Feeds

back to article Intel (finally) uncages Nehalem-EX beast

Intel's switch to the Nehalem architecture was finally completed Tuesday with the launch of the Nehalem-EX Xeon 6500 and 7500 processors, the last of the Core, Xeon, and Itanium chips to get the Quick Path Interconnect and a slew of features that make Intel chips compete head-to-head with alternatives from Advanced Micro Devices …

COMMENTS

This topic is closed for new posts.
Bronze badge

tough choice

$3,157 for an 8-core 1.86Ghz 95W chip from Intel for 4 socket systems

$1,165 for a 12-core 2.2Ghz 80W chip from AMD for 4 socket systems

tough choice...

2
1
Anonymous Coward

I think you mean...

$3.2k for 16 very fast threads with fast memory and turboboost

$1.2k for 12 slow threads with much less slow memory

1
2
Coat

nothing to see, plz move on

no one needs this stuff, if it does not have proper resilience / RAS features.

stuff like processor retry, failover, pci retry / failover, memory raid and so on.

what's the point of having all that performance if you dont dare to put more than say 10 production hosts on it although you have enough cores to host 32 of them ?

I do not know of any shop willing to do more than 10 prod hosts on intel, precisely because it is not reliable. / resilient.

replacing mainframes ? give me a break........

just my 2 cts.

ac

0
0
Anonymous Coward

Ah, the mainframe guy still trying to justify his existence!

It is the same as the TCP/IP principle - you know that the IP layer will have errors, so you compensate for them in the TCP layer. You know a 'pc' will not be resilient, so you cluster/grid/RAC/virtual farm 2 or more of them to compensate.

So you get the similar performance for an order of magnitude reduction in cost.

1
0
Thumb Down

Unacceptable RAS

Nehalem EX may have a "fancy feature" called Machine check architecture recovery but it just allows it to report problems to the O/S but that is about it.

When the blades catch fire doing simple java work I certainly would never put a mission critical database on them. Servers are now less than 10% of an IT budget so spend a little extra to stop the proliferation of small servers. Most mission critical processors have about 1/3rd of the processor chip dedicated to reliability.

Nehalem EX details that were not in the announce:

Processor

- Limited ECC, some Parity

- Limited MCA and requires reboot

- No live socket deactivation

- No live core deactivation

- Only some processors have ability to deallocate on reboot

- Bad data containment can only isolate to app or VM

I/O RAS

- No Hot Plug I/O

- No PCI-e extended CRC

- No Enhanced Error Handling/Recover

Infrastructure

- No redundant DC-DC conversion

- No hard partitioning

- No redundant system clocks

- No redundant I/O Fabric

- No redundant service processor

Memory

- No memory page de-allocation

- No double-chip spare

- No redundant address/control pins on dimm

As far as "Glue-less" to 8 sockets...count the QPI's. With only four QPI's the scaling past four sockets is horrendous. One QPI is for the I/O the other three are for the other three sockets in the four socket box. Anything past four is poor scaling unless you have a proprietary chipset.

Matt B

1
1

A.C

Nothing to see? Who are you trying to fool? You work at IBM, right?

4 socket Nehalem-EX does 723 specint_2006 which is quite respectable.

4 socket 8-core AMD Opteron 6136 does 571 specint_2006 which is ok. I think this machine with 128GB RAM costs 30.000 USD? I expect the 12-core version gives better performance. Not as much as 50%, but still a boost. If it gives 50% more performance, then it is almost 900 specint_2006.

4 socket POWER7 does 1010 specint_2006 which is best in class. But I suspect that machine costs 175.000 USD? What is the correct price? If the price is correct, you can buy six AMD Opteron servers for the same price. I bet these six AMD servers are faster than one POWER7 box? Jesper, how many AMD Magny Cour servers can you buy for the price of one such POWER7 server that does 1010 specint_2006?

.

When you go to many more sockets, 64 sockets and above, there is also Solaris if you need a good scaling operating system. You will not use Windows. Your only choice is a real Unix such as Solaris / FreeBSD. And of course, Linux. I suspect a 256 socket Nehalem-EX with Solaris/FreeBSD/Linux will beat the crap out of any 64 socket POWER7 machine. At a much lower price. Just buy a couple of these Nehalem-EX servers, and you have good redundancy too. I think the x86 cpus are getting really fast. That is good for Sun as Solaris runs on x86 as well.

