Sun christens once and future Supernovas

Last week, Sun Microsystems confirmed it's on track to ship its "Supernova" servers based on its "Rock" UltraSparc-RK processors before the end of 2009. And now the company has coughed up a few extra details. A link on the OpenSolaris site gives the product names for two out of what could ultimately be three or maybe four …

[Anonymous Coward]

Sun Rock is planned for end of 2009 but no systems

Sun's comment of end of 2009 is the chip announce not systems. This really bothered me when Intel did this with Itanium and now Sun is doing the same thing. How stupid is it to announce a chip when you are the only one who is putting it into systems?

Intel has quietly finally moved Tukwila chip announce to 2nd half of 2009 because of DDR3 support. The initial Tukwila's could not compete with Nehalem and since other than HP most sales are 4 socket all the alliance members would dump Itanium. As it is SGI will not admit if they are doing Tukwila or not.

This recession is going to be hard on Rock and Itanium. The economics of ROI can no longer be ignored.

[Anonymous Coward]

@ AC

"Sun's comment of end of 2009 is the chip announce not systems."

Cite your source, or STFU and GTFO.

Thanks!

[Anonymous Coward]

Source? ask the Dice man

I also heard that the instruction and data cache between the micro cores is a major problem because the pairs of micro-cores that share an instruction core share a data cache with a different micro-core. The cache sharing is trashing between the micro cores. Its bad enough there is only 2mb of cache on the chip vs. 30mb for Tukwila. So depending on whether you call it a 4 core or 16 micro-core chip it is still cache starved.

[Anonymous Coward]

Source

"I also heard that..." <snip>

You fail.

[Anonymous Coward]

DOes this factor in Sun's job cuts?

One wonders if the various internal teams involved in these products have factored in the current round of job cuts at Sun... I wouldn't be too surprised to see a littel slippage in these dates over the next 2-3 months as teams deal with the reality of losing 10-20% of their resource.

[deadbeef]

@ AC

"So depending on whether you call it a 4 core or 16 micro-core chip it is still cache starved."

Then you obviously don't understand how Rock works. Read up on the hardware scout.

[Matt Bryant]

A couple of codenames != a release date.

Or even confirmation of life. Before all the Sunshiners get too excited and start ululating, there were plenty of codenames for UltraSPANCedV servers before they got canned.

"....The cache sharing is trashing between the micro cores. Its bad enough there is only 2mb of cache on the chip vs. 30mb for Tukwila. So depending on whether you call it a 4 core or 16 micro-core chip it is still cache starved....." Lol, as predicted, feeble cores without the cache to go round. Maybe if Rock ever does make the light of day we can stop talking about Sun Vapourware™ and start referring to Rock as Weinerware™.

[Zen Zurt]

A word from the Pedants' Guild

Shouldn't that be "Supernovae"?

[Anonymous Coward]

http://www.geocities.com/sunsrockchip/

Looks like more info here

http://www.geocities.com/sunsrockchip/

[blargh]

Supernova?

Isn't a supernova the last big explosion that causes a Sun to go extinct?

Maybe the project name is prophetic...

[Anonymous Coward]

Re: but no systems

That's nonsense. Go put your FUD somewhere else HP-Boy. My company was just recently approached about possibly receiving some beta supernova units. So, a couple of questions for you FUD-Fucks:

1. Is Sun now giving out beta chips and not beta systems?

2. Where are you getting your information or are you really just making it all up like it appears?

Also, Intel is late because of DDR-3? Everyone's known they were late way before they just now announce that the issue is DDR-3. They're already 2 years late! Whatever happened to 2007?

[Anonymous Coward]

Itanium delay is official

http://www.informationweek.com/news/hardware/processors/showArticle.jhtml?articleID=213001916

Any words of wisdoom Matt B?

Curious how the 4 socket Nehalem and 4 socket Tukwila will compare....seems the delay is because Nehalem is better.

