Oracle and Fujitsu hook up on Sparc servers

Nahh.. no Venus IMHO, new M-series machines will IMHO pack Yellowstone falls

Well IMHO, the new M series machine won't be based upon SPARC64 processors.

If you look at the good old roadmaps from SUN, then the new M-Series servers won't be based upon the SPARC64 based machine, but rather the new 'beefier' CMT processors.

It looks like the CMT processors will branch into two different types, one with lots of cores and lot of threads, a kind of throughput computing cores, and then a more beefier one with better single threaded throughput.

If you look at this roadmap:

http://regmedia.co.uk/2009/09/11/sun_sparc_roadmap.jpg

And compare it to the recent Oracle roadmap:

http://regmedia.co.uk/2010/12/03/oracle_sparc_roadmap.jpg

Then it actually fits with:

T-model in 2010 is Rainbow Falls based.

M-moel in 2010 is Jupiter-E based

T-model in 2011 is Yosimite falls based.

M-model in 2012 is Yellowstone falls based.

T-model in 2012/3 is Cascade falls based.

What just doesn't match is the number of sockets, but hey.. these are foils made by marketing so.

So IMHO SPARC64 is dead, and the future strategic sparc processors are the CMT.

// Jesper

RE: Giving credit to Oracle (for its SPARC roadmap) where credit is due

Aw, that was just so cute and naive, I couldn't resist the temptaion to do a little troll-bashing!

"......Oracles reputation, and of course LJE's reputation, making this public disclosure on future product plans is quite aggressive to say the least....." What reputation? You are right that this is about reputation, but the problem for Larry is us customers perceive Oracle as having a bad server rep. Oracle has zero previous server building experience, they are just a software company, all they did was inherit the big pile of fail from Sun's server bizz (and lost a large chunk of the skilled engineers in the process). That came with a history of broken Sun roadmap promises (and Sun rubbishing SPARC64 when they thought they had Rock coming), and many said that Larry needed to stand up on day one of the Sun buy and give a firm, believable roadmap to stand any chance of stemming the defection rate of those old and let-down Sun customers. He didn't, he dithered and waffled about Niagara which was already too weak for the enterprise, hence the need now to try and rebuild customer confidence in Oracle's enterprise server offerings. Larry really needs to have a new top-end offering to fill the hole left by the death of Rock as the CMT offerings have not been able to provide the type of server needed. He had four options - use Power; use Itanium; make a deal with Fudgeitso to extend the very dated SPARC64 design one more time; or just abandon all those Sun shops running old UltraSPARC and M-series to IBM and hp. Given his ego (and that the Japs have paid for most of the new SPARC64 development), it was no surprise that Larry is trying to push the SPARC64 option. Problem is it has taken soooooo long, probably because Fudgeitso weren't keen, and many of the old Sun base have started migrating off SPARC already.

The problem is, even if the customers do decide to try Slowaris-SPARC64 one more time, what's next? A painful shift to a CMT design that has - so far - proven unsuitable to the enterprise apps they run? What if Oracle can't make the CMT option perform to the point where the CMT has some single-threaded performance? For many customers the simpler and safer route will be a switch to proven Power or Itanium (or Linux or Slowaris on Xeon) now.

All this has been predicted, discussed and rubbished by Sunshiners like yourself many times, and just as many times you all said "Niagara is sooooo good we won't need anything else", so you've just all been proven wrong yet again. If CMT was the answer, then Larry wouldn't need another extension to the SPARC64. The good news for you is that Larry has finally put some meat to the plan, but it may be too late. I said previously that the recession gave Larry some breathing space whilst customers delayed refreshing old enterprise kit, but that time is gone and hp and IBM are ramping up their latest enterprise server sales. TBH, IBM can probably afford to be a bit vague on their roadmaps for Power8 as they seem to have more standing with us customers, and Intel and hp have kept plugging the Itanium option many, many years after you Sunshiners repeatedly told us Itanium was dead (how many times have you been proven wrong on that alone!). Larry is the one with a reputation problem, and this deal with Fudgeitso may be too little, too late, to swing the buyers back behind the old Slowaris-SPARC combo again.

