Intel officially kills off its IA-64 CPU

[GoodBytes]

Intel quietly kills off it's Itanium line of CPU, which where CPUs using Intel's IA-64 architecture instead of x86, via a notification sent and it's CPUs line marked as End-Of-Life. Intel says that it will no longer sell the CPU to new customers. The latest model was the 2017's Itanium 9700 code name: Kittson. The Itanium 9700 was a 32nm process CPU, with models going from 1.73GHz to 2.66 GHz. It has little performance increase over the older model... from 2012, which features identical specs. The only specs difference is that the highest end model of the previous gen was limited to 2.53Ghz, and now 2.66GHz as mentioned.

 

Quote

Intel has unceremoniously, through a product change notification (PCN), discontinued the Itanium family of microprocessors. The Itanium 9700 "Kittson," which was released in 2017, is the final generation of Itanium, and its sales to new customers have stopped according to the PCN. The series has been marked "end of life" (EOL). Existing customers of Itanium who already have their IT infrastructure built around Itanium 9700 series, have an opportunity to determine their remaining demand of these processors, and place their "Last Product Discontinuance" order with Intel. The final LPD shipments would go out mid-2021.

 

Back in the early 2000's, Itanium was Intel's answer to jump from 32-bit to 64-bit. It was aimed to replaced x86 architecture, with a new 64-bit only architecture which the company called: IA-64.

Sadly, for Intel, it didn't catch on. Instead AMD64, by AMD, did. This is because AMD64 was a 64-bit add-on of sorts to x86 architecture, hence why we refer it to x86-64, which allows full backward compatibility with x86 without performance loss. While Intel's IA-64 needed software to be recompiled specifically for it, adding massive cost to enterprise to switch over, not to mention face with the issue of software support. As a result, AMD's AMD64 won the race. Oh and AMD focused its effort to target consumers, while Intel targeted enterprises server applications.

 

Source: https://www.techpowerup.com/252079/intel-officially-sinks-the-itanic-future-of-ia-64-architecture-uncertain

[paddy-stone]

4 minutes ago, GoodBytes said:

it didn't caught on

Should be "catch", hope you don't mind me correcting?

[rcmaehl]

5 minutes ago, GoodBytes said:

Intel quietly kills off it's Itanium line of CPU, which where CPUs using Intel's IA-64 architecture instead of x86, via a notification sent and it's CPUs line marked as End-Of-Life

Intel says that it will no longer sell the CPU to new customers. The latest model was the 2017's Itanium 9700 code name: Kittson.

 

 

Back in the early 2000's, Itanium was Intel's answer to jump from 32-bit to 64-bit. It was aimed to replaced x86 architecture, with a new 64-bit only architecture which the company called: IA-64.

Sadly, for Intel, it didn't caught on. Instead AMD64, by AMD, did. This is because AMD64 was a 64-bit add-on of sorts to x86 architecture, hence why we refer it to x86-64, which allows full backward compatibility with x86 without performance loss. While Intel's IA-64 needed software to be recompiled specifically for it, adding massive cost to enterprise to switch over, not to mention face with the issue of software support. As a result, AMD's AMD64 won the race. Oh and AMD focused its effort to target consumers, while Intel targeted enterprises server applications.

 

Source: https://www.techpowerup.com/252079/intel-officially-sinks-the-itanic-future-of-ia-64-architecture-uncertain

This has been a long time coming. Not sure I'll miss it.

[GoodBytes]

2 minutes ago, paddy-stone said:

Should be "catch", hope you don't mind me correcting?

Thanks!

[Taf the Ghost]

Been expected for a while, but I can't really say it'll be missed. There was some cool stuff about it, but it probably needed to have landed in the mid-90s for it to have worked. x86 CPUs would increase in single-core clocks by several hundred % during it's initial build-up period. Those insanely rapid node shrinks from 1990 to 2009 really did any other ISA's approach in until we started hitting silicon limitations. 

 

So, well...

 

 

F

[mariushm]

They killed it some time ago, but they were sued by HP if I remember correctly and were forced to keep producing it, because they had agreements and contracts with HP to guarantee availability for some period of time.

I guess that period is finished at the end of this year, or something like that.

[williamcll]

What was the use of these processors?

[GoodBytes]

58 minutes ago, williamcll said:

What was the use of these processors?

It was a different architecture, a different approach to do things at the CPU level.