I suspect in the future, we will see AIX on x86, because the x86 cpus are soon too good to not use, if this evolution continues. Who will buy an expensive Unix machine when it much much costs more, and have less performance? IBM has to port AIX to x86 or it will die. Solaris is already ported to x86, so Solaris is future proof, just recompile your apps for x86.

In the future, x86 will reign. Sure, the x86 train is buggy and squeaks, but it has a helluva speed and momentum. Any OS that does not support the cheap and faster x86 will die, because I dont see any incentive to buy a slower and several times more expensive system?

0
0
Thumb Down

sigh again again

Again your post is again full of various strange statements so it is kind of hard to know where to begin.

First on the spec_int part.

Well I'd rather wait with the SPEC_INT comparisons, Intel is reporting base numbers only. Which makes it kind of an invalid comparison, cause the specint numbers for Nehalem-EX could be significant better. So we'll just wait and see those ones.

As for

http://www.intel.com/performance/server/xeon_mp/summary.htm?iid=perf_server+mp_sum

Now what caught my eye is these benchmarks:

SPEC_JBB2005

Fujitsu PRIMEQUEST* 1800E 64 cores 8 sockets -> 3,321,826 BOBS/

IBM System x* 3850 X5 32 cores 4 sockets -> 2,012,730 BOBS.

This you have to compare to

IBM Power 750 Express 32 cores 4 sockets -> 2,478,929 BOBS

Which would mean that a 3.0 GHz power 750 would be faster, than the x3850. Now as for price then a half filled x3850 costs half that of a half filled power 750.

And SAP 2 Tier.

Fujitsu PRIMEQUEST* 1800E 64 cores 8 sockets -> 16.000 users.

IBM System x* 3850 X5 32 cores 4 sockets -> 10.450 users

This you have to compare to

IBM Power 750 Express 32 cores 4 sockets -> 15.600 users

So to match POWER7 you need a big boy 12U PrimeQuest 1800E, Now I don't have the price of that box, but other PrimeQuest boxes I have seen are in the same price range as a power 780.

But AC's comments on RAS features on Nehalem EX are relevant. These are the same reasons why people buy SD/MX000 and POWER 7XX boxes.

Again you get what you pay for.

Just admit it, lowest clocked POWER7 is as fast as the highest clocked Nehalem-EX and when the number of sockets grow.. then POWER7 is a sure winner.

And I don't know where you get your 256 socket x86 Solaris boxes from.. Well.. Sure SGI makes some but it's not like they support Solaris on that machine. Sure compile it yourself.. if Oracle hasn't cut support for something you need. As they are doing on other platforms right now.

And MC based boxes.. lets see some numbers rather than your mad ramblings.

// Jesper

1
0
Thumb Down

errr I thought we were talking about x64?

Jesper, one small point that makes the above comnparisons irrelevant......

PRIMEQUEST IS BASED on ITANIUM!!!!!!!!!!!

0
0
Anonymous Coward

Primequest 1800E

... is Xeon-based. Go ahead. Google it.

0
0
Pint

Try to..

check your facts before typing in bold :)=

Beer cause it helps on allergies.

// jesper

0
0
FAIL

Kebabbert, had a look at VMware's price list lately ?

an 8 core power6 blade with 8 GB is list price around 17K usd.

an 8 core intelblade is perhaps 10K + 60K vmware licenses the next 4 years.

so what is expensive ? 17K or 70K ?

The power blade does have all the stuff you need to run it reliably.

it can run eqasily 10-60 VM's without flinching.

yes. you read that : it says 10-60. people do it every day

so if you need reliability and scalability, it is still something better than intel what you need.

and it's cheap, on top of that. and linux on it is for free. really.

ac

0
0
Coat

ah but...

it still cannot play Crysis at a decent frame rate.

0
0
Anonymous Coward

Did Matt forget to log in?

A post above (AC 07:45) has the signature "Matt B" at the end.

Matt Bryant, perhaps, judging from the 'sceptical' content?

Matt, if that is you, how many of the features you list does today's Itanic have?

