Intel Details Eight-Core Poulson Itanium Processor 102
MojoKid writes "Intel has unveiled details of their new Itanium 9500 family, codenamed Poulson, and the new CPU appears to be the most significant refresh Intel has ever done to the Itanium architecture. Moving from 65nm to 32nm technology substantially reduces power consumption and increases clock speeds, but Intel has also overhauled virtually every aspect of the CPU. Poulson can issue 11 instructions per cycle compared to the previous generation Itanium's six. It adds execution units and re-balances those units to favor server workloads over HPC and workstation capabilities. Its multi-threading capabilities have been overhauled and it uses faster QPI links between CPU cores. The L3 cache design has also changed. Previous Itanium 9300 processors had a dedicated L3 cache for each core. Poulson, in contrast, has a unified L3 that's attached to all its cores by a common ring bus. All told, the new architecture is claimed to offer more than twice the performance of the previous generation Itanium."
Why? (Score:4, Interesting)
I was under the impression that Itanium was all but dead. I'm guessing Intel must be contract bound to bring out new versions.
Re: (Score:1)
If that was the case, why bother making performance improvements inside the core? Why not just move it to 32nm and double/triple the number of cores / socket?
Though I agree, this was likely a significant loss on intel's books.
Re:Why? (Score:5, Funny)
In death, an agent of project Itanium has a name
His name is Robert Poulson
Re: (Score:1)
I came for this joke. /. does not disappoint.
Re: (Score:3)
Yeah, I'm sure there was a big argument with oracle threatening to sue when Intel said they were dropping Itanium architecture several months ago.
Re: (Score:2)
Re: (Score:3)
The recommended ASP is ~$4000/tray. Anyone know how many itaniums are there in a tray? Multiply the unit price by 200k, and you'll get the cash that Intel would be making on those.
But honestly, there are some markets Intel should attack w/ this CPU. For starters, supercomputers. The platform from Cray discussed yesterday - that one looks just perfect for a whole bunch of these. There are quite a few supercomputer projects in a number of countries, and Intel should target the Itanium at all of them.
Re: (Score:2)
But honestly, there are some markets Intel should attack w/ this CPU. For starters, supercomputers. The platform from Cray discussed yesterday - that one looks just perfect for a whole bunch of these. There are quite a few supercomputer projects in a number of countries, and Intel should target the Itanium at all of them. That alone would have a bunch of them flying off the shelves.
Er, no. Itanic is just an over-grown, over-engineered DSP. The GPUs that they use as co-processors in supercomputers these d
Re: (Score:1)
The EPIC architecture (which is looked upon as a continuation of the development of the WLIV architecture) is significantly different from other more wide-spread architectures and perhaps the performance issues are there because people have not yet f
Re: (Score:2)
RISC was actually the optimal CPU architecture. CISC had a lot of things, such as variable instruction lengths, different modes of addressing and so on that complicated the hardware. RISC simplified some of that by reducing the number of instructions that were needed since all the programming was done in higher level languages like C, but still kept techniques like branch prediction, speculative execution and register renaming in the CPU itself. As a result, RISC never had problems maintaining compatibil
Re: (Score:1)
Re: (Score:2)
Another approach would be to add an abstraction layer between the hardware and software very much like what is done with virtualization, Java, ZFS, LVM, DirectX, Crossbow et al. That would make the software more independent of the underlying hardware...
Isn't that basically how CISC works nowadays?
Re: (Score:1)
As an abs
Re: (Score:2)
Re: (Score:1)
Re: (Score:1)
It's GOOD to be an important supplier to a black project with a black budget. *cough* NSA *cough*
There will be trainloads sent to Bluffdale, Utah, in boxes labeled as containing Donnie and Marie CD's. I imagine much if not most of the development was done by SAIC contractors with TS clearances. There will no doubt be a few thousand crippled versions marketed though the normal channels.
Re: (Score:2)
The rest of us jerk off to pictures of girls instead of conspiracy theories. Try it one day!
Re: (Score:1)
Great news for Oracle databases! (Score:3)
What's twice a small number? (Score:3)
My leaky/biased memory says these machines were a speed disappointment. Is this doubling going to make them faster or slower than an x86?
--dave
Re: (Score:3)
https://en.wikipedia.org/wiki/Itanium#Itanium_.28Merced.29:_2001 [wikipedia.org]
One of the traps Intel tends to fall into, at least according to someone I know who worked there during the Itanium "hype days," is that the architecture team does not communicate with the compiler team. Both Itanium and x86 fall into this trap, although x86 is far more illustrative of the problem (most compil
Re: (Score:3, Interesting)
Perhaps Intel fell into the trap of not communicating with the compiler team, but HP certainly did not.
