You'll all be shocked to learn that Sun Microsystems appears set to delay the release of the Rock processor. Word reached Vulture Central this week of troubles in the land of Rock. Sun hoped to ship the 16-core SPARC dynamo by the end of next year. Now, however, we're hearing that early versions of Rock have struggled to …
Rock doesn't rock after all!
Absolutely no surpise there then!
BTW last paragraph is incorrect, Sparc Enterprise servers with four or more sockets are all Fujitsu not just the processors.
Sun do manage to stick a badge on the front though, well done.
Some are Fujitsu(SPARC64). Some are not(UltraSparc).
actually Sun & Fujitsu jointly own final assembly plants in Scotland, US, & Japan
A tad wrong there chap!
AC, if you look at some Sun employee resumes on LinkedIn, you can put the whole picture together yourself. For convenience, here it is as a summary:
According to the Global Supply Chain Manager at Sun (who I'd guess has no reason to fib on his resume at http://www.linkedin.com/in/richardkenyon ) much of the Fujitsu & Sun server is farmed out to contract manufacturers. So you shouldn't have posted "Enterprise Servers....are all Fujitsu". Its not even a believable claim! Your statement implies Fuji makes the servers because Fuji is 100% vertically integrated. That is beyond belief and not even supported by Richard Kenyon's LinkedIn profile.
I did some more investigating then to get the real story.
On two other employee insider LinkedIn resumes, there is mention of jointly financed & owned final assembly factories in Scotland, US, & Japan. So your statement "are all Fujitsu" is refuted again.
I'm a forensic accountant and CPA, but I don't do global tax accounting. Therefore, I'm not sure how the tax accounting for the US & Japan works out. It hurts my brain to think about it. It also hurts my brain as to how they split costs & profits 50%-50% with the Yen & Dollar constantly changing. If the accounting has to be done on a quarter-to-quarter basis (at least in the US), that means that as the factories were built they had expenses to report in both countries. And to be fair to both parties, they would have to adjust for currencies quarter to quarter.
AC, why don't you find that out for us and then post here? Maybe you can redeem yourself even though your own view of yourself seems to be "Anonymous Coward" by your own hand. If you solve the accounting mystery for me, then you can get a "well done" gold star to attach to your Register UK posting as I will personally mail you one! I used to be a school marm and I think I still have some. If it doesn't stick well to the screen, you can use Scotch tape.
Richard Kenyon's LinkedIn Summary:
An experienced people and engineering manager currently accountable for a team of 20, located in Europe and the USA, responsible for all supply chain execution aspects of New Product Introduction through to End of Life of Sun's High End Server Line, a multi billion dollar business. I currently specialise in the management, communication and analysis of information throughout the supply chain that is required to monitor in as near real time as possible this significant supply chain. I also specialise in relationships management, working with contract manufacturers and am also leading the execution team introducing Sun's latest flagship products in partnership with Fujitsu with Fujitsu, a industry unique competitive partnership. Current bugdetary control is $100m+ for delivery of material purchases into a multigeography outsourced supply chain.
I believe I have a unique career centric approach to employee management and development, that results in improved employee morale and a greater awareness on behalf of the employee to do what they need to do to succeed in their careers and meet the goals of the business. Projects are assigned to employees whose career goals are most closely aligned with the experience to be gained from working on the project. This sets the incentive amongst employess to pay careful attention to their own goals and objectives in a given year of employment to maximise their chances of working on projects that will help them achieve their stated career goals.
Richard Kenyon’s Specialties:
Supply Chain Management, High End Server Engineering, Information Analysis, Advanced Statistics, Significant Management Experience, Global Team Leadership, Joint Venture Relationships Management, Employee Career Development, Engineering Process Management, Product Reliability Management, Data Center Management, Software Engineering
Shipping on time and Clearing up Anonymous Coward's Stupidity
The UltaSPARC IV and IV+ shipped on time.
The N1 and N2 processors shipped on time.
It certainly looks like the follow-on processors to the N1 and N2 are going to ship on time, too. So where does Mr. Vance get his info? Competitors?
It is true that UltraSPARCIII and V took forever. Sun learned from those mistakes.
