Sun Sacks UltraSparc V and 3300 Employees 408
bender writes "According to this article, Sun Microsystems has cancelled the next generation UltraSparc V processor even though the chip had already taped out. Perhaps this has something to do with the recent partnerships with AMD and Fujitsu?"
They are working on SPARC.NET (Score:4, Interesting)
Re:"Stupid! You so STUPID!!!" (Score:3, Interesting)
Re:Beginning of the End (Score:4, Interesting)
Well, I don't know that Sun is in the same marketspace as AMD/64. Personally I'm still wondering what will fill the gap in every corporate data center I've visited recently. There is a strong trend that I see.... everybody has their "x86" room- it is usually PC Servers running Win32 or maybe Linux. But here is the real trick- the "x86" room is always intranet type apps and *maybe* the rare external web site that gets lower volumes.
The rest of the datacenter might be things like Sun 6500's, 10Ks, or holy shit, a 15K or two. What fills the gap here? I'm starting to see more and more large IBM servers moving in. I guess IBM is really going to capitalize?
Also, BTW, a lot of shops now only have a "token" mainframe as I call it. A 390 box that sits at the back of the datacenter happily running whatever few legacy servicing systems might be left that will undoubtably be maintained for years to come. The IT guys still attached to these boxes as admins or programmers are an interesting breed. Talk about skittish folks.
Closer than you think (Score:5, Interesting)
0. Gosling leaves Sun for IBM.
1. All Sun hardware will run on AMD
2. Sun will port
3. Java bytecode will target the CLR
4. Sun/MS/HP vs. Intel/Dell/IBM/Linux
5. Apple keeps innovating
Perspective (Score:5, Interesting)
This tree was a HUGE oak tree though - had to be 100 feet tall at least, with a trunk that was probably 5 feet wide. And it sat there in a big planter box waiting to be "planted." The transportation costs alone must have cost a fortune.
The point is, while the industry began plunging into the abyss, Sun was farting around buying full-blown oak trees to make their campus look "pretty" - while other companies were working to stay afloat.
It seemed then that they had their blinders on, and while a fair amount of companies are stabilizing now here in the valley, they seem to be trying to stop the bleeding a bit late.
Perhaps if they'd spent less time farting around with building campuses and more time on building their market, they'd be in better shape. After all - if you let your employees go, who's going to look at the trees?
Just a thought... it seemed symbolic to me of what was wrong there - perspective. Shame though... they're so much more likeable than MS.
Yes. (Score:5, Interesting)
After the election, HP and IBM will be doing some as well, but it it unknown how much.
Short sighted plans (Score:4, Interesting)
"Then, in late 2006 and 2007, the company will release Niagara, a multicore, multithreaded chip."
Sun will somehow finish a significantly more complex processor when they give up on this one? IBM, AMD, and Intel will be four times ahead of Sun in three years. By killing the UltraSparc V, Sun has to execute perfectly in an arena they've stumbled in the past.Obvious... (Score:3, Interesting)
No more Sun in EDA? (Score:5, Interesting)
But now EDA vendors are starting to support AMD64. With Sun's announcment, the performance gap is going to get wider. No Ultrasparc V. Niagara and Rock won't help, even when they get here.
"The technique, which won't result in chips larger than those from competitors, sacrifices the ability to perform one task extremely quickly for the ability to do multiple independent tasks simultaneously"
No good. No good at all. How long before Synopsys, Cadence, and Magma do the unthinkable and actually drop support for Sparc/Solaris?
Re:Obvious... (Score:5, Interesting)
I've been working with the HESE (High-end Server Engineering) group there for almost 4 years making ASIC support chipsets for this cancelled SPARC program ("Eagle"). We had already taped out one, had the first design for another cancelled two years ago, and were 70-80% done with it's replacement (they switched from InfiniBand to PCI Express, which was smart, but resulted in tossing away about $80M in development and lots and lots of cancellation fees from my company.)
ASICs that should have taken 8-12 months tops were scheduled for 2+ year development cycles, then Sun's delays stretched that out even more. It was frustrating for me, since I spent a lot of time waiting for netlists and constraints, and aside from 2-3 key (lower-level) people I worked with who were competant, I saw so much waste and stupidity in the Sun management organization that I often got mildly depressed about it.
