Sun has to switch to Itanium?

[Yousuf Khan]

Not offering any opinions myself, just relaying an article:

http://www.osnews.com/story.php?news_id=6757

Yousuf Khan

 

[Robert Myers]

Not offering any opinions myself, just relaying an article:

http://www.osnews.com/story.php?news_id=6757

Sounds like a pretty reasonable move (if you're, uh, not living in an
AMD/Opteron/x86 echo chamber). SGI tried to sell its boxes as servers
and couldn't. SGI has the box, Sun has the customers. Intel might
even find some sweeteners to help the deal if Sun moves before it's a
charity case.

RM

 

[Nick Maclaren]

Sounds like a pretty reasonable move (if you're, uh, not living in an
AMD/Opteron/x86 echo chamber). SGI tried to sell its boxes as servers
and couldn't. SGI has the box, Sun has the customers. Intel might
even find some sweeteners to help the deal if Sun moves before it's a
charity case.

That is very much oversimplifying the case - anyway, the technical
advantages of the Altix have very little to do with the Itanium
and a lot to do with memory management. If Sun bought SGI, it would
make more sense for any new range of Altix to use Opteron in it.

But I doubt the technical issues are the ones that dominate in this
case.


Regards,
Nick Maclaren.

 

[Yousuf Khan]

Sounds like a pretty reasonable move (if you're, uh, not living in an
AMD/Opteron/x86 echo chamber). SGI tried to sell its boxes as servers
and couldn't. SGI has the box, Sun has the customers. Intel might
even find some sweeteners to help the deal if Sun moves before it's a
charity case.

But will Intel be able to sweeten a deal enough without requiring Sun to
stop selling Opterons? If Sun had to stop selling Opterons then it would
really be the end of the line for them.

Anyways, this guy's hypothesis is out of left-field. Why would Sun buy SGI?
Sun spent a good deal of time back in the 90's destroying SGI to the point
of where it is now.

Yousuf Khan

 

[Yousuf Khan]

That is very much oversimplifying the case - anyway, the technical
advantages of the Altix have very little to do with the Itanium
and a lot to do with memory management. If Sun bought SGI, it would
make more sense for any new range of Altix to use Opteron in it.

If that's the case, then why wouldn't SGI just start selling Opteron Altixes
themselves? They don't need a buyout from Sun to do something like that.

Yousuf Khan

 

[Robert Myers]

But will Intel be able to sweeten a deal enough without requiring Sun to
stop selling Opterons? If Sun had to stop selling Opterons then it would
really be the end of the line for them.

Respond to you? Respond to Nick? Respond to an entire mob of
club-swinging, stone-throwing AMD salesmen (one of whom would surely
like to take another swat at labelling me a kook)?

You've all just been out in the sun too long. Money, money, money,
money, money. Intel's got it, AMD doesn't. HP, IBM, SGI, and Dell
will all be selling high-end Itanium boxes and Sun and Pathmark (or is
that the name of a Drug Store? I get confused so easily) will be
selling Optera.

Were I Intel, I'd say, "Sure, you go right ahead and sell them
Opterons. How much money you got in your ad budget, anyway? You want
us to pay for Itanium ads so you can buy Opteron ads? Nah, don't
think so. We'll pay for the Itanium ads. You keep right on selling
them Opterons for, um, continuity. Just don't spend any or _our_
advertising budget on them, okay?"

Anyways, this guy's hypothesis is out of left-field.

Yeah, it is out of left field. Still makes sense. Remember, VMS, an
operating system for real men, will be running on Itanium. Wouldn't
Sun like to go back to the days when people thought of Solaris almost
the same way?

Why would Sun buy SGI?
Sun spent a good deal of time back in the 90's destroying SGI to the point
of where it is now.

Why would Sun settle with Microsoft?

RM

 

[Yousuf Khan]

Respond to you? Respond to Nick? Respond to an entire mob of
club-swinging, stone-throwing AMD salesmen (one of whom would surely
like to take another swat at labelling me a kook)?

Well there is good reason to keep selling Optera. Namely price &
performance. You get to sell a lot of software with Optera because you'll be
selling a lot of units of it.