ROCK incorporates features not found in IBM's processors or Intel's. Furthermore, most of Itanium's SPEC numbers are rubbish compared to even the N2 processor. IBM's power6 is only available in one mid-range product. Why? Because it's a hot, power-hungry monster.
As far as Anonymous Coward is concerned, Sun was intricately involved in the design of the large 4+ socket Enterprise servers. It is true that the products were mostly designed by Fujitsu, but it is an empty-headed insult to insist that Sun engineers did nothing more than mount the bezel. It's no wonder that the idiot who posted that comment self-describes himself as a coward.
Re: Sun & Fujitsu jointly own final assembly plants
The manufacturing plant in Scotland is entirely owned by Sun and is one of 2 plants they own worldwide - the other is in Oregon.
All of their low-mid range kit is handled by contract manufacturers (including the smaller US IV and US T1|2 servers), but their highest kit (F25k & M8/9000) is built internally at both Sun owned sites and Fujitsu's own facilities.
/A different AC ...
the best of the bes, the men of men have MONEY!
I'd say the moneys on Intel ha ha ha ha ha!
I'd rather wait on Sun to get thier act together than run an IBM.
When are we going to see the realisim here?
Sun BUILD the servers??????
Maybe so but for Mx000 servers, all of the components are Fujitsu based! , OK for the M4 and M5 boxes, SUN build the cases, again well done!
Bottom line is the non Niagara based systems are All Fujitsu, Joshua, I can concede that Sun added some technology, the fans I believe?
Suns behaviour here is classic, getting their A$$ kicked with performance/price of USIII/USIV, can't beat them, join them. If Sun arte so great, why do they need to use FJ stuff anywway?
Wake up boys!
Not a massive delay.
All chip manufacturers usually slip their dates, and this isn't a massive slippage, and many first gen prototypes get tested with sections of the chip turned off or bye-passed, especially if you are bug-hunting. Usually, it's big news if a manufacturer DOESN'T announce any slippage between creation and release to market. Much as I would love to start calling Rock the UltraSPARC V TNG, it will probably still ship as Sun really don't have any choice, they have to supply something or admit defeat and settle for just integrating other people's CPUs (FSC's, Intel's and AMD's). If Rock does slip again, then I'd start looking for signs of Sun really hyping Slowaris x86 as this will be an almost sure sign that they will kill Rock.
BTW, Joshua - the UltraSPARC V didn't "take forever", it got canned. It will not EVER be released to market. It was aborted. Terminated. Died in the harsh light of reality. It was a dead parrot. Just get over it and move on. :P
The reason ROCK is late, or will have the same fate as V
ROCK's initial wattage was >300W it was supposed to be 200W at most. The Transactional memory interface didn't work either. It would also be more appropriate to call the chip a 4/16 core. (4 real/16 mini)
The spec's are here: http://www.geocities.com/sunsrockchip/
We will have Tukwila in the mkt by the time rock arrives, if it ever does.
Fujitsu did the chip, memory and interconnect, Sun did the I/O, Which the Fujitsu engineers were not happy about. Sun rushed into the agreement when they canceled V and are hoping customers will still buy USIV+ (with that nasty 150MHz interconnect) until ROCK comes out. Customers will have to wait a long time or move to an architecture that has a future. Any word on the OEM agreement with Fujitsu being extended past 2008?
Rose needs to get a life!
I made it clear that 4 sockets and above were FJ kit, Forensic accountant is a great title, however read and understand!
Sun own the 'assembly line' The silicon, Mother board and chipset design and manufacture is done by Fujitsu.
Stop believing the SUN FUD and B/S, and see the reality, SPARC64 IV was a fix for Suns failing roadmap, which (and back to the main point has slipped YET AGIAN!)
US IV on time? I don't think so, soon we'll be back to stray cosmic rays and alpha particles taking Enterprise Class machines down! See the articles on this.
BTW the Enterprise Class machines referred to were E10K's, bought from Cray, I see a pattern, don't you?
The forum suck factor is strong tonight...
...with A/C coming out of the woodwork trying to whip up a flamefest.
So, does it all mean that Sun proves incapable of getting modern, cutting-edge, high-power, complex CPUs out of the door?
Was Don Capellas right and the only ones left that are able to "do" these kinds of CPUs are Intel and IBM, with AMD orbiting the Intelsphere?