Their management is sorely lacking in hierarchy -- there are dozens of people with power to influence any decision (they are "stakeholders" as Sun calls them) yet never any one powerful enough to make a final decision, and many of these folks are too smart for the company good. Rather than pick a workable implementation and go with it, they would have meeting after meeting for months arguing about which way was "better". There was never any "main manager" who would step in and halt the endless nitpicking and force a decision. This delayed projects to an almost silly degree, and it's hard to believe how incessant it is unless you see it yourself.
So, just about everyone I worked with in Burlington was laid off. Some were given the chance to move to California to work on the SPARC stuff still going there, but most of their managers advised them that this program will also be cancelled within a few years, so unless they just wanted to go to California (few do), they should take the severance and run. Everyone I know did just that.
So, now the project I was working on for Sun that was cancelled and revived slightly differently once, is now completely cancelled. My company still got paid, but nothing like what we would have made had we gone to mass production (though even those forecasts were dropping steadily every year before cancellation). Worse, we had 60+ engineers in Japan and four here in Mass. devoted to Sun, and we even turned down some projects last year because we didn't have the engineering resources to handle them. Now we wish we had those back, and our sales staff are hustling to bring in some more work.
It just makes me sick, since I always thought of Sun as the great, innovative company, and I was so thrilled to be able to work with them (at first), and now they fall apart in front of my eyes.
On the bright side, I did get some great free trips to Japan and Australia on a extra-juicy expense account during the initial design win when we were wooing Sun every-which-way. Even met my wife on one trip to Japan. So it's not all bad for me, but it sure sucks for Sun.
Re:Perspective - Wrong (Score:4, Interesting)
Re:Bummer... (Score:5, Interesting)
Eh? The UltraSPARC performed very poorly on things like SPEC, that were mainly CPU benchmarks. However, Sun machines generally performed well in real-world server scenarios, where the better architecture made up for deficiencies in processor power.
Additionally, the UltraSPARC processors weren't as fast as x86 but they scale much better and have no end in sight
How well a CPU scales is more a function of the machines memory and bus architecture than the CPU itself. x86 CPUs like the Opteron can scale very well --- its just that Sun machines are much more commonly equiped with the cross-bar memory controllers and other system support that you need to get a scalable machine.
whereas the x86 can't compete in large multiprocessor systems and are starting to show future caps in terms of power, heat, and size
I wouldn't compare SPARC so much with x86 as I would compare it with PowerPC, the former Alpha, PA-RISC, and Itanium. Relative to the other major RISC architectures, SPARC CPUs themselves were never very impressive.
Re:No more Sun in EDA? (Score:2, Interesting)
I'm an EDA admin and people love using x86 boxes, and complain when stuff runs so "slowly" due to requiring a large amount of memory as it had to run on SPARC boxes. So yeah, we're getting Opterons.
They won't drop SPARC/Solaris for a while, but I can't say news like this is good for Sun in the EDA space. Multi-threading apps to run on these multi-core chips is difficult, and they charge you a license per core so it'll be expensive to use it!
So instead, you'll run it single threaded on your brand new Opteron/POWER4/Itanium2 and be happy in single threaded land.
Re:"Stupid! You so STUPID!!!" (Score:1, Interesting)
Then companies like Fujitsu and formerly Tatung repackage the newer hardware without the unfortunate design compromises that made the new boxes undesirable (such as the RAM limitations and insistence on out-of-date bus technologies), port the old software to the hot new cheap box, and sucks up all the money Sun was expecting.
Re:Sun excised the SPARC VI proc and decided on Ro (Score:2, Interesting)
Re:Perspective (Score:3, Interesting)
Most of the strategies that come to mind were tried by one or more now defunct companies. Silicon Graphics decided to go Wintel and what good did it do?
You can fault Sun for not being economical, but when your basic business goes out of style, belt tightening just prolongs the inevitable.
fujitsu primepower so much better than ultrasparc (Score:5, Interesting)
The PrimePower 850 just blew away the V880, even with 2 less cpu. The PrimePowers use Sun Solaris and are 99.9999% * compatible because (I didn't realise this) that Sun do not own the Sparc design, Sparc Consortium do. I do not believe that Fujitsu will buy Sun outright because they simply do not have the money and have been doing lots of expensive merging of various subsidiary companies this year to save costs; e.g. the old ICL has become Fujitsu Services along with some other straggler companies including Fujitsu's Sun reseller company.
