The PS3
has just been 1-upped.
Sun Microsystems looks poised to lead the "mainstream" multi-core race for at least a couple more years. By late 2009, the server maker should deliver a third major revision of its Niagara processor which will have 16 cores and an astonishing 16 threads per core, The Register has learned. Today, Sun sells an eight-core " …
Too bad Sun can't boost their stock like they can core and thread counts. Another soft quarter and we're going to have to vote Jonathan and the board of directors off the island so that an appropriate corporate raider can gut this company like a fish. Sorry folks, at this point I just want my money back. Ashlee, is there a reported clock speed for the Niagra 3?
"....The problem, however, seems to be that the UltraSPARC Tx boxes eat up sales of other low-end systems, which has led to a steady decrease in shipments at Sun, while revenue has been flat to falling...." Oh, and the fact that competitors seem to be growing market share, i.e. converting Sun customers away from Slowaris, that has nothing at all to do with Sun's decline?
Anyway, tape-out is usually the signal for a Sun chip to get canned like UltraSPARC V, so until either Rock or KT actually hit the shelves you shouldn't be filing these stories under "Server" but under "Vapourware".
Sounds to me like Rock is dead. 2nd Tape and no news is bad news. Their stock is $2 and change (ignoring the 4-1 reversal) And it is very deceiving to say each core does 16 threads. If it round robins treads then each thread is only getting 87.5 MHz. Pick your metric but it's not both.
Mr Mojo Risin
Azul has been using HW based TxMem for years (hence, the Sun lawsuit), and does not require recoding. Perhaps the OS (and the OS' VM system need recoding), but apps should not. The whole idea of an OS is to abstract HW from applications. Anyone with a rudimentary knowledge of computer science knows that.
I have not heard anything on TxMem for Niagara 3, but given the number of threads, it makes sense. I think 256 threads is about the upper limit on most machines today. Tukwila on Superdome and SPARC64-VII on SPARC Enterprise should bring that to 512 threads. I don't think any machine can go above 512 threads per OS image without a significant change to thread and memory management.
I would not be surprised if SW based TxMem is in place in Solaris and HP-UX today.
...as always on the reg. pitty they might financially underperform right now - and who doesn't in this market - but they certainly are among the most innovative companies around.
if the will be able to monetize their products remains to be seen, but I must say we are very happy with our t2 boxen - and we came from x86! solaris too has been a pleasant surprise so far. just wish they get their act together on open office and we probably would move the whole institution front-office to back office to sun hard/software.
and jeremy - last time I checked solaris was open sourced and running on intel just fine - like on my macbook pro here...
You'll get it for free with the JVM though. Fujitsu or not, somebody needs to save the company from Wall Street. Sun's workstation pricing is an embarrassment to human intelligence. Who buys these things??? If I were CEO I'd either price the workstations to sell and scale the product line or cut them entirely. Jonathan might be right about a lot of stuff, but he has to take some steps to stay in the game. If the door is open, going private would make a lot of sense.
Jeremy writes : "No one uses Solaris anymore - everyone that matters has already upgraded to HP-UX running on the world's fastest HP servers. Sun and Schwartz-boy should've ported Slowaris to Intel and made it open source."
Jeremy, I'm sure you're just being amusing, but for those that might think your post is genuine, the open source edition of Solaris (for Intel/x86 or Sparc) can be downloaded at ....
http://www.opensolaris.com or http://www.opensolaris.org
Does anybody *really* use HP-UX anymore? - It's dead already isn't it?
For those unaware peaking in :
if anything Sun is a lot more relevant to the world than HP is.
Even IBM only manages to keep it's on in the high-end of the Unix server business but find it's mid-range a bit too expensive to sell for mainstream purposes.
No if any player in the Unix world is bound to become extinct it's going to be HP although they have a strong Blade play which is keeping them standing.. for now.
How many cores and how power efficient it is doesn't matter when a single core yield half the performance of the cheapest intel-core. And the float performance is so low Sun does everything to confuse the documentation of it. Since the clock-rate is the same, this is probably only a die-shrink and just adding more cores, a very simple job compared to redesigning the internal pipelines for higher clock range. (die-shrink comes from processing technology and not design)
So I believe Sun has seen the end of SPARC, they just want to squeeze some more juice out of it until the X86 line grows more momentum.