You've all just been out in the sun too long. Money, money, money,
money, money. Intel's got it, AMD doesn't. HP, IBM, SGI, and Dell
will all be selling high-end Itanium boxes and Sun and Pathmark (or is
that the name of a Drug Store? I get confused so easily) will be
selling Optera.

So far all of those Itanium sellers haven't made much of an impression with
their Itanium sales. Even with all of Intel's money. I guess Intel simply
doesn't have enough money to make the Itania completely free.

Were I Intel, I'd say, "Sure, you go right ahead and sell them
Opterons. How much money you got in your ad budget, anyway? You want
us to pay for Itanium ads so you can buy Opteron ads? Nah, don't
think so. We'll pay for the Itanium ads. You keep right on selling
them Opterons for, um, continuity. Just don't spend any or _our_
advertising budget on them, okay?"

They could try that, but as we see with IBM and HP, they keep right on
selling Opteron boxes, but they advertise a lot of Xeon boxes.

Why would Sun settle with Microsoft?

Well, in general, when you start a lawsuit, the expectation is that you will
settle it eventually, in or out of court, it doesn't matter, but the end
result is always a settlement.

Yousuf Khan

 

[Tony Hill]

Not offering any opinions myself, just relaying an article:

http://www.osnews.com/story.php?news_id=6757

I'm not really makes any sense at all, he's totally ignoring the
customers that Sun sells to!

Just how, exactly, would the Itanium fit into any of Sun's line-up and
how is it even remotely better than what they've got planned? Sun
sells to two main markets, their "Network Facing" servers, ie database
servers, web servers and the like, and their "Data Facing"
workstations. Which of these would the Itanium fit into?

One thing to remember before we begin: Going forward over the next
5-10 years, probably the most important aspect of processor design is
going to be performance/watt. Previously power consumption was low
enough that it could generally be ignored for servers, that is no
longer the case.

First off, for their servers they are looking for throughput. High
single threaded performance is pretty pointless as compared to high
multithreaded performance. Sun is very well positioned in this market
because of their software, Solaris is probably the best operating
system on the planet when it comes to scalability (with the possible
exception of AIX). However the hardware side of things helps Sun out
here as well. Good I/O is very important for performance in this
market while raw number crunching isn't such a big deal.

So, what solution does Sun have? A little chip called "Niagara". 8
cores, each capable of executing 4 threads simultaneously. It's
single threaded performance won't even be half that of the Itanium,
but for multithreaded performance it will SMOKE anything Intel has in
the pipeline before "Tanglewood" in 2007/08, and it will do so with
about half the power consumption (or less) and a smaller die.

As for the workstation market, here Sun is dead in the water, but
Itanium won't save them. x86 is going to steamroll over all other
workstation architectures. It's already BY FAR the dominant
workstation platform, and the only reason why x86 hasn't totally taken
over the market is that it wasn't 64-bit. That, of course, has
changed with the Opteron (and soon with the Xeon as well). The
Itanium had it's chance to make an impact on the workstation world
over the past few years, and it failed miserably. Now that 64-bit x86
is a reality, NOBODY is interested in much else. Even if an Itanium
is a little faster, who is going to spend $25,000 for an Itanium
workstation when they can get 95% of the performance from an Opteron
or Xeon system costing only $5000? It's even worse for the large
chunk of workstation tasks that focus on integer performance where the
Opteron is easily faster than the Itanium.

Sun does still have a market for SPARC workstations to support legacy
applications, but the Itanium is obviously no help here either.


Really I'm not sure that there is much of a market for Itanium except
as a replacement for PA-RISC and Alpha, neither of which were hugely
successful chips (in terms of sales, both had some pretty decent
performance). Itanium is a pretty good chip for HPC work, but that's
not really a market Sun competes in anymore. I suppose by buying SGI
and selling their Altix systems, Sun could get back into this market.
However given that SGI is bleeding even more red ink than Sun is, I'm
not sure that would be a very good idea. Besides, IBM looks to be
STRONGLY pointing the way forward in HPC with their Blue Gene/L
systems. Those who are familiar with this technology will know that
it bares a resemblance to the Niagara chip mentioned above in that
Blue Gene/L consists of many simple cores on a single chip rather than
one complicated core.