I would say that Fujitsu PrimePower are about 1 year ahead of Sun in terms of power & speed and in our tendering process were a lot cheaper as well.
Probably worth mentioning that I didn't buy Fujitsu in the end because the machines were not certified to use Oracle RAC - instead, I went for HP (linux) - the business benefits for linux outweighed the change from solaris.
* PrimePower won't run SunCluster - that scared me a bit about fujitsu's compatibility claims.
Writing on the wall... (Score:5, Interesting)
I think Sun started dying when they started to push remote framebuffer devices as a viable business solution. Besides costing more than a PC, it required extensive reworking of the network in many cases. They killed off (then brought back) Solaris on Intel when sticking with it might have slowed down Linux adoption in the data center (people looking for cheap hardware -- PC servers -- are generally not looking for Sun boxes). Sun was riding high on the dot.com and Y2K booms but they were too slow, too entrenched to react when the landscape changed. Their hardware can no longer keep up with equivalent priced Intel machines with equivalent availability features. Hell, even the Apple machines are eating into traditional Sun markets in research and academia. Why? Their low-end, slowest machines are still $1,200 more than Apple or Intel.
Don't get me wrong. I liked Sun and still do. I want them to survive not only because it makes my skills more valuable, not only because they were largely friendly to open source, but because they have developed some cool technologies. But they have to change. Maybe these moves are a good thing (they can't be worse than the previous path). But they have to do more: quit being so wishy-washy with Linux (either embrace it fully or compete against it); make Java easier to install on Linux (I don't care if it's opened up or not); make Solaris9/Intel as functional as the Sparc version (where's SMC? At least make a Linux SMC client); lower the hardware prices to be more in line with the industry (even if this means putting together an IA32 or IA64 machine).
turnabout sucks eh? (Score:3, Interesting)
My, how things have changed!
Not that Apple didn't deserve criticism in that era (I worked for a successful project that is still underway, however) but there were some damn fine people there that didn't deserve to be ridiculed.
Pardon me while I enjoy a certain amount of schadenfreud at Sun's expense.
And yes I feel terrible for the Sun people that were let go, its a rough market right now and they are (as I am) just pawns to the powers that be, that don't have any compunction about playing with peoples' livelihoods. I have no ill will towards the workers at all, just toward their executives.
Re:Beginning of the End (Score:3, Interesting)
H-1B Fallacy: SPARC64-V versus UltraSPARC-V (Score:4, Interesting)
Another interesting point is that the SPARC64-V was made almost exclusively by native (Japanese) engineers. Fujitsu, as a matter of traditional Japanese corporate policy, does not hire H-1B workers.
Sun hired hordes of H-1B workers. About 66% of the people who worked on the UltraSPARC-V were former/current H-1B workers. This observation proves the fact that H-1B workers are not needed to create high-technology.
Here's the sweetest part: Sun will sell re-badged Fujitsu servers, starting in 2006. I know. I work in Sun's server department.
Installed software base cause of cancellation? (Score:3, Interesting)
This is probably the real reason behind the cancellation -- moving to the UltraSparc V would have obsoleted the installed base of software (or at least would have required code changes to get the benefit of the new architecture).
And then the article goes on to say that after all those customers port their software to the V (at some huge expense), they'd have to port their stuff again to the next generation of Ultra Sparc processor. No wonder it was killed -- IBM learned that lesson back in the System/360 days. The last thing you do is prevent existing programs from working on your new machine -- because at that point the customer will say: "Well, we have to rewrite our code anyway, let's see what other hardware vendors have to offer."
Chip H.