The poor margins of the X86 hardware sales is a more severe problem. Where is Sun going to get its future revenue from, Their current business model is unable to get margins on commodity hardware based on volume.
"pitty they might financially underperform right now - and who doesn't in this market"
Well apparently HP and IBM, who seem to be riding out the storm pretty well in comparison based on their last few quarters...
Now my question - 2048 threads sounds great - but can the solaris kernel actually do SMP efficiently at this level? I've seen no evidence that this is the case - unless someone can point me at some literature to the contrary...
Paris, cos like Sun she knows you can never have enough threads
If you cracked 16 cores to 2GHz the chip would be caning a good few hundred watts at peka load - very difficult to cool and expensive to run. Typical use for systems like this would be a bigass file/http server or a db that serves a lot of very boring queries. Neither of which needs much CPU, just enough to push the memory aorund and get it out of the ethernet port.
How many of the bitter people on this forum got bitten by Sun's falling share prices or got laid off by them.
For pretty sensible people your being a bit panicky.
The 128 Threads is obviously not round robin scheduling that would not make it any different to what intel and everyone else has been doing for decades. Instead it's a variation on sueprscalar processes (known as simultaneous multithreading). Each thread can be executed in parallel (up to 128 in total).
I think sun are still pretty innovative but they are getting a beasting by the markets trying to recoup some of what they lost. I dont think it would take much for Sun to return to somewhere near the heights they were at in the mid 90's.
I have observed that those who had shares in Sun and sold in 99 are the ones in electronic and computer company car parks with BMW's and porsches. Those who held on into the 00's are driving 10 year old mondeos. Maybe that's where the bitterness comes from.
*I am in no way, and have never been, affiliated with Sun. By the time I graduated from uni they were already on their knees!
As an architect I have designed a number of very large scale Oracle based systems and there was a time when SUN were the only choice.
But... those days are long gone and the truth is it is going to take a lot for me to move away from commodity INTEL/AMD kit running linux!
With clustering going mainstream in Oracle RAC and servers being cheap enough to put on a company credit card the days of making big bucks out of hardware are dead and buried and resurrection would take something very special!
I like SUN, always have, but it's time to face up to the truth.
@Morten Bjoernsvik: "How many cores and how power efficient it is doesn't matter when a single core yield half the performance of the cheapest intel-core"
What are you talking about? This is totally application dependant. Most Java web applications for example need throughput and decent, but not fantastic single-threaded performance. Individial Azul cores are slower again and these boxes absolutely fly in terms of throughput.
According to Richard McDougall
http://www.acmqueue.org/modules.php?name=Content&pa=showpage&pid=328&page=5
Solaris scales up to thousands of threads and hundreds of processors.
I suppose performance would depend on the type of job. I suspect these Niagras are good for web servers (lots of concurrent accesses) but maybe not so hot for database work.
As ever, YMMV.
RE: Tom
"....Does anybody *really* use HP-UX anymore? - It's dead already isn't it?..." Yes, outside of all the large telcos, the majority of industrial, financial and medical players, basically all the FTSE 100 companies, and the majority of the FTSE 500, hp-ux is comatose! Of course, inside those areas it's being used in preference by ex-Sun customers quite happilly! Oh, did I mention that Integrity is the only enterprise server range with positive growth, and that Sun's server sales are in continual decline?
RE: AC FUD
"....No if any player in the Unix world is bound to become extinct it's going to be HP although they have a strong Blade play which is keeping them standing.. for now...." Aw, someone else that forgot the printers! And the market-leading PC sales. And the the number one worldwide server slot by shipments and revenue. And being the number one in disk-based storage. And number one in management and information software. Hold on a sec - HP are the World's largest IT company, of course they're going to fold when you compare their healthy spread of leading technologies to a one-trick-pony hasbeen server vendor like Sun! How many years has it been since Sun went four consecutive quarters in the black?