Mind you the comment about the number of HyperTransports is perfectly valid.

0
0
Silver badge
FAIL

RE: Did Matt forget to log in?

Dear Sunshiner,

Why is it you poor, deluded fools feel the need to pretend you are someone else when posting? I know you spend many an hour pretending to have technical knowledge, or that the Sunset didn't happen or that Rock isn't dead (is it just sleeping?), so maybe it's just habitual. Anyway, you forgot to put on the SP&L ("still pointing and laughing") - you're as poor at forgeries as you are at IT!

".....Matt Bryant, perhaps, judging from the 'sceptical' content?...." Now, why would I be sceptical about Nehalem EX? If you had bothered to read my previous posts you might have realised I am looking forward to the uber Nehalem.

"....Matt, if that is you, how many of the features you list does today's Itanic have?...." Whilst I can see why TPM's attempt to turn the EX launch into an Itanium attack piece would appeal to you, the reality is Nehalem EX is going to take big chunks out of ALL the UNIX lines - IBM's Power, Fudgeitso's SPARC64 and Soreacle's Niagara. The difference in effect on those vendors is that it will be predominantly eating the UNIX market from the bottom up, so Niagara will be hit hardest, followed by SPARC64, then Power, then Integrity. You may disagree (without any logical argument to support your rabid rantings), but the recent Gartner market figures showed that Sun is selling nothing but low-end, pSeries seems to be spread across the midrange, and hp are dragging in the top-end deals (check the figures in the thread http://forums.theregister.co.uk/forum/1/2010/02/24/gartner_q4_2009_servers/).

Those low-end systems are going to be at threat from the standard Nehalem EX boxes that will be coming out and scaling to eight sockets, same goes for a chunk of the mid-range, but the high-end should be a lot safer (unless we see a tier 1 vendor making a 16-socket Nehalem EX box). So, at the monemt, it looks like Itanium in hp Integrity has the least to fear from Nehalem EX, and seeing as hp is the leading x86 vendor it is also likely that hp will also be the vendor to gain most from any eating of the UNIX low-end. And seeing as I am all for hp ProLiant and Integrity (in my opinion, they give me the best performance at the best price with the best management tools and support), that means I'm actually quite happy with the idea of more scaleable Xeon boxes. And I'm also sure pressure from AMD will keep the prices of those scaleable x86 boxes down too!

Oops, did I not sound sceptical then? Sorry!

/SP&L

0
1
Megaphone

unacceptable RAS

Those "problems" with Nehalem EX are from HP and used to compare Tukwila to Nehalem EX.

Just written in check box / no check format powerpoint.

Which is why it takes two x86 boxes in a software cluster to equal the RAS of one RISC/EPIC box.

0
0

A.C

Lets face it. x86 servers are here to finally challenge Unix servers. Sure, x86 lacks RAS, but it has lots of performance, and is CHEAP. In the future, x86 will be even faster. You must be able to offer your OS on x86

0
0
Anonymous Coward

That says it all

As anyone working with larger systems/environments know, acquisition cost isn't everything, RAS features as the ones in POWER7 (even Power 5-6) are very important when it comes down to consolidating and virtualising your environment. Downtime costs a lot of money and not all applications are suitable to scale out for redundancy.

When you say "In the future, x86 will be even faster".... No shit Sherlock, everything will be faster in the future . POWER7 is the fastest chip, fastest socket and the fastest systems. I bet next generation of Power will be even faster... AND they have the RAS features to make it possible to have extremely high uptime. AIX shows highest uptime of all operating systems in several reports.

0
0

This post has been deleted by its author

Boffin

SPECint

Isn't really what you need to test a highly parallel, shared memory system - at least for HPC. What you really need is to strong-arm those systems with some communication heavy CFD code running MPI or OpenMP. Something that requires each of your processors to share a lot of data and often.

In the past Opterons scaled better in this regard, even with Intel's (allegedly) AMD-crippling Fortran compilers, even in clusters - bearing in mind the likes of MPI often used shared memory on local nodes.

It'll be interesting to see how the new Xeon systems shape up; there's a strong market out there for 32-64 core HPC boxes.