The development of the EPIC concept at HP already started in 1992 as the to-be-successor for the HP-PA architecture. Look up e.g. the many joint research papers of CPU-architecture and compiler engineers for PlayDoh (or see e.g. http://www.hpl.hp.com/techreports/93/HPL-93-80.html for an intro to PlayDoh).
The compiler technology to do well for EPIC architectures was mostly available by the time IA64 launche
Re: (Score:2)
Might all be urban legend, or misremembered, or I might be on drugs.
Re: (Score:2)
Re: (Score:2)
This has nothing to do with availability, or even cost per unit factor. These machines were plagued with the same problem that Alpha servers had... they weigh in excess of 200lbs. Even as a hobby, for a free machine, I'm not paying shipping on that bastard
Re: (Score:2)
Just as the AlphaServer had desktop equivalents, so did the Itanium, which were discontinued.
Re: (Score:2)
HP had in the 90s acquired 2 VLIW companies - Multiflow and Cydrome - and already had a lead in VLIW compiler technology. Once they made the alliance w/ Intel, they had the grand vision of replacing both the x86 and PA-RISC lines w/ the successor Merced. As I pointed out above, leading RISC CPUs were already adapting MIMD techniques intrinsic to VLIW, while moving the dynamic analysis to the CPU to the compiler didn't do much for the CPU real estate, since they weren't using much to begin w/.
Yeah, Intel
Re: (Score:1)
The current Xeon E5-2670 (8 core, 2.6GHz, 2012) can do roughly 4x the performance of the previous Itanium 9350 chip (4 core, 1.73GHz, 2010), according to spec.org CPU2006 benchmark results. I think Itaniums do slightly better with FP, the Xeons win with INT.
But that's per chip, and the Itanium systems are going up to 8 chips, so a single 8 socket Itanium system was getting roughly the same performance (in 2010) as a 2-socket E5-2670 in 2012. I don't think the Xeons go up to 8-sockets.
Re: (Score:2)
I don't think the Xeons go up to 8-sockets.
Intel do have xeon processors that support 8-socket systems and afaict at least HP and supermicro make 8-socket xeon solutions (I think HP sell them as fully-built servers while supermicro sell them as a "barebones" system to which you add processors and drives yourself).
However afaict the processors that support 8 socket setups are both underwhelming (high core counts but low clockspeeds and still on nahelm technology) and expensive compare to those for 2-socket systems.
Re: (Score:2)
My leaky/biased memory says these machines were a speed disappointment. Is this doubling going to make them faster or slower than an x86?
--dave
The big issue, IIRC, is that Itanium was dead slow in its x86 emulation in the first few rounds. Intel's ideas as initially to emulate the x86 chips in software so that the Itanium wouldn't lose their x86 market and they could switch everyone over. They later went back and remove the software emulation and put an x86 die on to do the work in order to make it faster.
In native mode, I've never heard a complaint about Itanium and speed - only its x86 support mode.
Re: (Score:2)
I looked at some older TPC results, and see the previous Itanium delivering 4/7 the speed of the T5440, one of Sun's oldest threads-not-clock-speed boxes. Compared to IBM Power 7, Itanium delivered 4/10, so the doubling should being it up to 80% of the IBM.
Not to be sneezed at! Nevertheless, not competitive with Power, Fujitsu (Sun) M series or even the new Sun T4 boxes.
--dave
Re: (Score:2)
I looked at some older TPC results, and see the previous Itanium delivering 4/7 the speed of the T5440, one of Sun's oldest threads-not-clock-speed boxes. Compared to IBM Power 7, Itanium delivered 4/10, so the doubling should being it up to 80% of the IBM.
Not to be sneezed at! Nevertheless, not competitive with Power, Fujitsu (Sun) M series or even the new Sun T4 boxes.
One question that begs: were those TPC tests done on Itanium optimized well enough for Itanium? Or was there another bottleneck other than the processor?
One of the early on issues with Itanium was that it was hard for the optimizers to get right. I think they solved that, but I don't know when.
And of course it is hard to make apples-to-apples comparisons between architectures unless you have a reference system where the only thing you change is the processor, and verify that the code running on top of
Re: (Score:2)
Why waste a product run? (Score:1)
From Intel's view as an innovation company, it kind of makes sense to try out new stuff on a platform that will not matter that much.
And since they know HP will buy them, Intel know they will be field tested.
Re: (Score:1)
By 'more', don't you mean 'everything'? I'm no Itanium expert, but I was under the impression that the compiler had to tell it precisely which instructions it could execute in parallel in any clock cycle.