Say it ain't so!
With Niagara II already performing so well and Victoria Falls (2 sockets, to give 16 cores and 128 threads) apparently on the way, many (most?) Sun customers will not need anything bigger to run their enterprise.
LDOMS are not virtualization....its thread partitioning
Joshua, you are funny...and you might want to lay off the Kool Aid
Technical capability vs financial credibility
Is Sun incapable of getting modern CPUs out of the door? I'm sure that, given unlimited time, Sun could bring the Rock chip out and even meet the technical promises they made in the first place. After all, the UltraSPARC V was all taped out when it was canned for the simple reason it was too late, not because it hadn't met the technical goals.
The market doesn't stand still, and what is killing Sun's chip program is the same thing that killed Alpha, PA-RISC and MIPs - meeting the technical goals inside the timeframe that allows for good financial gain is getting tougher and tougher beacuse it takes such a large investment. Capellas was right in that only Intel and possibly IBM have that much dough and facilities, unless a new player rises in the Far East, and I don't think FSC has what it takes to give SPARC64 the technical edge to give it a longterm competitive advantage.
Sun, HP, Compaq, IBM and Sun all looked at the future of CPUs back in the 80's - originally they all agreed a replacement for RISC was needed, it was only the path that they fell out over. Even Sun started out onboard the Itanium path, only to jump off when they thought their market presence meant they could force the "Solaris on SPARC and nothing else" message down customers' throats regardless. It looks like Sun made the wrong play, and even if Rock does make it out of the door we'll see another sweetly embarrassing flip-flop from Sun when they are forced to announce Sun support for Solaris on Itanium or Power. Anything else will reduce Slowaris to an also-ran Linux clone.
Intel X86 architecture beats Sun
I think the main reason why Sun continues to sell non-X86 servers is due to application lock-in and familiarity with the company/sales team.
Quite frankly, just about every test I've ever seen, implementation I've ever looked at has shown Intel X86 running Linux as better performance per watt that a similar Sun architecture and significant price-performance over Sun.
I'm not familiar with the very high end product line E12K, E15K, etc. but again, those architectures that are able ot be moved can be moved and the resultant savings to the customer are immediate and significant.
Sun isn't as fortunate as IBM to have a lot of proprietary banking and medical applications written for (and ipso facto supported on) its Unix platforms. Thus, Sun is more vulnerable than its competitor in the market space where it's still able to make a few roubles. I suspect this is the reason why Sun has changed its stock ticker from SUNW to JAVA - it's ultimately planning to get out of the hardware business.
...the Rock is cooking but we can't yet smell it?
Performance per watt...
...is only really one way of appraising a system.
I'm not convinced that it, or even cost is always primary driving factor when it comes to architecture decisions for high availability projects.
You could, of course, look at reliability.
Uptime is kinda important in some environments.
I have worked on everything from very large single box systems, to distributed computing using thousands of machines.
I couldn't count the number of x86 machines I've seen go pop. Even really expensive (and new) blade servers.
I've yet to see a machine with a Sun logo on the front die, even seriously old ones.
I see no reason why cheap/less reliable systems don't have their place in the world; look at Google and their virtually disposable Linux machines. But not for certain functions.
I can't imagine any board going for cheap/ecofriendly when it actually costs them money when availability drops below 99.9%
As to the issue of proprietary applications, I'm afraid I'll have to call an untruth to that too - and is it really that relevant, as long as a C compiler is available on every system? Of course, I suppose you could always run things on Java....
Is it possible to argue that SPARC sells because of application lock-in, but also the lack of applications is an issue, anyway? Colour me confused.
seemed to have missed the point
......... that SMS have yet again dropped a road-map commitment!
People are not logically attached to big SMP (RISC) , but emotionly attached to it.......
Sun is becoming a software company, agreed, not bad for a hardware vendor!
My Little Pony
Remember about a year ago, when Schwartz was telling people not to buy SPARC64vi systems, and to wait for Rock systems in 2008? Makes him look pretty stupid now. Maybe he should spend a little less time blogging, and pay a little more attention to the systems business.
And by the way, the N1 and N2 systems have about the best performance per watt out there.
I aggree about the emotional attachement bit. You always love the first ...