RE: dave
The main thrust of your argument is sound, but what is a "peka load"? Are you measuring CPU heat generation by holding hot chips against parts of your anatomy!?!
The reason that applications didn't need recoding for Azul was that the transactional memory was handled by the Java VM. It's find if you have an abstraction model like that, but for anything exploiting it in binary you need to recode. Given that there aren't any common hardware standards for transaction memory then it's a bespoke job.
On the subject of the hard multi-threading and the speed equivalence, it's far from the case that you can just divide the clock rate by the number of threads on the core to establish the effective clock rate of an individual hardware thread whn the system is fully loaded. In the first instance, the later Niagara2 processor actually has two integer units and one floating point processor servicing each core with 8 threads. These hardware threads are grouped into two groups of four and hence at the pure integer processing level, then the worst ratio you get is 4:1, not 8:1. There are several other components to the core which are shared resources. The Niagara processor uses a very fine grained hardware thread model that makes use of the core resources that might be used by stalled threads. Of course it is always possible to come up with a worst-case scenario where all data and executables are held in processor cache and then integer throughput will be cut by a quarter. However, on large, real-world applications, this is very rarely the case and there will be a lot of conditions which will stall the tghreads and leave resources which can be exploited.
However, the Achilles heel of this technique is the single-thread performance. If you have a CPU-intensive application which depends on single-thread speed, then response times can suffer very badly. We had to reconsider the use of T5220s on a major call-centre application as, whilst throughput was good, response times were poor. In general most commercial transaction-processing apps only exploit multi-threading by running multiple transactions in parallel. Very, very few can exploit multiple cores within a single transactional sequence to any great extent.
The other thing to point out is the Niagara lacks much of the complexities that speed up single-thread performance on other processors. For instance, it has no proper super-scalar features such as parallel processing on a single instruction thread, no out-of-order execution and isn't hot on things like branch prediction (on the basis that any such stalls can be exploited by another thread).
The other interesting issue about hardware threads is that they are essentially virtual CPUs. At low levels of utilisation, traditional CPU measures can be highly misleading. A machine running at 40% busy might appear to have plenty of headroom, but in reality the invisible back-end resources in the core might be at 80% utilisation (with a Nigara you can actually get these figures - on the rather crude hardware threading on a SPARC64 V, Pentium Hyperthreading and the ilk then there's no way of finding out). So for hardware multi-threading, be very careful about linear extrapolations of CPU utilisation...
You could put even more Z80s in the same space. And yet it wouldn't automatically nuke the SPARC. Building high-performance computers is sometimes slightly more complicated than that (not that you'd know it looking at El Reg comment threads and the rest of teh internets nowadays...)
Sun was and still is late to the ball game with a GUI for system administration.
Sure the old hacks (myself included) knew the command lines by heart. (I go way back to Sun 3/60 days in the late 80's.)
But IBM and HP had SWIT and SAM making life a lot easier. HP also brought in OpenView to help make it easier to monitor the network.
Where's Sun?
Do they have a graphical console app to help manage the system?
Thats the first key.
The second would be the ease of adding third party apps, disks and monitoring.
The fourth? Partner with a database company or pick up PostgresSQL and make it their own to do hi performance parallel computing. (Along the lines of Informix's XPS.) Or work out a deal with IBM because IDS was the basic database platform that could actually take advantage of the massive parallelization of Sun's new cores.
The problem with Atom-based systems is not the CPU but the power-hungry motherboards available. It would be interesting to see what a purpose-designed board with sixteen Atoms could do, especially as it could run bog standard x86-64 Windows or Linux apps without a recompile or rewrite, but for the complexity and price you'd probably be better off with a quad-core Xeon anyway. The real problem would be a single-threaded app running on a multi-core Atom system would suck in just about the same way as a single-threaded app on Niagara, only it would be at a fraction of the cost and power bill so you wouldn't feel quite as much a mug.