[Anonymous Coward]

@ Geocities Sunrock idiot....

At whoever is posting (and probably created) the link to the Geocities sunidiot page with the worlds worst html coding on it, can you please update it?

Let me compliment you on the amazing complex form of hyper tags you did (ie: just stick a load of text at the bottom & hope the search engines pick it up). Interesting it references Matt Bryant in the same text chunk?!?

Any comment HP fan boy? Interesting almost every link reference in google is reg related & Matt is normally posting just before hand. Co-incedence huh?

Anyway, to speculate Rock has 4 CPU cores when Sun are producing 8 cores on T1 & 2. Does that actually follow any normal logic?

Making a prediction of 500w when a T2 CPU is 95w? Again, any brain cells in use when writing that sentence?

Credit to the amazing quality cartoon edit, it "almost" looks (to a 5 year old) as though you drew the cartoons yourself but Niagara is out now and rock is still going. Your not going to update it and show the world how relevant & cutting edge your page is?

Noo, of course not, lets spread FUD through boredom. I wonder if this is the worlds worst attempt @ guerilla market disruption.

Sunsrockidiot seems a more apt.....

[Anonymous Coward]

Details, details

This took about a minute of web searching...(so I present it for those that need spoon-feeding)

http://www.opensparc.net/pubs/preszo/08/RockHotChips.pdf

16 cores and 16 fgus in 4 clusters (see page 2)

problem in first silicon (ROCK 1.0) was that scout threading simply did not work. (this is one of the major reasons why ROCK is so late, the other's being it was just buggy as hell.)

On page 3 you'll see a reference to ROCK 2.0 (says in lab today). Anytime SUN goes from 1.0 to

2.0 means a major mask changes (all the way down to the poly layer for bug fixes. (A 1.0 -> 1.1 spin is usually a minor mask change for something that can be fixed in 1 or more of the metal layers.) Pay close attention to page 10, From this it would appear, and I caution using the word appear that they have scout threading working in ROCK 2.0...

[Matt Bryant]

RE: @ Geocities Sunrock idiot.... & Details, details

"Itanium delay is official...." That's old news, everyone knew the DDR3 addition was coming. Well, obviously everyone except you. Besides, this shows the advantages of the new busses - Tukzilla will hit the ground running with fatter pipes between real cores, with much larger cache, and with faster bus access to faster memory than Rock even if Rock ever does make it out the door.

".....Any comment HP fan boy? Interesting almost every link reference in google is reg related & Matt is normally posting just before hand. Co-incedence huh?...." Yup, coincidence, Mr Paranoid. I started with html when you had to write it by hand, and whilst I find the site amusing I must admit the lack of style is much like a Sun product, and far below the standards I would set myself.

"....Curious how the 4 socket Nehalem and 4 socket Tukwila will compare....." Well, here's the prediction - Nehalem will help Intel continue dominating the Wintel/Lintel market and is probably the only chip with which Sun will make any money, whilst Tukzilla will help hp keep dominating the high-end where Sun will continue to lose money and marketshare (by the way, that's actually my summary of a Forester prediction). I'm actually keen to see what Intel do with Atom as a multi-core Atom would make for an interesting low-power blade chip and also make Niagara completely irrellevant as well as rediculously overpriced.

"...16 cores and 16 fgus in 4 clusters...." So hold on a sec, you're saying the weiner cores have to be clustered together in quads to actually work? So it's really a four-core with four thread-engines per core? What are the overheads of that clustering? And all that still with only 2MB of cache.

"....appear that they have scout threading working in ROCK 2.0..." Not exactly a ringing endorsement, and still no statement from Sun to reassure the investors and customers, but then that would imply they don't think they've fixed it. And what about the crux of the whole design, the transactional memory?

[Anonymous Coward]

WOW 250Watts

I would hate to see what Rock 1.0 was....it might have actually been 500W.

The T2 is 95W. that certainly gives credence to the rumor that Rock will be dumped for Niagara3.

