HP contemplating dumping Itanium

  • Thread starter Thread starter YKhan
  • Start date Start date
YKhan said:
http://www.theinquirer.net/?article=25573

x86 VMS, Alpha resurrection, etc., etc.

The question is: where will HP implement its Non-Stop architecture?
AFAIK, that's where HP's remaining investment in Itanium is. If HP
finds another way to implement Non-Stop, that's it for Itanium.

As to Alpha, I think Intel has all the engineers. Safe bet that,
whatever happens, it won't be a surprise to Intel.

Just the fact that these rumors are floating around is a killer.
However grounded in reality, someone much closer to enterprise
computing than I am told me recently that nobody (as in nobody among
enterprise customers) is buying HP's Itanium boxes.

It's a fair bet that, no matter what Intel says, it has been planning
on engineering into xeon all the enterprise features it has been
advertising for itanium. That makes x86 Non-stop sound much more
likely than a resurrection of Alpha.

RM
 
Robert said:
The question is: where will HP implement its Non-Stop architecture?
AFAIK, that's where HP's remaining investment in Itanium is. If HP
finds another way to implement Non-Stop, that's it for Itanium.

Well up until recently Non-Stop was running on MIPS processors. There's
lots of companies making processors based on the MIPS core, including
AMD, Broadcom, Altera, etc.; it's mainly for the embedded market of
course. But I'm sure somebody could be pursuaded to create a server
processor out of their embedded processor designs.

I think the first indication that HP might be reverting back to MIPS
would be if they decide to extend out the EOL and EOS deadlines for
their existing Non-Stop boxes. "Due to increased customer demand for
additional support of existing Non-Stop servers, HP announces the
/Nonstop Non-Stop Program/ to extend the life of your Non-Stop
servers." Or something like that.
As to Alpha, I think Intel has all the engineers. Safe bet that,
whatever happens, it won't be a surprise to Intel.

I can't see Alpha making a reappearance either. Intel owns it
completely now, and they'll find a way to kill it before they'd ever
resurrect it.

That's why I think there's now a rumour about x86 VMS. Of course, Intel
could conceivably get a piece of the action on x86 VMS too, but there's
no way that Intel could put a stop to an x86 port of VMS. In this case,
Intel's best course of action is to simply join em rather than trying
to beat em. And VMS might experience an increased userbase for the
first time in a long time, with so many cheap hardware choices.
Just the fact that these rumors are floating around is a killer.
However grounded in reality, someone much closer to enterprise
computing than I am told me recently that nobody (as in nobody among
enterprise customers) is buying HP's Itanium boxes.

For example, I don't know anybody with HP-UX on Itanium, everybody is
still on PA-RISC.
It's a fair bet that, no matter what Intel says, it has been planning
on engineering into xeon all the enterprise features it has been
advertising for itanium. That makes x86 Non-stop sound much more
likely than a resurrection of Alpha.

Well, I'm sure x86 is always a possibility for Non-Stop, but if they
keep their userbase at MIPS, the users will be much happier (no more
porting). I can imagine for example, that AMD takes one of their
Alchemy embedded MIPS cores and puts it into an Opteron package and
they can interchange MIPS and x86 over the same motherboards.

Yousuf Khan
 
YKhan said:
Well up until recently Non-Stop was running on MIPS processors. There's
lots of companies making processors based on the MIPS core, including
AMD, Broadcom, Altera, etc.; it's mainly for the embedded market of
course. But I'm sure somebody could be pursuaded to create a server
processor out of their embedded processor designs.

I think the first indication that HP might be reverting back to MIPS
would be if they decide to extend out the EOL and EOS deadlines for
their existing Non-Stop boxes. "Due to increased customer demand for
additional support of existing Non-Stop servers, HP announces the
/Nonstop Non-Stop Program/ to extend the life of your Non-Stop
servers." Or something like that.




I can't see Alpha making a reappearance either. Intel owns it
completely now, and they'll find a way to kill it before they'd ever
resurrect it.

That's why I think there's now a rumour about x86 VMS. Of course, Intel
could conceivably get a piece of the action on x86 VMS too, but there's
no way that Intel could put a stop to an x86 port of VMS. In this case,
Intel's best course of action is to simply join em rather than trying
to beat em. And VMS might experience an increased userbase for the
first time in a long time, with so many cheap hardware choices.