0
0

Jesper

"Just admit it, lowest clocked POWER7 is as fast as the highest clocked Nehalem-EX and when the number of sockets grow.. then POWER7 is a sure winner."

Yes I admit that POWER7 is faster, I wrote it is "best in class". And I also wrote that the four socket AMD Magny Cour 8-core which costs 30.000 USD does 571 specint. I also wrote that the 12-core Magny Cour should offer higher performance. And I wrote that 4 socket POWER7 does 1010 specint, which is best. Didn't I?

I fail to see why the POWER7 is a sure winner when the number of sockets grows? Do you mean that the biggest POWER7 servers will be able to outperform an 256 socket Nehalem-EX? How do you know that? Tell us!

.

You also say that "to match POWER7 you need a big boy 12U PrimeQuest 1800E, Now I don't have the price of that box, but other PrimeQuest boxes I have seen are in the same price range as a power 780" and therefore Nehalem-EX will not be able to compete with POWER7. Because Nehalem-EX is as expensive as POWER7 when you compare them performancewise. Well, if that is true, then you have a point and then I am wrong. Let us wait and see the final performance numbers and final prices before we try to agree on a conclusion.

.

Regarding RAS on Nehalem-EX and uptime talk. I see your point, which is a valid one. At the same time, several Stock Exchanges such as NASDAQ runs on Linux on x86. If x86 suffices for NASDAQ big stock exchange system, then I think that the RAS of traditional UNIX cpus are not really necessary? Hence, the RAS of POWER7 and SPARC and Itanium is not really needed anymore. Hence, RAS of Nehalem-EX will be good enough. And performance of Nehalem-EX will be good enough to challenge POWER7, at a much lower price. (Jesper Frimann means that prices will be equal - let us watch and see if he is right or if he is just guessing)

0
0
Anonymous Coward

NASDAQ

NASDAQ dumped HP NonStop (Tandem) for rackable systems and has built an incredible amount of redundancy to reduce not avoid outages.

Unfortunately when Rackable bought SGI the SGI salesforce prevailed and the rep on NASDAQ is clueless....just goes to show you fast/simple can be expensive and complicated.

0
0
Gates Horns

Kebubski

"I fail to see why the POWER7 is a sure winner when the number of sockets grows? Do you mean that the biggest POWER7 servers will be able to outperform an 256 socket Nehalem-EX? How do you know that? Tell us!"

Cause POWER7 scales much better than Nehalem EX.

Again from:

http://www.intel.com/performance/server/xeon_mp/summary.htm?iid=perf_server+mp_sum

Nehalem EX based server.

8 sockets -> 16,000 SAP users -> 2000 users per socket.

4 sockets -> 10,450 SAP users -> 2613 users per socket

That is not good scalability.

And if you look at specint_rate2006. then POWER7 in the power 780 has a scalability factor of 97%, on specint_rate2006. and 96% on specfp_rate2006 going from 2-8 sockets. That is not good, that is damn good.

Now for Nehalem-EX the numbers are 86% on specint_rate2006, going from 2-8 sockets, that is decent. But again these number are not based upon the same box, and are only base numbers. So they can only be viewed as a preliminary result. (to be read useless)

But they still indicate a scalability of Nehalem that isn't really at the same level as POWER7, hence the more sockets the more the advantage of POWER7.

Well Nyse went from mainframe to Windows on x86 BIG CRASH cost them millions of millions and now the are going for Linux on x86 I gues. It would most likely have been much cheaper to have stayed on the mainframe. Or like Nasdaq stayed on Tandem.

And as for the RAS features.. I think you are very naive. But to your defense, so are an alarming number of people in topish management positions, they don't really consider how important their IT systems are today.

What do you think that a crash that puts a company's financial system out of action for lets say 4 days will do to a firm, let say alone a week or 2 weeks.

Generally companies today cannot function without IT. If their IT systems are down. Companies today literally 'live and die' with their IT systems.

// Jesper

0
0

Jesper Frimann

"...But they still indicate a scalability of Nehalem that isn't really at the same level as POWER7, hence the more sockets the more the advantage of POWER7..." Yes I agree that POWER7 scales better. I doubt 64 socket POWER7 will be faster than 256 socket Nehalem-EX. I am not so sure, as you are on this.

.