We learned more than a decade ago that relying on the compiler to tell the CPU how to work was insane. I have very not fond memories of early RISC CPUs which didn't have any instruction interlocks so you had to order instructions to ensure a calculation would be complete by the time you read the result.
Re: (Score:2)
If OOOE is out-of-order execution, itanium does oooe fine. It just expects compiler to tell more about it.
We learned more than a decade ago that relying on the compiler to tell the CPU how to work was insane.
Correct me if I'm wrong, but IIRC one very major problem with Itanium was that Intel, having designed it around this philosophy, never properly implemented (or were able to implement) the compilers it relied on to do this.
Re: (Score:2)
It would be nice for virtual machines. Branch prediction is where VMs shine. But I can't see Sun (the Java people) having a big interest in it, and LLVM wasn't so widely used when Itanium had mindshare.
Re: (Score:2)
If OOOE is out-of-order execution, itanium does oooe fine. It just expects compiler to tell more about it.
They have clairvoyant [skepdic.com] compilers now, do they?
Thank you HP? (Score:4, Interesting)
Re: (Score:2)
It's because they spent a shit ton of money porting software to it. They don't want to have to incur that cost again to port away.
Re: (Score:2)
Still who is going to buy it now?
Remember the Alpha? Slashdot ran on Alpha's for 5 years. They had a new version out and it didn't matter. HP wanted Itanium and purposedly made sure people wouldn't buy it and crippled the product line for the inferior Itanium. Makes you wonder why they bought it?
After Windows 2000 dropped support in RC 3, it didn't matter. Who in their right mind would invest in a dead platform?
This new chip could be 20x faster than a xeon and use 1/10 of the power! No one wants to invest i
Re: (Score:2)
Re: (Score:1)
I wonder what management is going to do or are doing? I expect they are already underway replacing them. I doubt HP is porting them to x86 or ARM as it maybe too late for those that are retiring these with win32 or Linux equilivent of different applications that do the same tasks. It is not like you can get an emulator for these but these are systems I would not want to invest a penny into anymore as it would be a penny lost 3 years down the road when intel stops production and I can no longer even get moth
Re: (Score:2)
If you think some brand name beige Linux box is going to replace a nonstop system do yourself a favour and come out of mom's basement.
Nonstop actually means what they say.
No. Stops.
Period.
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
Re: (Score:3)
Hell, on the OpenVMS side, it wouldn't shock me a bit to find out that they don't even HAVE a team any more that's capable of porting it to other architectures. They likely say they do, to fulfill government contracts that specify that OpenVMS can't be orphaned, but I wonder what the reality is.
Re: (Score:2)
The same thing Apple saw in MIPS
Re: (Score:2)
If HP ditch Itanium, they effectively ditch HP-UX. They can (have?) ported HP-UX to x86, but why would anyone pay top $ for HP-UX on x86 - they would just use Linux instead. Without HP-UX, they don't have a tier 1 platform & will be drowned by Red Hat & SuSE.
Meanwhile, Intel is busy building the RAS features of Itanium into x86 - as these get implemented into Linux, HP-UX will become irrelevent anyway.
IBM & Power have a little more headroom - be interesting to see how long it lasts.
Too bad (Score:1)
We already switched. ... ok a former customer I worked with already switched.
Thank you Oracle for convincing us that it is dead.
No one will touch it with a 10 foot pole. I hope HP wins the lawsuit agaisnt them and Intel also sued Oracle for damages. When they violated that contract it gave a lot of hurt for those who have invested so much in Itanium.
Now it doesn't matter as no one will touch it.
The most important improvement... (Score:5, Funny)
From TFA:
Poulson can issue 11 instructions per cycle compared to Tukwila's six.
These go to eleven.
You can still buy Itanium?!? (Score:4, Funny)
You can still buy Itanium chips? Holy crap. Are they found on the same aisle of the department store as the iceboxes and cotton gins?
It's simple: (Score:2)
This is just Intel putting on a show of competing with themselves so that they don't get accused of monopolistic behavior... :p
Behind x86 in process (Score:2)
This is an annoucment for a 32nm Itanium. Intel has been shipping 22nm x86 since spring.
Re: (Score:2)
Intel has been shipping 22nm x86 since spring.
It seems that in the Intel x86 world the higher you move up the product line the older the technology gets.
Intels x86 processors right now are best grouped by the sockets they use. There are basically three "current" (that is not "replaced" by a newer socket) sockets.
LGA1155 is the mainstream desktop and low end single socket server socket. This is the only socket for which 22nm parts are currently available.