But many so called professionals have spent far too long in hte Sun Ghetto, slow machines, ancient operating system rubbish applications messianic marketing, they need to get out more.
Sun has always been my C development platform of choice but not for the reasons you might think! Its the easiest machine to "segfault" I have ever come accross, and, Solaris has the most minimal POSIX support that can actually still get to be called POSIX, so if your POSIX C code compiles and runs on Solaris it will run on anything!
Beliefs and facts, buy a low power, multi-thread CPU from Sun.
Points raised and some facts
Assumption: Only IBM and Intel can make/design chips/CPU's
Fact: Sun niagara chip N1 has sold better than Itanium, new N2 is ramping up strong. If Sun can make/design N1 & N2 this implies that they make and design CPU's. Where the design of Sun chips is industry leading. Go buy one from Sun if you believe they cannot make one.
Assumption: Intel superior architecture
Fact: Every system has it's place and different purpose. Intel does not have the mult-core or on CPU encryption & ethernet of Sun. Sun does not have the smaller die size of Intel.
Fact/Anecdotal: Intel servers run hotter than Sun, anything that runs hot from a physics perspective is under greater strain. Anything under creater strain/heat etc will have a shorter lifespan. Look at the X4450 from Sun, neither IBM or HP can make an equivalent Intel server this small.
Customers who grew up with Intel came from the low end upwards. Sun came from low end to high end and does both. You get what you pay for.
Sun is leading the market.
IBM and Sun started with multi-core chips.
But Sun put more cores on a chip than IBM, IBM has not been able to compete.
Sun made the chip use less power. Intel followed.
Sun has encryption and ethernet on a chip. Intel and IBM either cannot or do not have the capability to do this.
Summary, let everyone pick on Sun, but look at the facts.
there seems to be...
...a lot of people posting as AC on this one. anyone would think you guys were a little unsure of what you've been spouting!
More cores != better performance.
" But Sun put more cores on a chip than IBM, IBM has not been able to compete. "
IBM doesnt want to compete in a core count competition! IBM has probably more experience of SMP than any other hardware company in the world (since 1972) and are well aware that throwing more processors into a box is not a get out of jail free card.
Most reasonably complex systems have points where the logic must be single threaded (allocating mutexes, locking a row in a table, ordering database redo logs etc. etc.) at some point the speed of a single processor will become the limiting factor in any multithreaded system (or at least any system you are likely to buy off the shelf today). After you hit about sixteen processers you dont see much performance improvement.
Sun, IBM, SMP and innovation
I am afraid that innovation does not come, can be sustained or maintained by the large companies. They all have innovation, IBM labs are innovative, Intel can make smaller chips, but reducing a Chip size (die size) is not architecure or industry leadership. Sun is very innovative and their products proove it, go buy one.
As far as SMP goes, yes MVS & z/OS as called has a long heritage, but solaris is has always been designed to multi-thread. Other UNIXes (including Linux which is Unix) may not multitask or thread aswell as Solaris. Maybe this is why competitors say that multithreading with multi-cores is not the way. Ask why competitors are scared and talk down multithreading systems.
If you cannot do it, then criticise those that can.
Sun, competitors cannot create/design CPU's or Operating Systems like Sun, thus they criticise Sun.
HP and IBM cannot design a well cooled, fast and compact 4 CPU Intel system like Sun e.g. the X4450.
If you want to consolidate many single threaded apps on several servers then with Solaris on T5000 servers is the way to go. Many customer are consolidating their 100's of web servers to 10's of Sun T5000 servers.
Someone from IBM said that the world does not need more than 5 mainframes, if we now say a CPU does not need more than 16 cores. Will that statement be valid in 10yrs. Will it survive the test of time.
Read my blog a couple months ago about Intel and Sun's role reversal. Sun has CPU design leadership and Intel moves to software.
Beliefs and facts, buy a low power, multi-thread CPU from Sun....
> Assumption: Only IBM and Intel can make/design chips/CPU's
> Fact: Sun niagara chip N1 has sold better than Itanium, new N2 is
ramping up strong. If Sun can make/design N1 & N2 this implies that they
make and design CPU's. Where the design of Sun chips is industry
leading. Go buy one from Sun if you believe they cannot make one.