I think I have seen that Cartoon before.....somewhere...what goes around comes around i guess.

I hear Gartner gave it's rating for Rock at only 30% chance of coming to market.

[No Use for A Name]

all that sun fud is getting silly

ever since ashley vance left for nyt the reg's reporting on sun has been going downhill. gone are those well researched and thoughtful pieces on sun; obviously benefiting by some good inside sources in the company.

and now?! when it comes to sun it's just the usual hack-job of an article, pulling together some bits and pieces that showed up elsewhere and taps into the current meme of 'sun is the new dec'...

and don't even get me started on the commenters here!

it's pretty laughable imho. while sun's numbers aren't looking to good its a pretty solid company actually creating revenue and whats more important their product folio looks well positioned given the challenges coming up. governments are beginning to pour tons of money into IT and they require open source for various reasons.

last not least - as someone who works in a shop that bought a couple of 5120/40's recently we are quite happy with our machines. we have been informally briefed on supernova by our rep and I'd say this looks like a solid offering. everybody I know in sparc-land holds of on new purchases until these come to market.

sun may well be the new apple: predicted dead over and over. I'd say wait and see.

[Matt Bryant]

RE: No Use for A Name

"ever since ashley vance left for nyt the reg's reporting on sun has been going downhill....." Oh, you miss his unquestioning marticles? Don't worry, I hear the NY Times isn't doing so well, so Ash may be back before too long.

"....and now?! when it comes to sun it's just the usual hack-job of an article...." Oh, you really do miss Ash's unquestioning marticles, don't you.

"....and don't even get me started on the commenters here!..." So you don't like the articles, and you don't like the forums, then why don't you just go read some other site?

"....it's pretty laughable imho. while sun's numbers aren't looking to good its a pretty solid company actually creating revenue and whats more important their product folio looks well positioned given the challenges coming up...." At least we know you're not a professional analyst then. Not wanting to put too much of a dampener on you when you're obviously pining so mcuh for ol' Ash, but one quick question - if Sun is doing so well and is so well positioned, why is it's market cap so poor?

"....sun may well be the new apple: predicted dead over and over...." The big difference is Apple had clever marketing and Apple diversified into new markets with products like the iPod and the iPhone, which have generated both profits and an image of a hip company going somewhere. Compared to this, Sun has stuffed up delivery, muddied the roadmaps and made bizarre purchases at silly prices without gaining any market advantage. When are you Sunshiners going to get tired of trying to make out you're so like Apple when the truth is Sun has innovated less than DEC did.

[Anonymous Coward]

Spoiler Here

Hate to spoil the party but has there ever been a new chip that was not late????

[Mark Spooner]

Re: "supernovae"

They probably took off the "e" at the end so they could copyright the name

[Anonymous Coward]

Re: matt's rant

"crux of the whole design, the transactional memory?"

Huh? Since when was TM the crux of the whole design? From what I've seen everywhere, including NDA's from Sun, TM was only meant to be a small part of the design that may or may not show merit. Since Sun was the only one doing it, and TM did not take up much if any space on silicon, they put it in there. The jury's out on whether TM actually works in general, let alone in this one case.

The crux of the whole design, as I 've heard over an over again, was to get around the slow speed of main memory (scout threads and threading in general) and to save on power and space. 250W for a 16 core system is great! Show me another vendor that could give you 16 cores for less than 250W's. I'll give you a clue, they don't exist.

Come on Matt, take your nonsense elsewhere.

[Anonymous Coward]

They are not full function cores

Rock, Niagara have KISS cores. keep it simple and stupid cores. You cannot simply divide 250 by 16 and do a comparison. The wattage of T2 went up significantly from T1 so Sun went to a Watts/thread marketing campaign. It just shows that Sun thinks customers are stupid. Unfortunately for Sun, customers understand that virtualization is what cuts costs and Sun has poor virtualization offerings which do not work on their whole product line. Will ROCK still partition threads per O/S or will it have a real hypervisor that provides virtualization?