We talked to some Sun reps about these chips when looking to upgrade a small cluster. Their verdict was that we'd be better of sticking with x86 (from Sun of course ;). While the cores and power consumption seem compelling, the low clock speeds and minimal floating point support mean you're not necessarily going to get a performance boost. It can be very application specific. Incidentally, sparc is nowhere on the top 500 list that just came out.
That said, I understand that Washington State University built a cluster using sparc for BLAST (a genomic data search tool) and, after they retooled it for that architecture, were thrilled with the results.
"....Over 15million downloads of OPEN Solaris and counting...." The point here is this needs to be a revenue stream to Sun to replace the small fortune it used to rook customers of with SPARC Solaris. So, Sun has had to pay for the server infrastructure and bandwidth to host all those downloads, what it wants to get back is SUPPORT REVENUE from people taking the downloads and putting them into a commercial use which they are willing to pay for (like they do for Red Hat or SuSE support). Either that or BUY Sun x86 kit to put the downloads onto. And here's the bad news for Sunshiners like you - they aren't buying the Sun kit, which is why Sun is desperate to get other more popular x86 vendors to sell and support Slowaris x86 for Sun; and they aren't buying support contracts to go with the Sun downloads. So Sun is LOSING money on Slowaris x86, which is why Scott McNealy is begging HP to support Slowaris on ProLiant - because HP is the number one x86 vendor by a massive margin and already sells shed loads of Linux support contracts.
In fact, the vast majority of downloaders seem to be just looking out of curiosity (straw poll amongst admins down at the pub, alcohol may have affected voter's responses). Last figures I saw was only 60,000 support contracts sold for ALL Slowaris x86 ever, a fraction of what Red Hat and SuSE have sold, and probably wouldn't even factor into a Microsoft market share conversation.
"....where have you been??" I've been working in the IT industry, mainly in large corporate environments. I assume you have been sharing headroom with Fazo?
Niagara is not a processor that is good at number crunshing.
It's designed to be able to handle many simple requests at once like you have with a webserver, file server or in some cases even database server (depending on what is queried out of the database).
It definately would never perform good in HPC but there is not a single top 10 server that could handle webserving, file serving as well as these Niagara servers. Horses for courses.
For some reason people don't seem to be able to graps this easy concept, yet you don't see everyone racing around with Ferrari's on the highways... Why ? Because they may be fast but they only seat two (at best) and .. well.. they're absurdely expensive for shopping or traveling purposes.
this is not my expertise but sun apparently has sold a cluster to the high performance computing virtual laboratory in canada: http://blogs.sun.com/HPC/entry/hpcvl_turns_to_cmt_for
I am also aware of at least one german university evaluating a t2 cluster right now.
uc berkeley recently had a paper comparing matrix-vector multiplication on x86, niagara and cell. the results where surprisingly good for niagara: 'astoundingly, the full system single socket (64thread) median results achieve 3GFlop/s, more than 3 the performance of a single socket of the 86machines.'
http://cscads.rice.edu/publications/paper19
cell of course beat the crap out of niagara too. anyhow - all that talk about niagara being slow is just plain bs. given the right application those machines deliver very good performance at low power consumption. and in few years time we will see x86 processors using the very same model that sun uses for t2 - simple, relatively slow in-order cores amassed on one chip. of course since intel is doing it then everybody will brag about how cool it is...
It's when too many people are overzealous about tell you that something is absolutely not important that one knows that that something must scareing the hell out of them.
The Hardware Compatibility List says it all. It's not only IBM, Dell and Fujitsu that make darn sure their servers support Solaris x86 and are on the HCL. No, even HP has gone through the trouble of certifying some of their servers for Solaris x86.
Only HP kinda does not like to say it out loud but they all LOVE Solaris x86.
And then there's the multithread play from Sun : everybody understands it but it's often better to dish out FUD than to do the right thing and catch up.
I love the tone of the discussions here (not). We know what Niagara is good for (large numbers of autonomous threads) and not good for (things which are dependent on single thread performance). The analogy Sun used was the difference between a bus and a Ferrari - if you want to get one person somewhere in 10 minutes use the Ferrari, if you want to get 100 people somewhere in 30 minutes use the bus. If you want to spread FUD, say "This bus is no good because I can't get there in 10 minutes".