Intel targeted the CPU for Servers as a 64-bit version its Xeon CPUs (before Intel integrated 64-bit support to x86), and it was supposed to be faster than x86, if their marketing material was to be believed. Intel probably targeted the server space due to the fact that beside Windows and select Linux distros, you pretty much had very limited library of software that works with this CPU (so it doesn't fit consumers needs). And probably you can add the higher profit margins that the company gets from the enterprise sector over consumer to the mix.

 

Itanium gain a reputation of being super hot chips, very power consuming, and as a result, very loud due to the beefy cooling solution that was needed. It obviously got better with each generation, but by the time it catched up to something much better, but still needs work, the needs of the market changed. The popularity of using GPUs for processing grew as OpenCL and CUDA hit the scene (with supported GPUs). And if that was not needed, Xeon which had 64-bit support at the time, offered similar performance for most applications, lower price tag, cooler operating (enterprise where really interested in lower power consumption as when you have many servers, the power bill quickly becomes an issue.. don't forget the A/C needed as well to cool the server room), and not to mention a wide range of software solutions available as it was a x86 based.

 

[Taf the Ghost]

3 minutes ago, GoodBytes said:

It was a different architecture, a different approach to do things at the CPU level.

Intel targeted the CPU for Servers as a 64-bit version its Xeon CPUs (before Intel integrated 64-bit support to x86), and it was supposed to be faster than x86, if their marketing material was to be believed. Intel probably targeted the server space due to the fact that beside Windows and select Linux distros, you pretty much had very limited library of software that works with this CPU. And probably you can add the higher profit margins that the company gets from the enterprise sector over consumer.

 

Itanium gain a reputation of being super hot chips, very power consuming, and as a result, very loud due to the beefy cooling solution. It obviously got better with each generation, but by the time it catches up to something much better, but still needs work, the needs changed. The popularity of using GPUs for processing grew as OpenCL and CUDA hit the scene. And if that was not needed, Xeon which had 64-bit support at the time, offered similar performance for most applications, lower price tag, cooler operating (enterprise where really interested in lower power consumption as when you have many servers, the power bill quickly becomes an issue.. don't forget the A/C needed as well to cool the server room), and not to mention a wide range of software solutions available.

 

Real problem, as explained to me back when these first came out, is that porting programs to it was a nightmare. If you built custom software for them, it was supposedly pretty great, but, then, why wouldn't you just use Power or SPARC or one of the few others still around at the time? It wasn't bad tech, but, except for the memory allocation, it really didn't fill a need for most of the server market. We talk about AMD's troubles getting back into the x86 server market, and it's pretty similar set of reasons it's a problem. With IA-64, you also had to redo all of your programming. 

 

It was destined to fail, but I've never heard a convincing path for it to have taken over.

[S w a t s o n]

What the fuck? They were still making Itanium shit in 2017? Who was buying these and better question what the fuck were they running on them? This is hilarious

[aisle9]

Itanium was a lot like How I Met Your Mother. Interesting premise, but after a few seasons it was clearly not living up to the hype and they signed a huge long-term deal to keep it going anyway.

[Stefan Payne]

On 1/31/2019 at 3:00 PM, GoodBytes said:

Intel quietly kills off it's Itanium line of CPU

Wait? They were still running or selling them??

But I never liked the EPIC Concept to move all the branch prediction to the software side. That looked like a dumb thing to do.

 

Quote

The latest model was the 2017's Itanium 9700 code name: Kittson. The Itanium 9700 was a 32nm process CPU, with models going from 1.73GHz to 2.66 GHz.

Ui, the made a new one just 2 years ago?? 

But 32nm in 2017? SRYSLY? 22nm was the thing for them at the time...

And even 14nm was available, according to Wikichip.

 

Quote

Sadly, for Intel, it didn't catch on. Instead AMD64, by AMD, did. This is because AMD64 was a 64-bit add-on of sorts to x86 architecture, hence why we refer it to x86-64, which allows full backward compatibility with x86 without performance loss.

Yeah, good for the consumers in the end.

and i think the EPIC Architecture was kinda bullshit anyway...

IIRC the Code was huge, the moved all the branch prediction and other stuff into the compiler...

And were basically In order CPUs.

 

But later they moved "back" to out of order excecution with branch prediction and all the bells and whistles...

 

But IMO 64bit was too late anyway. Should have come far sooner!
When you need shit like PAE and other extensions for "large memory", something is wrong. 