Now, things aren't entirely too bleak for Itanium. The idea of
putting many cores on a single chip is, at least in theory, VERY well
suited to the VLIW design that the Itanium purports to follow. VLIW
chips should have a very simple and low powered core with a small die
that gets quite a bit of performance by offloading a lot of work to
software. The problem here is that somewhere along the way the
Itanium seems to have become just totally buggered, but it is neither
simple or low powered and the die is huge (admittedly it's mostly
cache). However if Intel can get these aspects in line and get back
to the original theory of VLIW chips, putting a whole bunch of
Itaniums cores on a single chip could have a LOT of potential. This
is exactly what Intel's "Tanglewood" chip is all about, though the
shipping date of that chip has been rather fuzzy.

Still, like I've hinted at above, it'll have to make it to market and
get the software support behind it before the window of opportunity is
gone. Sun's kind of in the same boat with Niagara, though at least
they have an existing market to sell to, so that gives them more time.
Intel (and customers) have to build their market for Itanium products,
which is MUCH more difficult.

I've mentioned a few things about this before, but I'll mentioned them
again:

1. I think Sun made the right choice in killing off the UltraSparc V.
It wasn't going to be enough to save their workstation market from x86
and it wasn't going to give them the "throughput computing" that
Niagara will. The only dumb thing Sun did here was that they waited
too long, they SHOULD have killed the UltraSparc V back in 2001 or
thereabouts.

2. There are only three ISAs that I see having a really positive
future going forwards: x86, PowerPC and ARM (only counting higher-end
stuff here, Motorola 68xx and PIC chips will still outsell everything
else on the planet by an order of magnitude or two), Itanium looks
like it will replace PA-RISC as HP's primary server chip as well as
replacing Alpha as the choice for raw number crunching HPC stuff, but
neither of those markets look super promising. Large *nix servers are
continuously being replaced by smaller/cheaper x86 systems running
Windows or Linux. HPC is moving more towards clusters of COTS stuff,
or alternatively to stuff like IBM's Blue Gene/L. SPARC has an
existing market to continue support, but again that's getting eaten up
by x86 in a fairly large way.


Anyway, not only do I not really think that Sun "has" to switch to
Itanium, I'm not even sure that it would achieve anything. Maybe when
"Tanglewood" gets here, but that's still a bit of an unknown.

 

[Anil T Maliyekke]

If that's the case, then why wouldn't SGI just start selling Opteron Altixes
themselves? They don't need a buyout from Sun to do something like that.

Yousuf Khan

SGI's market is technical computing. Itanium is a good fit.

Anil

 

[Nick Maclaren]

|> |>
|> > That is very much oversimplifying the case - anyway, the technical
|> > advantages of the Altix have very little to do with the Itanium
|> > and a lot to do with memory management. If Sun bought SGI, it would
|> > make more sense for any new range of Altix to use Opteron in it.
|>
|> If that's the case, then why wouldn't SGI just start selling Opteron Altixes
|> themselves? They don't need a buyout from Sun to do something like that.

I have difficulty in believing that you are naive enough to ask that
question seriously. But, just in case you are:

1) Money. SGI is very short of it. Investing in a new chip
takes money to develop and test the hardware and software.

2) Contracts. Some of those with Intel may be exclusive, at
least for now. SGI signed the deal with Intel when Intel had the
whip hand.

3) Commercial politics. Intel bears serious grudge, and SGI
can't afford to be on the receiving end from Intel.

Only (2) would remain valid if Sun - or, indeed, pretty well any
other company - took over SGI.


Regards,
Nick Maclaren.

 

[George Macdonald]

Not offering any opinions myself, just relaying an article:

http://www.osnews.com/story.php?news_id=6757

Hmmm, I didn't see any mention of Fujitsu in that article. It was my
impression that would be Sun's fallback for the Sparc market systems. IOW
it's not the end of the world for Sun though times are likely to be hard
for a bit. In the Opteron space, I expect them to (eventually) come up
with something which sets them apart, at least as a total system, from all
the others and protects their current customer base.

Rgds, George Macdonald

"Just because they're paranoid doesn't mean you're not psychotic" - Who, me??

 

[David Svensson]

I'm not really makes any sense at all, he's totally ignoring the
customers that Sun sells to!

It does not make sense, but for other reasons...