For example, I don't know anybody with HP-UX on Itanium, everybody is
still on PA-RISC.




Well, I'm sure x86 is always a possibility for Non-Stop, but if they
keep their userbase at MIPS, the users will be much happier (no more
porting). I can imagine for example, that AMD takes one of their
Alchemy embedded MIPS cores and puts it into an Opteron package and
they can interchange MIPS and x86 over the same motherboards.

Yousuf Khan
Didn't the development of server class MIPS processors end some time
ago, in favor of Itanium at SGI? And wasn't the whole point of the
Alphacide and PA-RISCcide to get HPQ out of the processor development
business? It seems really unlikely that Hurd, while in cost cutting
mode, would approve the reincarnation of HP processor development.

del
 
Well, SGI is not involved in this, MIPS is a seperate company now from
SGI. They are now in the business of licensing their core to other
companies to manufacture as embedded processors, just like ARM
Holdings. I'm not saying HP would want to manufacture their own MIPS
cores, but instead let other companies do it for them. There's lots of
companies making MIPS cores these days, including AMD & Broadcom.

Yousuf Khan
 
YKhan said:
Well, SGI is not involved in this, MIPS is a seperate company now from
SGI. They are now in the business of licensing their core to other
companies to manufacture as embedded processors, just like ARM
Holdings. I'm not saying HP would want to manufacture their own MIPS
cores, but instead let other companies do it for them. There's lots of
companies making MIPS cores these days, including AMD & Broadcom.

Yousuf Khan

And if you want a nonstop server or a VMS machine using the same
processor that is in a laser printer or a toaster then you are in great
shape. On the other hand if you want a server class mips processor,
that might be a little harder to come by.

Don't you understand the difference between embedded processors and the
sort of thing HPQ needs?

As I understand it, SGI retained all the MIPS server/workstation IP and
design team.
 
Didn't the development of server class MIPS processors end some time
ago, in favor of Itanium at SGI? And wasn't the whole point of the
Alphacide and PA-RISCcide to get HPQ out of the processor development
business? It seems really unlikely that Hurd, while in cost cutting
mode, would approve the reincarnation of HP processor development.

I'd rather strongly agree, I *REALLY* can't see HP pushing to develop
any sort of new MIPS-based Non-Stop server. The cores being made by
Broadcom, PMC Sierra, etc. are just not at all what would be needed to
replace existing chips. HP would have to spend quite a bit of money
on either designing the chip themselves or contracting out the design.
Definitely not any sort of off-the-shelf part.

Taking things a bit further, I would tend to say that it's the lack of
desire to spend money on developing hardware that is pushing HP away
from the Itanium. While they don't develop the processor anymore they
ARE spending a chunk of money to put together the rest of the server,
not to mention the software that runs on these servers. It wouldn't
surprise me if Mr. Hurd is taking a good hard look at their Itanium
servers and seeing a lot of expenditures without enough dollar returns
and trying to figure out how to get away from that business.
 
Del said:
And if you want a nonstop server or a VMS machine using the same
processor that is in a laser printer or a toaster then you are in great
shape. On the other hand if you want a server class mips processor,
that might be a little harder to come by.

Well, of course this doesn't include VMS, as VMS has never been run on
MIPS. This is just a specific suggestion for Non-Stop only.
Don't you understand the difference between embedded processors and the
sort of thing HPQ needs?

Well in this day and age, when they're selling Opterons for embedded
applications, apparently there isn't much difference anymore. :-)

I can't see why MIPS wouldn't be good enough to run server applications
anymore. It used to run them once, add some cache to it, put it on a
modern state-of-the-art production line, and it can probably run server
apps again. The important thing about Non-Stop isn't ultimate
performance anyways, just reliability.
As I understand it, SGI retained all the MIPS server/workstation IP and
design team.

Well, somebody was supplying MIPS chips to HP before for Non-Stop, it
couldn't have been SGI.