"...Well Nyse went from mainframe to Windows on x86 BIG CRASH cost them millions of millions and now the are going for Linux on x86 I gues...." No, this is not true. NYSE migrated from HP-UX to Linux. However, London Stock Exchange migrated to Windows, with lots of problems. Now LSE is migrating to Linux + Solaris. I told you, I work in finance.

See, it is possible to correct someone else, without insulting him. Maybe you should try it yourself next time?

.

"...And as for the RAS features.. I think you are very naive...." Maybe. What I am trying to say is that x86 suffices for NASDAQs big stock exchange system, so indirectly you call NASDAQ's executives naive. Maybe you should steer NASDAQ's business up which handles billions of USD each day, then?

I think that if x86 suffices for NASDAQ, then x86 suffices for most other companies, yes? And Nehalem-EX has improved RAS, so Nehalem-EX should be even better, yes?

0
0
Linux

Hmm...

"Yes I agree that POWER7 scales better. I doubt 64 socket POWER7 will be faster than 256 socket Nehalem-EX. I am not so sure, as you are on this."

Well I didn't really say that a 256 socket Nehalem machine won't be faster than a 64 sockets POWER7 machine.

I don't really know any 256 socket Nehalem machine, besides perhaps SGI will most likely do one. But that isn't really a machine made for commercial workloads, but rather HPC.

There isn't any 64 socket POWER7 machine, and from what I've seen there will be a 32 socket machine with 256 cores. For HPC you might be able to connect an unspecified number of power 775's together, the same way as SGI does.

On specint_rate2006, which is embarrassingly parallel, I think a 32 socket POWER7 will be about equal to a 64 socket Nehalem box and it will beat it on specfp_rate2006.

On benchmarks like TPC-C and SAP 3-Tier and perhaps some of the heavy app server benchmarks, the story is different, quite different.

But again the highend x86 marked is tiny compared to RISC/EPIC/CICS.

"However, London Stock Exchange migrated to Windows, with lots of problems. Now LSE is migrating to Linux + Solaris. I told you, I work in finance."

You are correct, I stand corrected.

But it still doesn't change the fact that they most likely would have been better off keeping their mainframes. If it ain't broke don't fix it.

And taking a Big Iron UNIX server/Mainframe/HP3000/VMS whatever where the business logic is written in Cobol/PL1/C/C++ and port it to a .NET or Java stack on a distributed solution is.. (IMHO absolutely bonkers) a huge project. And one that often put the very business at risk. Seen it several times. But often it isn't something that gets put in the news.. you only learn about it through your network.

Big Migration projects always have impacts on business, and often this is quite a big impact.

"Maybe. What I am trying to say is that x86 suffices for NASDAQs big stock exchange system, so indirectly you call NASDAQ's executives naive. Maybe you should steer NASDAQ's business up which handles billions of USD each day, then?"

NASDAQ is in the US. I am in a tiny country called Denmark, just south of Sweden you know :)=

And having been in the business for quite some time, I've been involved in some way with many of the big Companies in the Nordic.

And to quote one of my favorite films "I've seen things you people wouldn't believe."

I've seen companies break their backs and go down and being sold due to IT problems.

Now there is nothing wrong with LINUX, I love it. Even my wife who is a nurse uses it on her computer, as do I on the machine I am using here.

And sure you can make a big centralized solution using Linux and _good_ high quality x86 iron. I have no problem with that. Will it be just as stable as a solution on POWER? Not likely but it will be damn good. 99.9% availability ? Sure no problem.

But... you have to have the right people, the right documentation level, the right ... and and and...

What I really talk out against is the "we just buy a bunch of the cheapest blades and .. and and.."Windows and NAS and spaghetti and lots of cheap people and and and .. "

That doesn't work, and that is what many companies are doing, and paying a very high price for.

As I understand it Nyse had their trading system running on Mainframe and perhaps HPUX as front ends. Today that is AFAIK AIX on POWER + Linux on x86 for ftp. But that is still big centralized system.

With regards to LSE.

It really has to be seen with regards to LSE if their new system is better than the old.