And can we grow out of "Slowaris" - it is one of Sun's jewels, rock solid and stable. It isn't an accident that you can run a piece of code compiled ten years ago on today's OS on today's hardware, it's far sighted design in both the hardware and the OS.
I work in a large commercial institution, that moved away from Sun back in 2001, and is now moving back. HP-UX is a joke, we punted Itanium out of here years ago. We can't afford Oracle any more after their recent price hike, and Sun's the only company delivering a commercially supported open source DB (two in fact). So for all the whinging about them, they seem more relevant to me than either HP or IBM, both of whom seem to have blown their brains out when they embraced open source, but forgot to ship open source products.
Unless I missed the irony of the post about HP-UX and Solaris.
If not meant as ironic, I can't help but ask:
What platforms can hp-ux run on, and is it open?
Itanic and PA-RISC.
Oh, great range there. PA is dead. Itanic has a limited life, will be completely overrun by X86 when the new CSI architecture ships. Basically, it is a product nobody but Intel and HP have any future plans for. Even those are shaky.
And - When was the lat time you were browsing the source of hp-ux? Open? No.
What about Solaris platforms and openness?
Sparc? Yep
X86, both 32 bit and 64 bit? Yep
Highly available or commodity hardware? Yep
Open Source? Yep - http://opensolaris.org
(Granted, that's Opensolaris, not Solaris 10)
Free? Yep (Though you pay for support, though still less then Redhat...)
And - for the 87.5Mhz comment, there is little more I can say but... Get some idea before commenting on something you clearly know nothing about. You are assuming that a) There would only be only pipeline and b) the single pipeline can execute only one instruction per cycle c) The cost of waiting for memory is zero d) The cost of waiting for I/O is zero
All of these are bad assumptions, and the very reason the niagara series systems are a $1bn business...
I think HP sells more x86 solaris systems than Sun. Just goes to show you the OEM agreement does not mean much.
Richard:
I would like to know the person who wrote the Solaris download virus..that is the only way I can make sense of 15M downloads and Sun is still dying fast. The US population is 300M. So 5% of the population downloaded solaris?...I doubt more than 1% of the population knows what solaris is.
"Very, very few can exploit multiple cores within a single transactional sequence to any great extent." .... By Steven Jones Posted Tuesday 24th June 2008 12:24 GMT
I concur, Steven. And whenever the Opportunity is So XXXXStreamly Exploited? Does that become A.N.Other Revenue Stream/Honey Money Pot?
Scandanavian dDeep C Phishing is such a Private Delight ... AI Treasure Trove.
And the Jolly Roger for Pirates, on Course Enlightenment Missions, of course.
With more that 100% growth in a $1b business the Niagara series is solving real customer problems and migrating to this technology is easy with Binary Compatibility guaranteed by Sun for any Solaris 8 or higher application. The real value is that you do not have to continuously change applications and processes over time with the introduction of new OS or hardware revisions.
That makes it a very low cost platform to stay on.
CPU speed is no longer the major bottleneck in many applications - Oracle for example is highly threaded and is a perfect candidate for this technology.
With virtualisation at the heart of many projects - memory, storage and IO is where the battle commences. That is why Sun packs double the memory into their X64 range and is investing in new virtualised IO technologies.
On a last point re:- "The problem, however, seems to be that the UltraSPARC Tx boxes eat up sales of other low-end systems"
These systems are Sun's entry level SPARC servers - you get 4-cores / 32 threads as a minimum with an X64 price tag.
"only way I can make sense of 15M downloads and Sun is still dying fast. The US population is 300M. So 5% of the population downloaded solaris?."
Sorry to puncture your balloon, but there is more to the world than the US, and Sun sales outside the US exceeded its sales in the US some 15+ years ago. Solaris downloads show a similar trend. The Microsoft monopoly is driving many non-US users towards open source, and Solaris is one of the OSes benefitting from that.
coming from a legacy 5000+ seats microsoft installation we evaluated linux vs. opensolaris. decision has yet to be made but opensolaris has come out very attractive, not least because of the security model and integration with id-management products sun offers.
we would get support for os, openoffice and mysql from one vendor. being able to run linux software with little to no modification is also a big plus. we liked zfs for our storage-requirements but there are some headaches with older storage solutions we still have to keep around.
what sucks is the lack of drivers for commodity hardware, missing package system and open office still has ways to go to be as easy to use as office.