 

Especially if you look at DEC Alpha, wich was 64bit from the beginning or so, back in the 90s.

And Intel bought DEC!

So why didn't they build on that instead??

Well, sadly the Slot A Alpha never happened

[AnonymousGuy]

1 minute ago, Stefan Payne said:

Wait? They were still running or selling them??

But I never liked the EPIC Concept to move all the branch prediction to the software side. That looked like a dumb thing to do.

Itanium was basically killed off 6 years ago but HP was paying $ directly to Intel's pocket to keep supporting / "developing" it.  That's why the most recent one was 32nm because it was basically a rebinned Paulson part.  No actual engineering effort was spent on it.

[Mira Yurizaki]

34 minutes ago, Stefan Payne said:

But I never liked the EPIC Concept to move all the branch prediction to the software side. That looked like a dumb thing to do.

In hindsight, sure. But back then, it was the next logical step in improving CPU performance. But the concept wasn't abandoned entirely. Elbrus uses VLIW with an x86 emulation mode and seems kind of promising with running it (https://www.extremetech.com/computing/205463-shadows-of-itanium-russian-firm-debuts-vliw-elbrus-4-cpu-with-onboard-x86-emulation). NVIDIA's Project Denver did the same thing with ARM.

 

And I'm pretty sure compilers are doing this now since eliminating branching where possible is low hanging fruit to increase application performance.

 

34 minutes ago, Stefan Payne said:

But 32nm in 2017? SRYSLY? 22nm was the thing for them at the time...

And even 14nm was available, according to Wikichip.

Smaller nodes would likely be taken up by higher volume orders and if you still have a production line for larger nodes that aren't in much use, why not use them?

34 minutes ago, Stefan Payne said:

IIRC the Code was huge, the moved all the branch prediction and other stuff into the compiler...

And were basically In order CPUs.

 

But later they moved "back" to out of order excecution with branch prediction and all the bells and whistles...

You can have out-of-order execution without branch prediction.

 

34 minutes ago, Stefan Payne said:

But IMO 64bit was too late anyway. Should have come far sooner!
When you need shit like PAE and other extensions for "large memory", something is wrong. 

Why?

34 minutes ago, Stefan Payne said:

Especially if you look at DEC Alpha, wich was 64bit from the beginning or so, back in the 90s.

And Intel bought DEC!

So why didn't they build on that instead??

Well, sadly the Slot A Alpha never happened

Uh, DEC was acquired by Compaq in 1998, then merged with HP in 2002. I don't see where Intel fit in this.

[Coaxialgamer]

On 2/1/2019 at 10:26 PM, Stefan Payne said:

 

But I never liked the EPIC Concept to move all the branch prediction to the software side. That looked like a dumb thing to do.

Doing so is actually better when looking at the EPIC/VLIW philosophy because it can enable better performance. Doing so in hardware isn't very feasible or practical. The problem is that the IA64 compiler never really become good enough.

[imreloadin]

On 2/1/2019 at 4:26 PM, Stefan Payne said:

Ui, the made a new one just 2 years ago?? 

But 32nm in 2017? SRYSLY? 22nm was the thing for them at the time... 

Did you really expect Intel to put forth any actual effort into a platform that was essentially dead on arrival?

[Jito463]

On 1/31/2019 at 8:00 AM, GoodBytes said:

Back in the early 2000's, Itanium was Intel's answer to jump from 32-bit to 64-bit. It was aimed to replaced x86 architecture, with a new 64-bit only architecture which the company called: IA-64.

I still believe Intel did this because they wanted to ditch their cross-license agreement with AMD.  If they created a new architecture that was completely devoid of anything x86, then AMD would have no legal claim to it.  Fortunately for us, it didn't pan out.

[Stefan Payne]

On 2/3/2019 at 3:15 PM, Jito463 said:

I still believe Intel did this, because they wanted to ditch their cross-license agreement with AMD. 

Yes, that's highly probable as well as the whole X86 patents in mind, wich would run out soon as well.

With a completely new architecture you avoid that and own all the patents of the architecture, thus excluding competition...

 

Quote

If they created a new architecture that was completely devoid of anything x86, then AMD would have no legal claim to it.  Fortunately for us, it didn't pan out.

Yeah, I agree with that.

And in the future, there might be some basic X86 Implementations out of china that are legal because the Patents are no more...