Yousuf Khan
 
Tony Hill wrote:

Taking things a bit further, I would tend to say that it's the lack of
desire to spend money on developing hardware that is pushing HP away
from the Itanium. While they don't develop the processor anymore they
ARE spending a chunk of money to put together the rest of the server,
not to mention the software that runs on these servers. It wouldn't
surprise me if Mr. Hurd is taking a good hard look at their Itanium
servers and seeing a lot of expenditures without enough dollar returns
and trying to figure out how to get away from that business.

What you're talking about here, though, is not HP abandoning Itanium,
but HP abandoning an entire line of business. PA-RISC is EOL. The
boards HP is making now can take either PA-RISC or Itanium. If HP is
going to be selling high processor count SMP boxes, they're going to
have to do the engineering for something, even if it's x86. Going to
x86 would mean they'd have to port VMS and HP-UX. The costs of moving
the software to Itanium are largely sunk costs.

There is a completely different explanation for the slowness of HP
Itanium sales: people are in no particular hurry to move to Itanium,
because Itanium technology is still on the steep part of the learning
curve. A presentation from HP says that processor utilization on Unix
boxes averages 17%; why would anybody be in a hurry to upgrade
anything?

HP's business model is a different story. If HP wants to remain a
head-to-head competitor with IBM up and down the line, they have to
continue making high processor count SMP boxes, even if it's
unprofitable.

I generally don't pay any attention to proprietary Unix wars, but
here's an article from last fall about the future of HP-UX

http://www.itjungle.com/tug/tug102804-story01.html

So, yet another way of asking the question would be whether HP will
port HP-UX to x86, and my suspicion is that the port is already
underway (along with a port of VMS). Porting HP-UX to x86 wouldn't be
a good sign for Itanium, but it's not quite the same as HP abandoning
it.

RM
 
Robert said:
I generally don't pay any attention to proprietary Unix wars, but
here's an article from last fall about the future of HP-UX

http://www.itjungle.com/tug/tug102804-story01.html

Good article. I remember the little spat between Sun and HP on that
one, but didn't see an article speculating about the possible existence
of an HP-UX for x64.
So, yet another way of asking the question would be whether HP will
port HP-UX to x86, and my suspicion is that the port is already
underway (along with a port of VMS). Porting HP-UX to x86 wouldn't be
a good sign for Itanium, but it's not quite the same as HP abandoning
it.

It wouldn't surprise me, HP kept denying it had any interest in Opteron
right up until it introduce a complete line of servers and workstations
based on it, and became the #1 Opteron OEM.

Yousuf Khan
 
Tony Hill wrote:



What you're talking about here, though, is not HP abandoning Itanium,
but HP abandoning an entire line of business.

Yup, that's about it. I'm not saying that it's a sure thing by any
means, but I wouldn't be surprised if it's crossed the mind of Mark
Hurd (and others) on at least one occasion.
PA-RISC is EOL. The
boards HP is making now can take either PA-RISC or Itanium. If HP is
going to be selling high processor count SMP boxes, they're going to
have to do the engineering for something, even if it's x86. Going to
x86 would mean they'd have to port VMS and HP-UX. The costs of moving
the software to Itanium are largely sunk costs.

There is still the question of maintaining and supporting the software
and hardware. If these systems aren't selling now and don't seem
likely to sell in the not-too-distant future then it's a waste of
money. Or perhaps it's a business best separated out from the rest of
the company in an effort to run more efficiently.

It certainly seems likely that HP's current plan for restructuring
focuses heavily on the balancing act they are currently doing between
the low-volume/high cost space where Itanium tends to live and the
high-volume/low-cost space where x86 tends to live.
There is a completely different explanation for the slowness of HP
Itanium sales: people are in no particular hurry to move to Itanium,
because Itanium technology is still on the steep part of the learning
curve. A presentation from HP says that processor utilization on Unix
boxes averages 17%; why would anybody be in a hurry to upgrade
anything?

And hence the reason why it just might not make sense for the company
to pursue this market aggressively.
HP's business model is a different story. If HP wants to remain a
head-to-head competitor with IBM up and down the line, they have to
continue making high processor count SMP boxes, even if it's
unprofitable.

Well, as hinted to above, right now HP is trying to be a direct
competitor to both IBM and Dell. The general consensus seems to be
that they are failing on both fronts.
 
Back
Top