And if you sum up what LSE have lost/spend of extra money on their move away from Mainframe... this article gives you just a little clue:

http://www.computerworlduk.com/technology/networking/messaging/news/index.cfm?newsid=18429

Original cost + 40M£ to Accenture + 20M£ writeoff this year.. + lost sales + ... + ... makes what 100M£++ It would have been cheaper to stay on the Mainframe.

// Jesper

0
0

Jesper Frimann

Regarding NYSE stock exchange, as someone writes

"The IT people at the NYSE lose their entire yearly bonus if their uptime drops to less than 99.99%. They use Linux, but it took a 15 year migration project to get off of HPUX. Even so they use way too much hardware and alarms go off if any machines have a load average of more than 0.1. They do not believe in putting any kind of a load on their machines, they are afraid of performance slowdowns. They know full well that Linux does not behave well under load."

So what I have heard NYSE is not using IBM AIX. Traditionally, Solaris have been heavily used in Finance and Telecom. Now many finance companies migrate to Linux. I have never heard of a single stock exchange that use IBM AIX, actually. If you could provide links about exchanges running AIX and POWER I would read them with great interest.

0
1

Here's a link

"The IT people at the NYSE lose their ..."

Well I don't know where you have that from and it might very well be right. But usually Big Places run many different OS'es, on many different Hardware platforms.

"So what I have heard NYSE is not using IBM AIX. Traditionally, Solaris have been heavily used in Finance and Telecom. Now many finance companies migrate to Linux. I have never heard of a single stock exchange that use IBM AIX, actually. If you could provide links about exchanges running AIX and POWER I would read them with great interest."

http://searchdatacenter.techtarget.com/news/article/0,289142,sid80_gci1254860,00.html

// Jesper

0
0

Chip Select

Does anyone know the Chip Select for the Nehalem EX (number of ranks each memory channel can work with). If the number is still 8 as with Nehalem and Westmere EP, then even with 4 channels, the max number of quad rank DIMMs each socket can have is 8 even though each socket has total of 16 DIMM slots. So the 16 slots turn out to be just marketing.

Also, do I understand that the max memory throughput is 1066MT/sec compared to 1333MT/s for Nehalem?

0
0
Anonymous Coward

Well.

Isn't it better to build a trading system with fewer components than to cluster it with hords of cheap intel boxes? Less admin, less space, less energy etc.

By the way, trading systems always have more service windows compared to larger backend systems such as bank applications and systems for manufacturing.

Thats why larger systems such as mainframes and big-iron unix boxes still have a large market.

0
0

A.C

"Isn't it better to build a trading system with fewer components than to cluster it with hords of cheap intel boxes? Less admin, less space, less energy etc."

Probably not, as NASDAQ is using cheap x86 with Linux for their stock exhange system. I think x86 is good enough. It is all about price. Obviously, Unix servers would suffice for NASDAQ's needs, but now they have chosen x86.

Ergo, x86 is good enough, when we talk about RAS. NASDAQ's system handles billions of dollars each day. Nehalem-EX also has increased RAS. That would be a good choice for NASDAQ.

Buy three cheap servers for the price of one expensive server. Then you have good redundancy, if the OS can handle it. I have heard that IBM AIX have much of the redundancy in hardware (IBM brags about their machine offers redundancy no matter which OS you run), whereas Solaris has much redundancy in software.

0
0
Pint

Re RAS (@Matt)

Hello Matt, AC0908 from Wednesday here again.

Why do you assume that anyone questioning is a Sunshiner? I'm not.

But nor do I approve of all the ridiculous "you need IA64 for decent RAS" rubbish that's been put around for years, and as this article and your lengthy but unsupported "shopping list" comment shows, continues to be put out.

As you well know yourself, and indeed have said here previously, AMD64 on Proliant does very nicely for hardware RAS thank you. In fact now that IA64 systems and x86-64 systems are both Hypertransport based [1] there can't be much difference can there.

So given that, all Compaq needs to make the most of its AMD64/Proliant hardware is decent software - HP-UX, NonStop, VMS, on a small enterprise-class selection of Proliants. There'd still be a place for IA64 for a year or two, until AMD64 catches up with current IA64 massive-memory massive-SMP capabilities.

You know it makes sense. So do HP and Intel. But so much personal prestige at HP HQ and Intel HQ depends on it *not* happening, so we may have to wait till certain individuals "pursue other business opportunities".

