Intel should have made a stronger statement than it did about its Itanium processor at this week's launch of the "Westmere-EX" Xeon E7 high-end server processors, considering the bashing that Oracle has given the chip in recent weeks. However – for whatever reason – Kirk Skaugen, general manager of Intel's Data Center Group, …
sip the koolaid
Funny how the massive slippage of each generation of Itanium is not brought up when talking about HP rosy Itanium roadmap.
>Intel should have made a stronger statement than it did
At some point Intel is going to figure they have lost enough on Itanium and pull the plug. My guess is they only haven't done so far is because they don't want to get into an ugly patent fight with HP over the IP they borrowed.
2 year cycle my arse
Well Poulson was a different development team and Tukwila was at least 1 year late after the 2 year delay. Kittson is only a fab of the exact same chip so expect it to make the 2 year ship.
If there is any future chip after kittson it will be only a fab die shrink.
Q) Is there plans to have a chip after Kittson
A) Intel: No comment
Q) Are you working on a similar "Aries" emulator for Itanium software on x86
A) Intel: No comment
Oh and BTW...the reason NEC has follow on Itanium systems is because they just OEM the HP boxes.
Re: sip the koolaid
I'll get to that. One story at a time, lad.
Re: sipping the kool aid
"At some point Intel is going to figure they have lost enough on Itanium and pull the plug."
Surely that's the one thing that *won't* happen. The longer they keep ia64 in production, the more they get to recoup the initial R&D costs. (That is, the few billion they blew in the nineties discovering that it wouldn't work.)
I expect Intel to keep ia64 "current" for as long as there is someone out there writing software for it. However, as the article notes, that's an increasingly limited number of people. Also, and with reference to the "dual instruction decode" suggestion made further down in these comments, if someone were to produce a decent VMM for ia64 on x64, that would probably kill it. Given the differences in architecture, particularly the wide floating point, that might be quite challenging, but not impossible. At some point, *Intel* might decide that it was the cheapest way to "keep ia64 current", much as IBM did when they scrapped AS/400.
If so many Itanium features have been ported to Xeon, is there any possibility that they can merge the cores and run both architectures, much like ARM has Thumb mode?
dual instruction decoders
An interesting idea, but would that be feasible given that it's a totally different architecture (VLIW) ?
1) Is there any real need? Port/recompile the applicatons. Of course, I do not understand porting implications, not having used itanics before, but surely it can't be that completely unreasonable an option to consider.
2) imho (and I'm no chip designer) the x86 instruction set looks ugly enough, plus having to deal with legacy issues etc... Just adding further burden just seems perhaps a tad bit unwise.
Would you want die space on your xeon dedicated to emulating itanics? Or more of the other bits that are more likely to help you with your x86 code.
re: ia64 emulation
The Itanium architecture has some decent advantages over x86, so yes you can just recompile, but you'll be taking on board x86 limitations with it.
AFAIK, Itanium decode logic is a -lot- simpler than for x86, so it shouldn't take much die space.
ARM has Thumb instructions, Intel themselves put IWMMX extensions into the PXA line - so it seems a small stretch to fit Itanium and x86 onto one die so parts of the same application can run in whichever give the best performance/space metrics.
GCOS is a Bull product (actually two unrelated ones, GCOS 7 and 8), not a NEC one. NEC ACOS is derived from GCOS (ACOS-2 is based on the defunct GCOS 6, and ACOS-4 is based on GCOS 7) but it is not the same thing.
Why don't you have info on IBM's plans?
It is WIDELY known that IBM plans Power 7+ by December. It is WIDELY know that it will increase the cores-per-socket and some GHz enhancements with a die-shrink.
Why are you comfortable telling people that the info is not available when it is more-true that you have not researched it.
FYI, it is also WIDELY known that a (quite) minor GHz kick is coming out in May.
I have seen Power 8 and Power 8+ on IBM Executive presentations for two years.
re: Why don't you have info on IBM's plans?