I hope sun manages to ride out the current slump. looking at the way both markets and regulation (!) is heading they really do offer a very consistent software & support strategy, they are just a little to much ahead of time...
hp-ux indeed is a joke for the majority of markets. proprietary software and hardware lock-in - that is so 1992! and then the regulation nightmare when you deal with public clients on a large scale.
and one more word on niagara - if you look at the berkeley paper you'll see that in this application the cell beats the niagara by 2x, but then it runs at 3.2ghz versus 1.4ghz on the niagara. I really start to wonder what a 2.1ghz niagara would do.
btw - ours run at 1.2ghz and we are perfectly happy with the performance.
Telco's do. That is why they are priced at completely rip-off prices. Moving to an IP based voice system (no, not that crap Asterix/Skpe shit, the real heavy duty stuff), it seems Sun have given developers free/low cost kit (their high end kit naturally), so telco's need to buy high-end Sun kit to run the software controlling the network.
In the next few years I predict Sun's share price will rocket as all telco's move off the analog phone systems and since there is no software on the market apart for stuff running on Sun boxes....price rise in the future and higher profits are assured.
Mine's the one with a telephone cord hanging out the pocket.
Short Stocks: Bets build against Sun
NEW YORK, June 24 (Reuters) - Short interest on Nasdaq rose
almost 6 percent in mid-June, the exchange said on Tuesday,
suggesting an increase in bearish sentiment in the stock
market.
The following stocks saw increased interest from short
sellers, who bet a certain stock's price will fall. The
data reflect short trades with a settlement date of June 13.
SUN MICROSYSTEMS INC (JAVA.O: Quote, Profile, Research, Stock Buzz)
Short interest more than doubled in shares of the world's
fourth-largest business computer maker.
Morgan Stanley cut its rating on the company last month,
saying the server market is seeing a shift to more lower-priced
systems, causing Sun to lose market share due to its high-end
server exposure. The company posted an unexpected third-quarter
net loss last month.
Some 57 million of the company's shares were held short, or
about 7 percent of its total shares outstanding.
The 24/7 Wall St. Ten Worst Managed Companies In America
With the trading year almost half over and results from the first quarter out, 24/7 Wall St. has created the latest installment of its Ten Worst Managed Companies In America list.
This analysis is based on: 1) one year and five year stock performance relative to the major indexes and other companies in the industry, 2) the company's position in its industry both now and over the last five years, 3) whether management made identifiable and critical decisions which hurt the company, 4) a change in the company's relative market strength compared to its competition, and 5) whether the company could have identified mistakes and changed course quickly enough to avoid a catastrophe.
Some readers will think it is not fair to include companies which have had a recent change of management. While it may be true that a new CEO gets a "honeymoon", if his early, significant decisions do not create a substantial change in the company's fortunes there is little reason to hope for later improvement.
1. Sun Microsystems (JAVA). Sun's stock has underperformed for several years. It did bounce when Jonathan Schwartz was named CEO about two years ago, but his tenure has been a great disappointment, both for investors and employees. Sun has made several expensive acquisitions including buy-outs of MySQL and StorageTek. Sun's revenue remains flat and its operations swing between tiny profits and small losses. Sun's chief scientist and head of sales recently left the company. News from JAVA comes in the form of almost daily and not useful PR. The company has changed its ticker symbol and has done a reverse merger to move its share price into higher territory. Whatever chance Sun had to claim a major portion of the global server market is lost. IBM (IBM) and Dell (DELL) are in the final stages of beating Sun into the ground. Its bets on Solaris, Java, and open source software have failed. Sun's shares are off over 50% during the last year to $11.14.