LGA1356 is intended for low end dual socket systems but I get the impression it didn't really catch
Re: (Score:2)
It's easy to criticize Itanium in hindsight (Score:2)
If it hadn't been for AMD's 64 bit extensions, we'd all be running Itanium servers right now. AMD forced Intel to release a 64bit x86. If AMD hadn't, all of the effort that is being put into Intel's current 64bit chips would have go into Itanium and it would be a very strong platform. The alternative, PAE, sucked.
Re: (Score:1)
Well as long as you ignore that all the legacy x86 software that is still running today wasn't going to be ported to Itanium. People would have just stuck to x86 rather than spending billions on porting and rebuying working software.
Re: (Score:2)
The Itanium is still around? (Score:2)
Do these chips make the user's face glow blue ? (Score:2)
hothardware eh? (Score:2)
Sounds like the itanic all right.
They had one at LinuxExpo once, back in the day, allegedly running DeadRat, but we couldn't see it because it had overheated and they took it away.
Its name was Itanic Poulson... (Score:1)
its name was Itanic Poulson... its name was...
There are a hell of a lot of Itanium users (Score:4, Funny)
I'll be buying a number of systems with these in a few months when they hit the street and the budget's ready. I'll be able to virtualize a lot of our old PA-RISC boxes into a smaller and more efficient set of systems.
But you're right, they suck because you can't play Angry Birds on it.
Re: (Score:2)
Intel provides an Itanium reference board [intel.com] that makes it possible for other manufacturers to release OEM Itanium based systems. As a second manufacturer example, I've used on of Bull's Novascale Bullion [bull.com] servers. It wasn't very cost effective, but it did include 256 cores, and continued running just fine when one socket was damaged during shipping. The sort of applications that need that many cores and heavy redundancy against hardware failures exist, and no commodity hardware will satisfy them. There's j
Re: (Score:1)
Intel would have a wet dream if a few hundred thousand titanium systems were sold each year. I'd revise that number down a few orders.
The number of non-HP itanium vendors is very very very very narrow and getting smaller by the day. Niche products.
The market for 256+ core single image systems is also vanishingly small.
Re: (Score:2)
What I'm wondering is why Intel doesn't sit down w/ organizations that still support Itanium, such as Debian and FreeBSD, to optimize those OSs for the Itanium. Similarly, they could also work w/ the GCC and Clang guys to make sure that additional work gets done to finetune those compilers so that they can make the best use of these CPUs.
But these CPUs seem more ideal for supercomputing work, not the Google or Facebook types. Their needs, aside from Xeon or Opteron, can be met even by SPARC servers - th
Re: (Score:1)
What I'm wondering is why Intel doesn't sit down w/ organizations that still support Itanium, such as Debian and FreeBSD, to optimize those OSs for the Itanium.
Because it wouldn't benefit them economically?
Similarly, they could also work w/ the GCC and Clang guys to make sure that additional work gets done to finetune those compilers so that they can make the best use of these CPUs.
Why would they bother? They already dropped IA-64 support in their own compilers after version 11.1.
Re: (Score:1)
With the software requiring to fill your raised floor with Itanic servers, the cost of hardware usually is the lesser of your worries.
A project for you. (Score:1)
Go find an open source or commodity system that can be deployed in a heavily regulated power industry with SCADA systems.
Make sure it's so cheap that the difference in cost for buying Itaniums and this software will pay the millions in training people all over the country, interfacing in the financial and billing systems, as well as covering the cost of redeveloping all of the customized code that is required to operate coal, natural gas and nuclear plants.
Please call me when you're done.
Re: (Score:2)
Re: (Score:2)
Uh, no, Itanium could only run Windows Server 2008, but that too was Itanium II. For Itanium III, software would have to be recompiled, and since Microsoft has dropped support for that CPU, you can be sure that it won't happen.
Heck, even the Linuxes have dropped support for Itanium - the only exception being Debian. On the BSD side, only FBSD supported it initially, although looks like NBSD support might be there in version 6.0. But w/ the compatibility breakage b/w each generation, looks like it would
Re: (Score:2)
I ought itanic it n ceberg nd ank nto he ea
Re:Failtanium (Score:5, Funny)
Re: (Score:2)
That's funny, because I've personally worked on hundreds of them.
*woosh*
Re: (Score:1)
3 Words: VMS + Fortran + Mission Critical
If it's going to cost you 250M to migrate to another platform, or 20M to buy replacment itanic hardware,
which one are you going to do?
Companies and large institutions with these kinds of equations exist in many places.
They are the ones that actually needed computers when computers 1st came out.
Re: (Score:2)
Balls, that was supposed to be a 'funny' mod.