Just how, exactly, would the Itanium fit into any of Sun's line-up and
how is it even remotely better than what they've got planned? Sun
sells to two main markets, their "Network Facing" servers, ie database
servers, web servers and the like, and their "Data Facing"
workstations. Which of these would the Itanium fit into?

Most of them.

First off, for their servers they are looking for throughput. High
single threaded performance is pretty pointless as compared to high
multithreaded performance. Sun is very well positioned in this market
because of their software, Solaris is probably the best operating
system on the planet when it comes to scalability (with the possible
exception of AIX). However the hardware side of things helps Sun out
here as well. Good I/O is very important for performance in this
market while raw number crunching isn't such a big deal.

If you have a 16 core machine and 128 core machine and they both
perform a task with similar performance which one do you choose? What
is the obvious answer to that?

Do you have any evidence that HP-UX scales worse than AIX?

So, what solution does Sun have? A little chip called "Niagara". 8
cores, each capable of executing 4 threads simultaneously. It's
single threaded performance won't even be half that of the Itanium,
but for multithreaded performance it will SMOKE anything Intel has in
the pipeline before "Tanglewood" in 2007/08, and it will do so with
about half the power consumption (or less) and a smaller die.

SMOKE?
How can you know this? As far as I know very little is made public
about Niagara, but you make lots of claims here. I also think the
latest was that Niagara would be released in 2006, the same year
Tukwila is supposed to be released. Perhaps Niagara can be released
before Tukwila, but I don't think there will be over a year between.

 

[Nick Maclaren]

|>
|> > So, what solution does Sun have? A little chip called "Niagara". 8
|> > cores, each capable of executing 4 threads simultaneously. It's
|> > single threaded performance won't even be half that of the Itanium,
|> > but for multithreaded performance it will SMOKE anything Intel has in
|> > the pipeline before "Tanglewood" in 2007/08, and it will do so with
|> > about half the power consumption (or less) and a smaller die.
|>
|> SMOKE?

You're quite right - it's the Victoria annd Blue Nile Falls that
are "the water that smokes".


Regards,
Nick Maclaren.

 

[Yousuf Khan]

SGI's market is technical computing. Itanium is a good fit.

It seems to be Cray's market as well, and they're doing it all with two
different types of Opteron servers.

Yousuf Khan

 

[Yousuf Khan]

1) Money. SGI is very short of it. Investing in a new chip
takes money to develop and test the hardware and software.

Which seems to be the biggest calling card for Opterons. Don't have a lot of
money (or at least don't feel like spending it) to develop a server on your
own? Just go Opteron and buy from a bunch of prepackaged barebones systems
and engineer your own badge.

2) Contracts. Some of those with Intel may be exclusive, at
least for now. SGI signed the deal with Intel when Intel had the
whip hand.

This one is possible, but Intel only controls an advertising budget whip
hand. A lot of companies (such as HP and IBM) are simply getting around it
by not advertising non-Intel systems, but selling non-Intel nonetheless.
Some are even advertising them with a separate ad stream.

3) Commercial politics. Intel bears serious grudge, and SGI
can't afford to be on the receiving end from Intel.

Definitely SGI can't afford it, unlike Sun or various others. I wonder if
SGI would switch to 64-bit Xeons eventually though?

Yousuf Khan

 

[Nick Maclaren]

|>
|> > 2) Contracts. Some of those with Intel may be exclusive, at
|> > least for now. SGI signed the deal with Intel when Intel had the
|> > whip hand.
|>
|> This one is possible, but Intel only controls an advertising budget whip
|> hand. A lot of companies (such as HP and IBM) are simply getting around it
|> by not advertising non-Intel systems, but selling non-Intel nonetheless.
|> Some are even advertising them with a separate ad stream.

That is not true. Intel used its position to get some 'interesting'
clauses added into at least some agreements.

That was almost certainly impossible with either IBM or HP, both of
which were never in the subservient position that SGI, Dell etc.
were at the time of Intel's maximum ascendency. This MAY be part
of the reason that it is IBM and HP that have broken away from the
Intel line. The same applies, to some extent, to some Japanese
companies.


Regards,
Nick Maclaren.