The N1 has sold well but so has Xeon processors from Intel; per core the performance or complexity of the T1 chip is pathetic - it does great of edge applications but not so hot on core applications such as database serving due to it small cache size; the T2 is an improvement but it has still not addressed this issue. The T1 also had a single FP unit (shared across all 8 cores) which resulted in poor FP performance which required by most commerical applications and HPC kind of applications. To the point; just because it sold in huge quantities it does not make it a great processor - by that logic x86 Xeons would be the best processor out there. PS the real competition to the US-T1 was the x86 processor and not the Itanium2 9000 or POWER 5+/Power6.
> Assumption: Intel superior architecture
> Fact: Every system has it's place and different purpose. Intel does not
have the mult-core or on CPU encryption & ethernet of Sun. Sun does
not have the smaller die size of Intel.
Intel does in fact have multicore processors; on chip accelerators for ethernet or encryption is not a current requirement of x86 Xeon market space or that of Itanium. This is not terrible difficult to implement but requires real estate on the chip so the trade off designers have to make is the value of these functions to the cost in real estate on the chip.
> IBM and Sun started with multi-core chips.
No, IBM came out with the first commerical multi-core chips in 2001 with lots of features which Sun is still to incorporate in its processors including Simulataneous multithreading (running 2 independent threads through a single processor core on each clock cycle), a distributed intrerprocessor switch (which scales system bandwidth with the clock speed of the processor and / or with the number of processors in the server) a full 6 years later.
# Is Sun a strong competitor? Yes.
# Will they survive? Yes.
# Will they revolutionize the world with their processor designs? Maybe not.
The guy from IBM was right.
Thomas Watson actualy said
'I doubt if there is a market for more than 10 of these mainframes".
The mainframe in question being a warehouse sized vacuum tube powered monster specially commissioned to manage the hundreds of radar installtions etc. that defended the US from the Soviet menace in the 1950s. They actually sold 12 I think. Just goes to show he knoew his market.
I would also like to point out that its applications that need to multithread and as Sun is not actually a software vendor yet. (They are a major giver away of software but as nobody actually hands over money they cannot really call themselves a "spftware vendor". Multihtreading Java applications seem just as slow and unweildy on Sun hardware as any other platform.
And if sun are so good how come dont float to the top of the TPC benchmarks which these days is an IBM/HP only club (with a guest appearance by Fujistu on Intel).
Scope creep, TPC, Itanium, cores etc
To many subjects to cover, but good questions raised.
Yes, OK it was 10 m/f and not 5, but do you get the idea. Underestimation, have we underestimated multi-core. Intel now is following Sun wih multi-core CPU's.
Is TPC a good measure of performance, this question will run and run. All CPU's wait at the same time. Put a Ferrari in a traffic jam, see how fast it can go.
How long can Itanium last before HP/Intel kill it. Will itanium move to X86 with a virtualisation engine above it, including the overheads. Itanium is a dangerous path to follow, I did not invent the Itanic slogan.
Sun now has the T2 with more floating point units, thus any comparison with previous generation T1 chip is outdated.
Future issues will be power and cooling IBM/HP recommend more cooling, why not stop the power consumption hot chips first. Install Sun T5220 and you may not have to update your cooling system. Use prevention rather than cure. Sun can prevent you going down the hot datacenter path, this saves money. If you are not a drug/power addict, you may not need to go into rehab/extra cooling. Get the idea, do not make the high power mistake and then you may not have a cooling problem. Many webservers can be consolidated to T2, see:
For the record on the manufacturing of the Sparc Enterprise servers, everything written here is to some degree wrong in terms of the details.
All I can say is that Sun outsources some sub components and final product assemblies to regular EMs/CMs, sources chips and some other sub components from Fujitsu and also manufactures some product internally. So there you have it in black and white. Sun does not have a factory in Japan.
Supply Chain Manager
Is Valdis Filks related to amanfromMars??
Not quite sure which actually posts a better argument but amanfromMars actually scores more points for technical knowledge.