Re:Worse financial situation than we think? (Score:5, Interesting)
SGI tried to stop fighting and work with Microsoft on Fahrenheit among other things. They pretty much cratered the company in the process. Not sure anyone remembers Fahrenheit, but it was an attempt to develop a next gen 3D API beyond Direct3D, OpenGL and Performer. It became very obvious from day one that it was mostly designed to divert SGI's attention from backing OpenGL against Direct3D. SGI was dreaming of defining the 3D standard for all those millions of Windows desktops. Microsoft wasn't concealing the fact that everything going on there was irrelevant unless it could be shoehorned in to Direct3D:
http://www.theregister.co.uk/1999/11/29/ms_quie
Microsoft simply DOESN'T work with its competitors. It very rarely works with its partners. Its partnerships tend to be a smaller companies who think by partnering with Microsoft they are going to be "made men" in the mafia sense of the term, but at some point if the project is successful or not Microsoft will, one way or another "whack" the partner. The one strategy most likely to lead to a desirable outcome is for the small company to sell its assets to Microsoft, probably for less than they are worth, but still make a tidy profit and run.
Microsoft seduces, it bullies, it uses slight of hand misdirection, it uses, it simply doesn't partner. One thing Scooter used to have right at SUN, you either fight Microsoft or you die, the only other viable strategy is to look small, don't get to profitable, and hope Microsoft doesn't notice you before you cash out.
Re:Worse financial situation than we think? (Score:2, Interesting)
Whoever has the most troops, normally, wins regardless of quality (i.e., the Soviet Union in WW2; sure, there's exceptions, but that's the rule)
Whoever has the most money in American business, normally, wins regardless of the quality of product.
"Quantity has a quality of its own" -Lenin
"Congress is the best money can buy" -Will Rogers
The law means power, real power. Law, money, government, i.e., congress:
"LOMAX: Why the law? Cut the shit, Dad. Why lawyers? Why the law?"
"MILTON: Because the law, my boy, puts us into everything. It's the ultimate backstage pass, it's the new priesthood, baby." -The Devil's Advocate
Whoever has the most money to throw at lawyers to keep the opposition in court, to lobby senators (former lawyers) wins in the end. It's a numbers game which translates to a dollars game. Perry Mason, John Grisham is a fantasy. There aren't any cinderalla stories when it comes to law. Money rules it. Microsoft has an ocean of money to keep even large corporations in court, making them build bigger and bigger legal departments within the corporation sucking more and more money off of the widget development the company should be doing instead.
At the corporation I work at, our legal department has grown exponentially over the past few years. Lawsuits reign supreme and permeate IT and every other department. We manage IT now always wondering "what will legal think?" It wasn't this way just a few years ago in '98 or '99.
Make the competition sink money into legal matters, into the law, into lawyers, drain a corporations money by making it go to court -- drain the swamps and the Seminoles wither from within....
Re:turnabout sucks eh? (Score:3, Interesting)
I worked at DEC. Scott took great pleasure in using DEC's situation to his advantage. DEC got in bed with MS, Scott jumped up and down. DEC moved to Alpha, Schott jumped up and down. DEC/CPQ/HP cancelled Alpha, Scott jumped up and down.
I don't think Scott is jumping right now.
I feel bad for the Sun employees however. Very bad.
Re:Yes. (Score:3, Interesting)
For a MS shop, I would expect that you could layoff 10% or more, but not so for a Unix shop.
Take a look at the costs of the systems.
With Unix, you have a high up-front costs of buying the system. From there, your day-to-day operations are fairly low costs due to how much work a single sysad can do.
With Windows, you get to buy cheap computers (1 time cost), but then you start paying high dollars for programs (anti-virus, office, etc) and several times more sysads.
So where does Linux fit in this, it uses fewer of the cheap computers combined with low number of sysads. So that will save lots of money for somebody switching from MS, but quite a bit less for a Unix shop.
The real problem here is that Sun simply pays costs for their hardware (far less than you or I do) and the number of sysad do not change if you are running Linux or Unix. So will it save Sun 9%? Very doubtful.
Re:Not what it looks like (Score:1, Interesting)
Rock will have the a ability to create two threads from one (some sort of "thread level paralellism", besides the clasical ILP), in order to maximize CPU utilization. Dont forget that Solaris has the most advanced thread implementation on the planet. They will laverage this advantage.
As for workstations, chances are that they move back to a third party processor (probably Opteron) as they did with the original Sun 1 (a Motorola 68.000 based workstation), back to the roots baby!