Settle down Francis. As TPM said, IBM does not provide PUBLIC roadmaps (not very good ones at least). If TPM has seen the NDA Roadmap, then he would have to sign the NDA and could not publish it. I am assuming that since you've seen the "Executive presentations" then you have signed a NDA and should not be publishing your statements such as you are.
Now if someone were to let the IBM NDA Roadmap leak, then I believe that TPM could link to it...
"sign the NDA and could not publish it"
You're not serious are you?
Half (maybe more) the "roadmap" content at sites like this comes straight from the vendors NDA presentations, and both sides know it. It's a game that suits both sides, most of the time. Plausible deniability and all that.
re: "sign the NDA and could not publish it"
Yes, very serious. TPM cannot show the slide that he got, but as I stated, he can show the slide that you slipped to him in a dark alley.
Intel Oracle No show
Amusingly an Intel Exec was listed as giving one of the Key Notes at a local Oracle Cloud Summit conference here today and not surprisingly was a no show.
Where's Matt Bryant, telling the world that Itanium and Power are dead because Intel and IBM did not publish public roadmaps?
Is Itanium dead because Intel is releasing a new version of it's x86 architecture?
Can't wait to see the comments from the marketing hypocrites on the next non-Itanium & non-Power CPU articles!
About those dark alleys...
I guess that you have come back to my point. Why haven't you been able to assemble (on background or whatever) some general information about NDA roadmaps.
If you state that there are no roadmaps, you are being obtuse. To say the least...
Strong statement? Already out there.
As an hp customer, we got a PDF via our hp salesgrunt, outlining Intel's commitment to Itanium. It is very black and white - they are making Poulsen and Kitson. There's even a sneaky dig at Oracle in it. Sorry, NDA stops me passing it on.
And you may want to think twice about the "dark alley" method of getting round an NDA as the material would be covered by the NDA you already signed even if it came from someone else. Whilst there's not much the vendor can do legally, they can make sure you don't get any future NDA material, a real pain for a juh-nah-list.
Here is the letter
1) "Strong" commitment to Itanium
2) Targeted to $15B RISC/Mainframe market...well with only HP-UX, VMS, Non-Stop and without any Oracle product it's a pretty week arrow going after that target
3) 14,000 applications..is that counting the HP-UX software that needs the Aries emulator?
4) Tukwila more than doubled the performance of its predecessor (considering it has twice the cores its a lame statement)
5) Poulson will be delivered on Intel's newest 32nm technology (won't this be old technology by the time 2012 rolls around when everyone else is on 22nm?
6) "We are currently starting exploratory work for what comes after Kittson" (would this be an emulator on x86?)
7) continue to migrate....to industry standard solutions running on Intel Itanium...(talk about a stretch on the term industry standard)
Beam me up Kirk
re: Lame commitment
I hate agreeing with the super fudster when attacking the lame fudster, but this "black and white" commitment is pretty weak, for all the reasons that Allison points out. So, Intel and HP, what does come after Kittson? Seeing how Kittson will be released ~2014-2015, it is not hard to expect that it will be supported for 5-6 years at least, which puts us into the 10yr range that Intel and HP are espousing. Seeing how nothing will be running on it, how can they seriously expect us to buy what they're selling?
RE: Lame commitment
This should be fun!
"1) "Strong" commitment to Itanium...." And.... What, you're not denying it's a strong commitment? So, how does IBM make a "strong" commitment to Pee8? Seering as there is NO public commitment to Pee8 at all, no public statement from IBM at all, by Allison's own measure, Pee8 is vapourware and never going to arrive. People in glass houses....
"......2) Targeted to $15B RISC/Mainframe market...well with only HP-UX, VMS, Non-Stop and without any Oracle product it's a pretty week arrow going after that target...." IBMers always get frantic when someone mentions targetting their precious mainframe racket (remember PSI?). Strange that Allison assumes all IBM mainframes run nothing but Oracle software - not the last time I checked! Looks like Allison needs to do some homework on her own kit whilst she swots up on other vendors' offerings!