The simple fact is the whole Rock development is like a pizza delivery service. You get a fleet of scooters and they zip around delivering more pizzas than one guy in a lorry could. But, what happens when your job is delivering grand pianos? What are you going to do, break the piano into scooter-sized loads and then stick it all back together at the other end? Or use the lorry? Itanium and Power are the lorries, and Xeon (and Opteron) are pretty big pickups. Rock will always be a scooter. Sticking more wheels on the scooter or stapling a load of them together does not make a lorry. Itanium, Power, Xeon and Opteron are flexible platforms that can handle more business applications better than Rock, the real core applications businesses relie on. Rock is better for some edge applications, the pizzas of the business world.
And Ash has already slapped me for ridiculing the x4450, the most hyped non-event in x86 server history. Just in case you wandered in late, presumably because you were busy graduating kindergarten, I'll reiterate the main thrust of the argument - Sun made the x4450 because they can't make a 4-socket blade that fits in the same chassis as their 2-socket blades. Blades are better at being denser, having lower power requirements, and better cooling than the equivalent number of racked servers. I don't expect you to believe me as your fantasy world is liable to collapse if you do, so I also don't suggest you read up on blades at Gartner, IDC, or other independent sites.....
facts are always better than fiction
Fact: If Sun does not extend the OEM agreement with Fujitsu they will not be able to sell the US64 quad core systems in 2009. Another end of life product for Sun
Fact: Being able to put 10 blades in a 10U enclosure does not count as blades. They are horizontal 1u servers.
Fact: Eight SPANCII cores on a chip with almost no cache/core is good for benchmarks, but not for dynamic workloads, or virtualization.
Fact: T2 is the most expensive chip for software licensing. Six Oracle licenses/chip.
The crowd cheered when T1 was announce at 2 licenses and is now in shock and awe.
Fact: Never compare the T2 with a scalable system/architecture with more than one chip
Fact: The T2 chip uses 120W up from T1 95W. This is why we have a ridiculous metric of Watts/Thread. The T2 also requires incredible amounts of memory compared to other architectures, which is why you have to buy 64GB for a one chip/8core system.
Fact: Montecito and Montvale are both only 104Watts.
Fact: The T2 chip is not 89.6GHz, and I have no idea how that is not considered lying.
Fact: LDOMs are not virtualization, they are partitioning of threads
Fact: The ROCK chip systems above 8 sockets have been canceled, so Sun has exited the >8 socket space forever.
Fact: The Road to Redmond is through Sunnyvale
Fact: SUNW is down to $4.80, once you get past the marketing antics
Fact: its after 3am in London
Valdis from Mars, no but close.
This could be a first scoop for The Register, yes I am an alien. But maybe I am not allowed to call myself that for politically correct reasons, my alien status has been revoked since the country I live in is part of the EU now. I did live in a large EU financial metropolis and work in the middle of a large concentration of IT, where most of the rest of that country looked upon those city dwellers as aliens. And I believe still do.
I also now live closer to the northern lights than most of the population of europe. Apparently where the "blond" aliens have integrated into the population. But I have no starship, just a SAAB that runs of alcohol (serious clue here). When I drink my cars fuel, I do believe I have my own starship.
Well we have had some good chip discussions. A quick wrap up for people from another planet could be that Itanium is a truck nowadays, not a sinking ship, Sun's Rock chip will be edible in the form of a pizza and people still believe that power per thread is not important.
I love the computer industry.
Summary, get yourself a pizza, buy a new cooling device to support the old hot CPU designs and be careful of aliens.
All this goes to show......
There are 5 types of lie in life.
Lies, damn lies, statistics, roadmaps and benchmarks!
RE: All this goes to show......
You forgot to include anything a salsman says before the PO goes through!
"If Sun does not extend the OEM agreement with Fujitsu they will not be able to sell the US64 quad core systems in 2009."
I do not believe this is correct. As I had the deal explained to me, once the agreement ends, both vendors can bring out new products which are not part of the "APL" alliance. But those products currently under the agreement will continue to be sold until an agreed to EOL.
What matters is when Fujitsu's "Jupiter" chips arrive. If they arrive prior to the ending of the agreement, they are covered.
But I also heard the agreement always covered the Jupiter based systems.
Fict versus faction
Good lord, AC, do you write the HP "Rilly Rilly True Stories" ?