[Jesper Frimann]

RE:They are not full function cores

Well virtualization is not good for business. Going from partitioning your server, like on a M class machine to a shared pool of processor resources, like on POWER systems would mean that the average utilization of boxes sold would jump quite significantly. Hence you would sell fewer, and possible smaller servers to replace old ones.

Sure using containers helps, but that glove doesn't fit all IMHO.

// Jesper

[Matt Bryant]

RE: Re: matt's rant & Jesper Frimann

RE: Re: matt's rant

"....Since when was TM the crux of the whole design?...." Sun have long maintained that TM would allow them to keep all the cores spinning and not just the majority stalled. Scout threads was the other technology that was supposed to help this happen. Without either, Rock promises to suck when running real world heavy-thread apps much in the same way as Niagara has. Sun needs both working optimally to stand a chance of matching yesteryear's Power and Itanium, and the deafening silence from Sun is not going to reassure customers that can see new chips from Intel and IBM on the horizon.

RE: Jesper Frimann

Virtualisation is actually very good for most businesses. It allows you to consolidate a lot of under-utilised rack servers with direct-attach storage (also usually under-utilised) into SAN-attached virtual instances. This saves on floorspace, power, administration, storage and server hardware costs. Management think VMware is the best thing since that Kylie underwear ad, so much so my big problem is explaining why we don't want to put all servers into virtual instances. When we wanted to introduce Integrity Virtual Machines into production all I had to do was tell them it was like VMware but for Itanium and they were all for it.

[Anonymous Coward]

Re: Matt's rant on Matt

"Sun have long maintained that TM would allow them to keep all the cores spinning and not just the majority stalled."

I've read Sun's lab report on TM and they specifically stated that it was not expected to improve ROCKs performance on most workloads and possibly a bit more on others where TM is utilized such as DB's. As for Scout threads, they have heavily pumped these up as important to ROCK. The NDA's that I've seen are extremely impressive, but are based on expected performance increases, not real (seeing how the product is not released yet...).

Of course, since no one else has TM, the fact that it is present is not a bad thing and could only be a good thing.

You're full of it as usual Matt. Are you sure you don't work for HP Matt? You're so good at confusing the facts and making unsubstantiated comments.

[Bill]

Re:They are not full function cores

That may be true for Niagara, but it's not true for ROCK. That's the whole point of ROCK. Of course, going core to core, I bet you're right that the ROCK cores will be slower than comparable MHz Cores. However, the fact still remains that who really cares about core performance in servers, except when dealing with Oracle? Oracle's not the only game in town. All I know is that total throughput will obliterate the competition. Why do you think that Niagara is growing over 30% quarter to quarter and that's with the single thread penalty. Just think how well these boxes will sell when you get rid of that single thread penalty... The only ones that don't see the benefit of this idea are HP and IBM as they have nothing to compete against it. Even Intel's seen the light.

[Jesper Frimann]

RE:Matt Bryant

Well I agree with you on TM. They need something to address, one of the main weaknesses of the Niagara based servers, and I would also presume ROCK, their locking problems.

On your comment to my post. I think that my post was a little unclear. What I meant was that Virtualization is bad for the Server vendors Hardware revenue.

If you take a partitioned physical server running at around 20% seen from a whole physical box perspective, and upgrade it to a new virtualized server where perhaps the cores are twice as fast and runs at 60% utilization, due to reuse of unused CPU resources, then you will actually only need 1/6th of the number of cores to run the same workload, compared to before. Hence you cut the number of resources you need and you will normally also be able to go down one or two sizes in type of server. Eg from a 32 way (highend) to a 8 way server (entry level).

This will mean that the server vendor that is selling you the server will see a significant drop in revenue, compared to if it had sold you a new box that would run at the same utilization.