"......3) 14,000 applications..is that counting the HP-UX software that needs the Aries emulator?...." Nope, that's pure Itanium-ready software, as she well knows (hence the question mark so she can IMPLY the point). By the way, Alli, how many apps are there out there for Pee7, as in recompiled for Pee7, not old Power apps that don't run to their full potential on Pee7 without a full recompile and probably a rewrite? A few hundred at most, maybe a thousand?
".....4) Tukwila more than doubled the performance of its predecessor (considering it has twice the cores its a lame statement)...." Strange - Pee7 doubles the cores over Pee6 but didn't double the performance, as admitted by IBM's own presales team......
"......5) Poulson will be delivered on Intel's newest 32nm technology (won't this be old technology by the time 2012 rolls around when everyone else is on 22nm?...." And the public roadmap showing a Pee8 CPU at 22nm is.... Oh, there isn't one! I'll try and look surprised that Intel have a more detailed and believable public roadmap than any of their competitiors.
".....6) "We are currently starting exploratory work for what comes after Kittson" (would this be an emulator on x86?)...." Seeing as IBM haven't made any public "strong" commitment to Pee8, maybe you should hope there is a Power emulator in next gen Xeon.
"...... 7) continue to migrate....to industry standard solutions running on Intel Itanium...(talk about a stretch on the term industry standard)...." Anyone can buy an Itanium CPU from Intel, make their own or buy a mobo for it, and make their own servers. The fact is that hp are simply far better at it than the rest of the industry. Including IBM, who did try very hard to compete with their own IBM Itanium servers (yes, we tried the x450 for MS SQL on Itanium and went hp Integrity instead, and that was about as level a playing field as IBM could hope for).
".......Beam me up Kirk" Sorry, Alli, but given the shabby quality and inaccuracies in your posts, you'd probably not meet Star Fleet standards! News of The World, maybe.....
resorting to derogatory now?
I am shocked Matt.
1) I hear Power8 is baked but IBM does not want to give details because it has a lot more than just the typical cores/cache/threads/execution units as other chips are doing. I don't see a lot of info about Kittson either which I hear is just a fab enhancement of Poulson.
4) Power7 has 4X the cores of Power6 and also increased core performance for a 5X per footprint increase compared to Tukwila's 2X (as stated by Intel and many HP executives)
5) I hear Power8 will be 22nm just like Kittson but should show up in 2013 if IBM keeps on their 3 year cycle.
6) No one in the industry doubts IBM commitment to Power
7) Itanium is what it is and I think will be killed by Intel just like it killed PA-RISC, MIPS, Alpha
In a restrospect it is a successful chip, but really only successful in what it has done for x86
Matt, this is not about HP. It's about how Lew Platt and Rick Belluzzo decided to trust Intel with the future of HP's enterprise systems and HP and us the customers have had to pay for the poor decision. HP wanted to get out of the chip business but found themselves still in it because they had to make the glue chips (SX2000, SX3000) for the Itanium to get it to scale past 4 sockets.
It's been a disaster and the latest spat with Oracle is the final nail in the coffin. Our HP rep before he left said that they were told to have every customer write a letter to Oracle. How is that going?
I think you missed the Kirk joke....and it is Allison...not Alli
RE: resorting to derogatory now?
"I am shocked Matt....." <Sigh> Did you really expect everyone to just blindly accept your musings?
"....1) I hear Power8 is baked but IBM does not want to give details because it has a lot more than just the typical cores/cache/threads/execution units as other chips are doing...." So, according to you, it's groundbreaking and different, so much so that IBM doesn't want to tell anyone it's groundbreaking and different. Yeah, and when was the last time any vendor missed an opportunity to sing their own praises? The silence from IBM only makes some of us wonder if Pee8 isn't as far down the road as IBMers make out, and whether IBM have started another rehash of the design.
".....5X per footprint increase compared to Tukwila's 2X ...." What, on those carefully crafted IBM benchmarks? Real World testing with real apps and data doesn't agree. If the IBM performance blurb actually played out in production then Itanium sales would have stopped rather than ramping up, and nobody would be buying anything but Power, which is very obviously not happening.