Fict: "If Sun does not extend the OEM agreement with Fujitsu they will not be able to sell the US64 quad core systems in 2009. Another end of life product for Sun"
Yes, and if Intel finally decides to pull the plug on Itanium, HP is totally hosed. I truly doubt either will happen, but they make nice spooky stories around the campfire, don't they?
Fict: "Being able to put 10 blades in a 10U enclosure does not count as blades. They are horizontal 1u servers."
Hah? A blade is a blade unless it isn't?
Fict: "Eight SPANCII cores on a chip with almost no cache/core is good for benchmarks, but not for dynamic workloads, or virtualization."
Yeah, having good benchmarks really sucks. And dynamic workloads and virtualization are precluded why?
Fict: "Never compare the T2 with a scalable system/architecture with more than one chip"
That's not a fact; it's an admonition. And a weird one at that. T2s work great in many circumstances where you might use a scalable systems architecture as well. And when it doesn't? Well, use a scalable systems architecture. Horses for courses, as you say on that side of the pond.
Fict: "The T2 chip uses 120W up from T1 95W. This is why we have a ridiculous metric of Watts/Thread. The T2 also requires incredible amounts of memory compared to other architectures, which is why you have to buy 64GB for a one chip/8core system."
Again: hah? It uses more power because it does more: you know, 64 versus 32 threads? It uses more memory because it *supports more threads* and thus *takes on more workloads*. It requires the *same amount of memory per workload* as any other chip. Chip designs generally have no impact on memory footprint, broadly speaking. Certainly there'd be no difference within the same instruction set architecture.
Watts/thread was used as a metric from the very first with T1, so why is this somehow a whitewash for T2?
Fict: "Montecito and Montvale are both only 104Watts."
...delighting sheer dozens of Itanium customers.
Fact: "The T2 chip is not 89.6GHz, and I have no idea how that is not considered lying."
It's a REALLY STUPID metric. It's not lying, but it's really odious logic. Surprise! I almost agree with you.
Fict: "LDOMs are not virtualization, they are partitioning of threads"
I think it was the prime minister of France that once defended one of their country's nuke tests by saying, "it was not a bomb, it was a device that exploded."
If I can run up to 64 separate OS instances on a single T2 chip and dynamically shift allocation of cores and threads amongst those OS instances, how is that anything but virtualization?
Fict: "The ROCK chip systems above 8 sockets have been canceled, so Sun has exited the >8 socket space forever."
...and the Itanium processor now takes over the world. You win! Just ask IDC!
Fict: "The Road to Redmond is through Sunnyvale"
The wind is in the buffalo. John has a long mustache. My hovercraft is full of eels.
Fict: "SUNW is down to $4.80, once you get past the marketing antics"
Oooo, so close. First off, if you'd just said Sun's stock price was down, you'd have been in the paint. But calling it marketing antics just turns it into more bullshit. Why? Because using share price as a standalone metric is meaningless. Overall market value is all that matters.
Sun split their stock *three* times after the bloom was off the boom, so in reality, their stock price should be in the $36-37 range. But really, meaningless.
Your other gaffe, of course, was missing the opportunity to poke fun at the ticker change to JAVA, because, hey, Sun sucks and HP roolz. I mean: Gwen Stefani!
"LDOMs are not virtualization, they are partitioning of threads"
Huh? Why do you get to define what is, and what is not, virtualization.
Okay, I say VMware and IBM Micropartitions are not virtualization, they are partitioning of processor cycles.
LDOMs are almost identical to IBM's pSeries LPARs, except they use logical CPUs (threads) rather than physical CPU cores. No, they are not like IBM's CPU percentage based Micropartitions, but LDOMs do have a very similar shared I/O system to IBM's Virtual I/O Server.
When you have 32 or 64 threads per processor, why not use threads as your atomic unit rather than time-shared percentages? Also, keeping threads within a partition means the thread context (registers, caches, etc.) are preserved. When you time-slice a single thread context, you have to flush caches. This means much lower overhead, and more predictable response.
- Geek's Guide to Britain INSIDE GCHQ: Welcome to Cheltenham's cottage industry
- Analysis Spam and the Byzantine Empire: How Bitcoin tech REALLY works
- Game Theory Is the next-gen console war already One?
- VIDEO Herschel Space Observatory spots galaxies merging
- Apple cored: Samsung sells 10 million Galaxy S4 in a month