Im really expecting wide Solaris Opteron support from ISVs, since this will easy worstation deployment for end users. Nowadays, for Linux, you have some ISVs that only supports RedHat 7.3 (Landmark, etc.), while others supports SuSE, forcing end-users to have dual-boot or vmware implementations in order to mantain ISV support for the high-priced software (petrol apps, etc.). Whats even worst, is that is common for new libraries to be incompatible with old ones (glibc 6.22 and 6.23 and more) what forces ISVs to perform extensive re-certification. Thanks to binary application guaranty (http://www.sun.com/service/support/sw_only/solar
I still thinks that Sun drop the ball with many bad choices, but replacing US-V to with a extremely different processor (as Rock) is the best way to cut through the chat. Either Sun will raise or fall from this desition. If it really works, a Rock + "asynchoneus logic" processor will position them on a hole new game, forcing all other competitors to perform an expensive (time & money) catch-up.
If it fails... I doubt services will save them. As my father once told me when he was CEO of a service (telco) company "To the customers eye, service is always bad. After they get used with any new improvement, they will start to complain again requesting some further improvement, until their complain is solved, then the hole thing starts over again." Thats why long term out-surcing contracts tend to end really baddly. Is not the quality of the service, is human psiquis...
Thats why Sun, beeing a engineers company, will be far better with serving value added products (with huge differentiators) than services.
Re:No more Sun in EDA? (Score:1, Interesting)
Rock will have the a ability to create two threads from one (some sort of "thread level paralellism", besides the clasical ILP), in order to maximize CPU utilization. Dont forget that Solaris has the most advanced thread implementation on the planet. They will laverage this advantage.
As for workstations, chances are that they move back to a third party processor (probably Opteron) as they did with the original Sun 1 (a Motorola 68.000 based workstation), back to the roots baby!
Im really expecting wide Solaris Opteron support from ISVs, since this will easy worstation deployment for end users. Nowadays, for Linux, you have some ISVs that only supports RedHat 7.3 (Landmark, etc.), while others supports SuSE, forcing end-users to have dual-boot or vmware implementations in order to mantain ISV support for the high-priced software (petrol apps, etc.). Whats even worst, is that is common for new libraries to be incompatible with old ones (glibc 6.22 and 6.23 and more) what forces ISVs to perform extensive re-certification. Thanks to binary application guaranty (http://www.sun.com/service/support/sw_only/solar
I still thinks that Sun drop the ball with many bad choices, but replacing US-V to with a extremely different processor (as Rock) is the best way to cut through the chat. Either Sun will raise or fall from this desition. If it really works, a Rock + "asynchroneus logic" processor will position them on a hole new game, forcing all other competitors to perform an expensive (time & money) catch-up.
If it fails... I doubt services will save them. As my father once told me when he was CEO of a service (telco) company "To the customers eye, service is always bad. After they get used with any new improvement, they will start to complain again requesting some further improvement, until their complain is solved, then the hole thing starts over again." Thats why long term out-surcing contracts tend to end really baddly. Is not the quality of the service, is human psiquis...
Thats why Sun, beeing a engineers company, will be far better with serving value added products (with huge differentiators) than services.
HP Knows Linux Well (Score:3, Interesting)
Re:Beginning of the End (Score:1, Interesting)
Re:H-1B Fallacy: SPARC64-V versus UltraSPARC-V (Score:3, Interesting)
This is true of UltraSPARC-IV, not UltraSPARC-V.
Sun's UltraSPARC-V was going to be a traditional continuation of the SPARC line vis-a-vis bigger faster more Hertz. Sun's next generation processor is going to focus on non-traditional approvements vis-a-vis multi-core processors like 2, 4, 8 processors on a chip. Something like 7 CPU cores for ALU and 1 CPU core for FPU or 6 ALU 1 FPU 1 IO core.
Dubbed Throughput computing.
IBM's revenues (Score:3, Interesting)
I am not disagreeing about IBM's hardware sales, but IBM has become a services company, and they leverage services to sell the hardware.
According to this report of IBM revenues [infoworld.com], services were $10.4B of total $21.5B for 2003Q3. Almost half the revenues are from services, and the profit margin on services is much higher than the margin on hardware.