 

[Yousuf Khan]

That is not true. Intel used its position to get some 'interesting'
clauses added into at least some agreements.

That was almost certainly impossible with either IBM or HP, both of
which were never in the subservient position that SGI, Dell etc.
were at the time of Intel's maximum ascendency. This MAY be part
of the reason that it is IBM and HP that have broken away from the
Intel line. The same applies, to some extent, to some Japanese
companies.

The only thing that seems to be allowing some of these aforementioned
Japanese companies to break away from the line is that they are no longer
afraid of Intel. No written contracts (which Intel is too smart to do), just
implied threats of supplies being cut off. But now it looks like the
Japanese don't care, and neither do HP or IBM.

Yousuf Khan

 

[Nick Maclaren]

The only thing that seems to be allowing some of these aforementioned
Japanese companies to break away from the line is that they are no longer
afraid of Intel. No written contracts (which Intel is too smart to do), just
implied threats of supplies being cut off. But now it looks like the
Japanese don't care, and neither do HP or IBM.

Er, there assuredly WILL be written contracts with Intel. But I am
certain that they are absolutely kosher. If there is any pressure
that isn't, I agree that it won't be in writing, but I have no
evidence that Intel has stepped over the legal line. Intel doubtless
has a large supply of M'Learned Fiends who would jump on anyone who
implied the contrary ....

It is perfectly legal for a contract to require an exclusive product,
and I have reason to believe that some contracts with Intel do just
that. The point is that Intel could get away with it when they had
the whip hand, which they never did with HP and IBM to the extent
that they did with SGI and Dell.


Regards,
Nick Maclaren.

 

[Tony Hill]

If you have a 16 core machine and 128 core machine and they both
perform a task with similar performance which one do you choose? What
is the obvious answer to that?

The "cheaper" system of course, with cheaper being measured as some
sort of TCO (ie a rather fuzzy number).

SMOKE?
How can you know this? As far as I know very little is made public
about Niagara, but you make lots of claims here.

Enough has been made public about Niagara to figure this one out.
It's a chip with 8 simple SPARC cores on it, each capable of running 4
threads. Even if the individual Itanium chips are MUCH faster,
they'll only be running one or two threads until "Tukwila" (as you
hint at bellow, the code-name "Tanglewood" has been changed to
"Tukwila"). That means that the Itanium will have to be MUCH faster
than the individual Niagara core to perform the same in a heavily
multithreaded environment, at least 4 times faster (assuming dual-core
Itaniums), and quite possibly much more than that if the SMT
performance of Niagara works out like Sun says it will. Considering
that Intel's plans for Itanium over the next few years consist of
little other than increasing the cache size and going to dual-core
designs, I don't think they'll come close. Increasing cache size
helps, but it's a case of fairly rapidly decreasing returns (how much
does your hit rate increase when going from 9MB of cache to 12MB? I
would guess it's MUCH lower than 1%). Dual-core is a good start, but
they'll be rather late to the game (at least a year away still).

I also think the
latest was that Niagara would be released in 2006, the same year
Tukwila is supposed to be released. Perhaps Niagara can be released
before Tukwila, but I don't think there will be over a year between.

Niagara is scheduled for 2005, though I expect it to be delayed until
2006. Tukwila is also scheduled for 2006, though as I mentioned
previously, Sun has an existing market so they have a bit more
flexibility in timing. Intel needs to build their market so they need
to execute much better. Customers will delay for a little while to
get an easy upgrade rather than make a difficult upgrade right away.
They will not delay in order to make a difficult upgrade if such a
change can be avoided altogether.

 

[Michael S]

So, what solution does Sun have? A little chip called "Niagara". 8
cores, each capable of executing 4 threads simultaneously. It's
single threaded performance won't even be half that of the Itanium,
but for multithreaded performance it will SMOKE anything Intel has in
the pipeline before "Tanglewood" in 2007/08, and it will do so with
about half the power consumption (or less) and a smaller die.

"single threaded performance won't even be half that of the Itanium"
is sort of understatement. Even biggest optimists at Sun would be
surprised if Niagara achieves 25% of Montecito's single-threaded
performance. My bet - under 20% on all but most ugly zero-ILP
workloads. Most likely we will never know since Sun isn't going to
publish single-threaded benchmarks.