And to be fair to SUN, they are using a combination semi fine grained partitioning on their Niagara boxes and containers to achive a higher utilization on the servers they sell. So they have some of the virtualization effect. But no where to the same degree as for example people using VMware or POWERVM are.

// Jesper

[Anonymous Coward]

RE: jesper

"But no where to the same degree as for example people using VMware or POWERVM are."

So which single socket 16 domain systems does IBM sell? I haven't heard of them yet. IBM and HP are stuck with domaining on the high-end, while Sun is doing domaining all the way from a low end niagara to a highend m9000. Granted, the m9000 does not have the granularity that a powervm has, but it does have Containers with Solaris. Also, since Solaris run's on X64, you can still run VMware or even xVM Server from Sun. IBM was pretty impressive with their early adoption of VM technologies on Power, but I doubt that they will ever push those capabilities down to lower end servers and the licensing for IBM's domaining is extremely expensive.

[Matt Bryant]

RE: Re: Matt's rant on Matt & Re:They are not full function cores

RE: Re: Matt's rant on Matt

"....I've read Sun's lab report on TM and they specifically stated that it was not expected to improve ROCKs performance on most workloads and possibly a bit more on others where TM is utilized such as DB's...." Sounds like not just a Sunshiner but a Sun employee to me. What you may be unaware of is what your salesforce and channel partners have been doing for the last two years, and that is hyping Rock beyond belief. Whilst your lab reports may hold the truth, the messgae going out to potential customers (and here I'm going on a mix of what I have received and a straw poll of sys admins and IT managers in other comapnies) is along the lines of Rock being a super-chip capable of running any existing app twice as fast as any other chip, transactional memory being some wundertech which will keep all the cores spinning, scout threads will make cache hits comparable or better then Intel's, etc, etc. As Rock has got later and later, and as Niagara has not lived up to the hype, these messages of Rock's super-powers have become even more desperate and - frankly - rediculous.

"....Are you sure you don't work for HP Matt?..." Geez, you want it in semaphore or morse code!?!?! Losing count of the times I've said this. I do not work for hp. Hp is not my employer. I do not get paid by hp to do work for them. Got that? I doubt it, it's so much easier for you to paint anyone dissenting with your views as a competitior's employee rather than face the facts that customers out there think the Sun is setting.

RE: Re:They are not full function cores

"That may be true for Niagara, but it's not true for ROCK....." Hold on a sec, Bill, only a few threads ago you were saying the niagara cores were not weiner cores, make you rmind up. But before you do, please go and compare the transistor count for Rock and the current available chips and try and explain how sixteen real cores can have so few transistors?

"....All I know is that total throughput will obliterate the competition...." Then you know nothing, because there are no Rock chips anywhere other than some lab silicon which may not even be the final cut, and I doubt if it's being fed tests such as running a major billing system with real data. So all you have is your blind faith beliefs, and no empiical data. On what I've seen so far, it looks like Rock will be hard pressed to beat SPARC64 VI let alone the real competition of Power and Itanium.

"...Why do you think that Niagara is growing over 30% quarter to quarter and that's with the single thread penalty...." Because Niagara is just eating Sun's old low-end, low-margin UltraSPARCIIi and older business, and it's not even mopping up all of that, a lot is being lost to other vendor's x64 kit. Even if it were mopping up all the old Sun low-end it still wouldn't generate enough cash to keep Sun in business, and the same for Galaxy. 30% of a small amount is still too little revenue to impress the analysts, hence the Sun-is-junk market cap.

"....The only ones that don't see the benefit of this idea are HP and IBM as they have nothing to compete against it...." Oh yeah, hp and IBM look so miserable with their massive market share advantages and healthy profits. They seem to be doing quite fine without Niagara or the Rock vapourware. Why else do you think Sun are so desperate to get hp, IBM and Dell signed up as Slowairs on x64 resellers - it's because Sun know they are exiting the server bizz soon, and they will then be reliant on other vendors shipping Slowarisx86 licences t survive.