This year-end report [gartner.com] states that all hardware sales increased, including the mainframes (z-series). But it points out that services revenue grew 17%, while total revenue grew 10%. Do the math. That means the non-services business only grew around 3%. If that trend continues, then in 3 years, service will account for 3/4 of IBM's revenues. Aren't statistics wonderful? While the growth of services may not be maintained, selling hardware keeps becoming more difficult, so these numbers are possible. The first report states that hardware revenues declined 1%, so you guess if IBM's hardware business is actually growing.
---
Software is included on the non-services half. The report states that IBM's software sales have flat-lined. If hardware revenues declined, then software must have grown some to offset the hardware decline to reach the 3% non-services growth. Most of the increase is because IBM keeps (successfully) pushing WebSphere, which competes against free software.
New business model:
1. See Free Software succeeding.
2. Develop proprietary version.
3. Use marketing and support organization of very large company to sell it.
4. Profit.
(I dislike business plans that include "and then a miracle happens". My current startup is depending on several of them, and they will give me ulcers, especially since I am expected to provide the miracles.)
The one real advantage of pushing WebSphere is that development is so complicated that IBM sells more services. IBM stopped pushing Lotus Notes because development is so easy that your receptionist can do it, so it generates much less money from services.
IBM has not been pushing Lotus Notes recently. That may change soon. Lotus Notes dominates because it allows business people to create business applications easily and quickly. Notes 7 will allow the use of DB2 as the internal database structure. Then it can scale to almost any application's needs. It could also mean easy use of DB2 for mobile applications. If they can maintain the ease of development, Notes could take a significant portion of the application market from MS and Java. The issue is whether IBM will market it well. They spent most of the last 7 years positioning Lotus Notes as a competitor to MSExchange. Notes is a much better email system than MSExchange (try administering/supporting both for a while), but Notes shines as an application platform, and IBM buried that message in the competition with MS for number of email users.
-- Back to SUN
My first thought was that the deal with MS included unwritten conditions that SUN would stop selling hardware that could not run MS software. Then I realized I was being completely paranoid, because even if Scott has absolutely no idea what to do next, he would not give up the Sparc for just $2B. Right?
Re:Worse financial situation than we think? (Score:3, Interesting)
Their recent hardware hasn't really been that good. So far in my limited experience Sun processors have failed more than Intel CPUs have. Go look in the archives - Sun's CPUs are slow AND not something I'd rely on. 2nd level cache probs, memory probs, go look it up.
Heck even the Athlons and Durons probably have a better reliability track record than the UltraSPARC III. Otherwise AMD would be dead now.
If you talk about the rest of the hardware, Sun shops from the same bunch as the x86 server makers for most of it.
Maybe it isn't all their fault - Texas Instruments has a lot to answer for. But who cares about that?
If I want high end SPARC performance - Fujitsu. Their software runs on Fujitsu. Their hardware either sucks or is the same as the rest. So the only reason to buy Sun stuff may be support.
Which is probably what Sun are doing - getting rid of the underperformers and focusing on what they are competitive in.
Sad, but what else can they do?
That "Throughput Computing" stuff only buys them a little time. As if the x86 folks don't care about throughput and real world performance. Proof: _HP_ is making Opteron servers.
Not much mercy:
http://h18004.www1.hp.com/products/server
Wonder who holds patents on "memory controller on CPU".
Re:Bummer... (Score:3, Interesting)
I have:
333MHz UltraSPARC IIi: Ultra 10 128MB RAM in one bank.
300MHz IBM G3: Old Clamshell iBook 192MB 222 PC-100 SDRAM
300MHz Intel Pentium II: Generic PC 384MB 222 PC-100 SDRAM
In Ubench, the x86 and PPC are much faster than the UltraSPARC (in the processing test, not memory). I can't remember how much quicker the PII was, but the little G3 was 2-3 TIMES faster than the UltraSPARC.
Building a full release of OpenBSD 3.4 (including X), the UltraSPARC took 7 hours and the G3 took 5. I assume the slow notebook drive slows the iBook considerably in this regard.
I was shocked.
They say that the UltraSPARC FPU is where its performance is impressive, but I've seen benchmarks which don't seem all that great to me.
From all accounts I have heard from, the Blade 100 with it's relatively highly clocked III, runs like a two legged dog, compared with generic Intel gear.