[1] Yes I know Hypertransport and CSI/Quickpath aren't the same really, just like AMD64 and Intel's copy of it aren't the same. But if Itanic and Nehalem are both using the same interconnect they by definition have pretty similar hardware RAS capabilities, regardless.

0
0
Silver badge
WTF?

Re RAS (@Matt)

".....Why do you assume that anyone questioning is a Sunshiner? I'm not....." But you have all the syptoms - inability to read a post and actually comprehend the content; no knowledge of how enterprise systems are used; no understanding of the business cycle; complete failure to actually do some research before posting; ability to contradict yourself in the space of three lines.....

"....As you well know yourself, and indeed have said here previously, AMD64 on Proliant does very nicely for hardware RAS thank you...." If you say I have said it before, why are you accusing me of saying "you need IA64 for decent RAS"!?!?!?

".....In fact now that IA64 systems and x86-64 systems are both Hypertransport based [1] there can't be much difference can there....." Except the one big difference you forgot - scale. Even Nehalem EX does not scale to the same extent as Power, Itanium or SPARC64 (though it already scales past Niagara). Whilst x64 is great for scale out, it cannot currently scale up as much as those enterprise UNIX chips with the same level of vendor support. Whilst there are ways to make it scale (after all, Unisys has been doing it for years), they are not mainstream, are expensive, and do not have 64-bit WIndoze or Linux apps to match. Until the whole x64 stack - hardware, OS and commercial apps - can scale to match Power, Itanium and SPARC64 there will carry on being a market for those CPUs in large SMP servers.

"....There'd still be a place for IA64 for a year or two...." <Sigh> Just give up now, you're not doing anything to help your cause. It will take a lot longer for the OS and commercial apps to appear for really large scale x64 instances. And then you will have to wean us customers off our SMP UNIX paltforms that we all like and trust. Believe me, I've done the research and the POCs, and we could strip out about 98% of our current UNIX and WIndoze servers and cobble together solutions using Lintel and OSS apps, but there is no way the business would get the same level of support which means risk, and the business is very adverse to risk. So we don't. At the current rate of development I'd say large SMP UNIX is probably safe for a good ten years if not more. And that's depsite my being a Linux fanboi. Should UNIX development stay ahead and we see a jump to 128-bit UNIX OS and CPUs then all bets are off and x64 is back to playing catch up again (though whether IBM, Intel or Fudgeitso have the will to make the jump to 128-bit is questionable).

"....But so much personal prestige at HP HQ and Intel HQ depends on it *not* happening....." Where do you get this stuff, from the bottom of a w(h)ine bottle? As it stands, Intel is the leading x64 CPU manufacturer, they would like nothing more than to have everyone switch to x64 because they would make MORE MONEY. And hp is the leading x64 vendor, so same goes for them, as it would seriously damage their opposition. What you fail to see is that hp has very cleverly balanced the two to meet the spread in market demand. Sun killed itself pitting UNIX against x64, whereas hp are happy to offer both. Whilst hp would be very happy to take the extra margin they make on all the attach on Integrity, they have realised that they need to be strong in x64 to stay strong in the UNIX field. What you mistake for "personal prestige" is just actually shrewd product management. When Integrity no longer makes money for hp then it will be retired, and likely enough hp-ux along with it as hp probably won't be as stupid as to go head-on with Linux and WIndoze like Sun tried (and failed) to do.

/SP&L

0
1

Microsoft just dumped Itanium

First it was RedHat now its HP's "friend" Microsoft.

http://www.microsoft.com/windowsserver2008/en/us/2008-IA.aspx

0
0
Silver badge
FAIL

RE: Microsoft just dumped Itanium

Still looking hard for the link to "Micorosft announces support for Windows on Power"..... Maybe you could help me, Mrs Kebabfart, you seem to be very close to the IBM FUD department? OH, you mean it has never been supported on Power? Why, by your argument, what a complete failure Power7 must be!

/SP&L

0
1
Boffin

Actually it was once..

http://en.wikipedia.org/wiki/PowerPC

Windows NT 3.51, the last good windows version.

And Solaris also btw.

// Jesper

0
0
This topic is closed for new posts.