Peace Frog
HP/UX .. hmm, i remember that. How do I install it in at VM on my x86 Xen box again? Actually more seriously I would like the chance to do that, get back up to speed on an OS that no-one I work with can handle.
To Dave,
"If you cracked 16 cores to 2GHz the chip would be caning a good few hundred watts at peka load - very difficult to cool and expensive to run."
well actually that's manageable. We handle 5kW / rack routinely (power, temp maintained sub 15C). Problem? OK admittedly we can't put 21 boxes in a 42U but who cares - there isn't a data centre that matters that is not power-constrained rather than space-constrained. If anyone can arrange a 200MW feed into a sensible part of the UK, then post :-)
Just can't resist the temptation to wind up such a frothing Sunshiner! It's like Mr Pickles is taking notes from the Sun Marketeering droids.
"With more that 100% growth in a $1b business the Niagara...." 100% growth of next to nothing is still next to nothing! And that "$1bn" is not Niagara alone, Sun is inventively counting in services, support and anything else they can think of! Niagara is miles behind it's true competitors (Opteron and Xeon) in market share, revenue and units shipped, and there is simply not enough existing Sun customers willing to buy it to give it a chance of getting out of it's niche. This is shown by the decline in Sun's market share - Sun customers are moving to other vendors' kit because they simply don't believe in Sun anymore.
"....The real value is that you do not have to continuously change applications and processes over time with the introduction of new OS or hardware revisions..." Really? 100% guaranteed compatibility? Forever? I think not. Otherwise Transitive wouldn't be selling licences for their emulator software to run existing SPARC apps on Linux just for a start, and Linux on x86 wouldn't have spent the last three years that Niagara has been available happily eating Sun's buisness.
"....That makes it a very low cost platform to stay on...." But not cheap to get on in the first place! And definately not cheap if you are banking your business on your competitors not stealing a massive advantage by using more innovative kit from HP, IBM or Dell!
"...Oracle for example is highly threaded and is a perfect candidate for this technology...." Every time I hear that Oracle runs "just great" on Niagara I nearly split a side laughing. Trust me, I've benchmarked it using real world data, and it doesn't. If Niagara could handle large loads like Oracle then Sun wouldn't need Rock or the badged FSC SPARC64 servers or even x86 servers. The truth is Niagara is just so niche it wouldn't survive on it's own without other CPUs providing more depth to the server range, hence the Sun love-in with x86 and SPARC64.
"....That is why Sun packs double the memory into their X64 range and is investing in new virtualised IO technologies....." Technologies which the rest of the market have already got in spades. In short, Sun is playing catch-up yet again, and just stuffing boxes with memory is not going to make up for their serious lag elsewhere.
"....These systems are Sun's entry level SPARC servers - you get 4-cores / 32 threads as a minimum with an X64 price tag...." And still being outsold by Integrity and Power, and massively outsold by x86. I wouldn't be surprised if Intel's Atom CPU as generating more revenue by this time next year.
So, to summarise - the customers don't believe in Sun, they're buying elsewhere; Wall Street doesn't either, which means even less customers will want to risk buying Sun; and the competition all seem to be doing much better, especially those like HP and IBM that have a much wider product portfolio. Think of the old IBM marketing slogan "No-one ever got fired for buying IBM" - no-one is going to think that about Sun in their present state.
"Sun Marketeering droids" if only Sun had a marketeering department then perhaps the stock would be better. Getting back to the point .... the technology does the talking.
Niagara is pushing the envelope with built in with 10 Gigabit, line speed encryption and virtualization built in as standard in most cases. Even the PCI Express IO is on the silicon. All that means fewer parts and points of failure, lower costs, and vastly reduced power consumption.
Again to reiterate - this is Sun's ENTRY level SPARC range and focussed on massively parallel network centric workloads. It is also suitable for many standard application workloads up to a point. For 80% of the time it does the job very well. For the other 20% you have other products in the SPARC range.
Just one last point - if the thread speed was upped to 2GHz then the low power advantage would be lost. The 2GHz + version is called Rock.
The technology is different - heresy to some even - but it gets the balance right between price / massive throughput and very low power consumption and is very popular as a result for those who care about such problems.