"....5) I hear Power8 will be 22nm just like Kittson but should show up in 2013 if IBM keeps on their 3 year cycle....." So you hear, but the rest of us don't hear any public statements from IBM. Until IBM actually put something out there there will be doubts as to what they can provide and when. You also included the fact that it all hinges on "if IBM keeps on their 3 year cycle" - IBM chips have slipped before, and if IBM is rehashing the Pee8 design then it's very likley to slip again. Not having a public roadmap with a date makes it easy for IBM to later say "we didn't slip, honest", but doesn't exactly inspire confidence.
".....6) No one in the industry doubts IBM commitment to Power...." Really? You personally asked everyone? Sorry, you missed at least one - me! Whilst I'm not saying IBM are definately not going to make a Pee8, it would be nice if they could make as public a commitment as Intel already have to Itanium. I already know what is coming with next generation Itanium, the fact that it will be socket-compatible making current new servers attractive, yet I'm left guessing by the lack of public statements from IBM regarding Power. Will it be another fork-lift Power upgrade? Will I need another AIX update to actually utilise all the Pee8's capabilities? Planning is a lot easier with some public roadmap info.
"....7) Itanium is what it is and I think will be killed by Intel...." A statement both lacking in any argument or logic. If you actually provided a reason why Intel would kill Itanium it might be worth the time to read, but all you have done is regurgitate the usual IBM FUD. Once again - Itanium keeps IBM on the back foot as it allows Intel to attack Power from above as well as below. Whilst x64 eats Power from below, Itanium threatens the IBM mainframe heartland. Please try and deny the whole PSI farce wasn't because Itanium nicely keeps the pressure on Power. IBM would love it if Intel killed Itanium, but that is exactly the reason why Intel won't.
"....I think you missed the Kirk joke...." Nope. You simply dsiplayed the same lack of Trekkie knowledge as you do server nouse - the beaming up was usually done by Scotty.
Die die die....
If it was any good it would have filtered down to the consumer market.
I don't get this notion that it is a server only architecture, this was the big next generation technology from Intel who thought 32-Bit x86 would die and Itanium would replace it.
But when it was released the Pentium 4 was faster and that was slower than the Pentium 3 at first. How on earth could Intel make such a rubbish CPU?
Re: Die die die.....
You're missing the point. Power, zArchitecture, and SPARC aren't in consumer parts either, but they're all arguably successful architectures. Itanium is a decent commercial-workload processor, currently hobbled by Intel's insistence on fabbing it on obsolete processes. Poulson should be competitive with Westmere-EX when it gets released, and depending on clock speed might be a bit faster.
GCOS and Itanium
You wrote : "Bull will still sell you an Itanium 2–based server or an upgrade to an existing machine. But none of these vendors launched machines last year that support the quad-core "Tukwila" Itanium 9300 processors."
You did not read the Press Release : We have launched the "novascale 9010" in May 2010. It is a GCOS 8 system that support "Tukwila". And in 2012, we will launch the next generation with Poulson.
"Intel hoped 32-Bit x86 would die and IA64 would replace it."
And they even managed to convince fools like Bonkers Bob Palmer at DEC, and his equivalent at Compaq, and the Board at Intel (who were desperate for some non-x86 success, and something to keep the x86 competition at bay).
How wrong could these folks have been? Well, given that Intel have hardly succeeded with anything other than x86, the runes should have been obvious from the start. But director did speak unto director, and the Alpha was subsequently canned way before its time, in favour of something which one day might deliver a bit more of what was promised so long ago. But even though IA64 was proclaimed to be the future for "industry standard 64bit computing", we know it never will be.
In parallel with all that board-level nonsense, AMD technologists showed that a compatible 64bit extension of the x86 architecture was not only possible, it was eminently practical, and even attractive to hardware and software people, and in general a much more sensible way forward than IA64.
Give it a year or three and IA64 will be history, just like the 4004, iAPX432, I2O, i860/i960, WiMax, and pretty much every other non-x86 product Intel have backed (they even sold off their ARM-based business, how mad was that?).