12470_div/12470_div.PDF) with 6 Dual-core Itanium chips.

I am considering moving over to X86 and am trying to understand what Xeon chips would be the equivalent or better than the Itanium.

On the RX7640 I have 4 Vpar's. All are running various instances of Oracle BD's both EE and SE. I am only running BDs.

Not sure if this amount of information is enough but am hoping it is. 

What would like to get is someone that could confirm the starting from a certain Xeon chip processing power will be equl to or better than 1 Itanium chip.

Thanks in advance,
Regards

This is very tricky to answer in real life. Most current x86_64 CPUs including the current Xeons are fairly narrow (3 or 4-issue) units with few registers (16) and large amounts of power and effort dedicated to out of order execution engines. Itanium is an extremely wide (11 issue) in-order chip with a massive 128 registers that depends very heavily on the compiler to keep itself occupied. The Itanium also has a potential very powerful FMAC FPU that Intel has yet to replicate on any Xeon. AMD only recently put an FMAC FPU on an x86_64 CPU in 2011 with the "Bulldozer" AMD FX series and 3200/4200/6200 series Opterons. Itanium performance as a result is VERY sensitive to the particular code it is running and the compiler it is compiled with. You can have a MASSIVE difference in performance if you have branchy low-IPC code compiled with a poor compiler or poor options vs. hand-tuned, high-IPC code compiled with a compiler with the right magic juju to keep all 11 pipelines as full as possible. I have worked somewhat briefly with Itaniums for HPC and they can be very powerful or downright suck. Also, there is no direct apples-to-apples performance comparison using identical OSes and program binaries between the x86_64 Xeons and ia64 Itaniums because they have different ISAs and require different binaries. 

My best guess is that your six-CPU Itanium system is probably on a good day about as fast as a modern six-core Xeon like an E5650 or E5-2620. The current Xeons are clocked much higher, have bunches of bandwidth due to their integrated memory controllers/point-to-point buses and have excellent SIMD capabilities and a bunch of cores. Your Itaniums are five years old, low-clocked, likely underutilized by less than perfect binaries, and bus-choked by their old FSB architecture. If you really want to guarantee that you end up with a faster system buy yourself a dual 8-core Xeon E5-26xx system or a dual 16-core AMD Opteron system. 
 

icebrian said:

Thanks Jay. I had actually already found these links. What I can see is that Intel sells the E7 as an equivalente it doesn't however include allot of information on why. Also I I still am in doubt if the there might be inferior Intel X86 processors to the E7 that could acomplish the same as the Itanium.

The E7 is an equivalent to the Itaniums because the E7s and the Itaniums are both very expensive, have the ability to be made into 4-CPU and greater systems, and have roughly similar RAS (reliability, availability, and serviceability) capabilities. E7s are very, very expensive because of their 4-CPU+ capabilities and you will do better with the two-CPU-only E5 series from a cost and performance standpoint.

In short, I recommend that you get yourself an inexpensive but decent desktop system such as an i5-3550K or AMD FX-8150 and run Linux on it, and see how it stacks up to the Itaniums running Linux. Base your server purchase accordingly based on the number of CPUs and clock speed you feel you need to get the performance you need.

 

[Jito463]

51 minutes ago, Curufinwe_wins said:

Actually... since you asked... You can hotswap Itanium processors. Yes. Seriously.

As impressive as that is, I'm leery enough of hot swapping SATA drives, much less hot swapping a CPU.

[Taf the Ghost]

I'm still amazed they were making chips for it.  Wow.

 

I do remember those days in the computer space. Itanium just wasn't a consumer-level CPU and there was already the issue with the x86-32 ecosystem being absolutely massive. The power of x86 has always been the "general purpose" aspect of it, and switching to an entirely new ecosystem just wasn't going to just happen. Yet, the server-side badly needed the memory address space.  (Which is most of what even consumer level activity is really effected by the instruction set.)

 

There was an open-source community building for Itanium, and some of the larger projects had specific builds for it.  Though I haven't seen one in ages.

[Vode]

9 hours ago, Ryan_Vickers said:

wait, they were still making those?

Itanium still makes Intel more money in the Server space than AMD gets, and it's only a tiny fraction compared to Itel's x86. lulz

 

Ironically Itanium "got killed" by AMD, whose product makes less money (right now) than said "killed" product.

[vanished]

Just now, Vode said:

Itanium still makes Intel more money in the Server space than AMD gets, and it's only a tiny fraction compared to Itel's x86. lulz

Really!? wow  seriously, I thought these were a thing of a bygone era...

[Vode]

6 minutes ago, Ryan_Vickers said:

Really!? wow  seriously, I thought these were a thing of a bygone era...

Yes, I was surprized myself when I read into it a couple months ago.

 

In their Mainframes and mission critical systems companies sometimes can't afford to upgrade to a wholly different system and hence just swap components. Especially when the software was specifically tailoed to Itanium.

 

Similar situation to the ATMs running WinXP or Win2000.