/SPAL!

bah

Bah posted literally one minute before I wrote this so bitten by moderator delay once again. Still have good rips on the garbage that SPARC is so won't withdrawl.

RE: Matt Bryant was wrong about dead architectures, let us count the ways!

Whilst it is amusing to point out that you have simply regurgitated a number of posts, you haven't actually proven the "error" in any of them, you have missed the most obvious fact that Sun did fail with their SPARC plans, and that only Oracle's money has allowed the disaster to roll on a bit longer. Please try and deny that whole Sunset thing happened, just for laughs!

".....Matt wrongly predicted ~~~ SPARC is dead...." I suggest you go read the quotes you linked to, you may notice a few chickens coming home to roost. As I predicted, Rock stalled and was killed and Sun had to switch to Fudgeitso's SPARC64 for their top-end servers. Sun sales had to make the embarrassing switch from rubbishing SPARC64 and urging customers to wait for Rock to instead pushing SPARC64. Also as predicted, Niagara was useless at replacing many of the low-end UltraSPARC servers in the market, to the point where Sun were forced to bring out the one-socket, SPARC64-powered M3000. Please try and deny the original Sun plan was that Niagara would face Xeon in the webserving heartland that used to make up 80+% of Sun's Slowaris base, whilst Rock was supposed to slug it out with Itanium and Power in the enterprise arena. Niagara hasn't moved out of the webserving niche because the single-threaded perfromance is still pants, and most of Oracle's existing Slowaris customers are still using apps that require good single-threaded performance. Nothing illustrates the unsuitability of the CMT design for the current industry apps like the fact that Oracle still needs Fudgeitso to produce another generation of SPARC64s.

/SP&L

RE: David

"The Oracle T3 platform over doubles performance over Sun T2 platforms... 400% throughput improvement comes at a cost uplift, but 100% performance increase is a actually a discount from the old platforms."

I guess you mean "The Oracle T3 platform over doubles the price/performance..." right :)=

With regards to T4 (Yosemity falls), then I think you might be wrong, I don't think it is a sure thing that the T4 will offer an aggregated throughput boost.

Now the clock frequency will be 1,5 times faster.

It will have half the number of cores.

The single thread throughput will be up to 3 times faster, which dictates a factor of 2 from something other than clock.

Now that will give the chip a 'best case' throughput of clock increase x other increase / factor of fewer cores = 1,5 x 2 /2= 1.5 times the throughput. But this dictates that the x2 'other increase' is attainable for all cores at the same time. I don't think that will be the case.

But lets see, what actually happens.

Now personally I think it's a great move, if Oracle Fujitsu kind of agree to merge SPARC64 and the CMT processor lines. IMHO it's terrible to have several processor lines running at the same time.

Specially when those processor lines are actually overlapping in products. For example if you look at the tuning settings on the more or less straight forward specintrate2006 benchmark then for the T3-4 and the M8000 the compiler settings are actually quite different, if you look at the compiler settings, and it's also true for the base settings. And these two machines both have 64 cores and are within the same performance range.

Which basically means that if you have to compile code that have to run as good as possible on both platforms, you have to aim somewhat in the middle. That is not an optimal situation. So if Oracle and Fujitsu somehow agrees on merging development teams and start to use one platform, then I think it would be a advantage for them in the long run.

// Jesper

RE: RE: Giving credit to Oracle (for its SPARC roadmap) where credit is due

Looks like Novatose has forgotten about the Sunset too! In case he's forgotten, he got the "Novatose" moniker for his continual hyping of the long-anticipated Rock server lines, even after it became obvious to the rest of us that Rock was a zombie just waiting for a mercy killing.

"....you spelled Solaris, and various other words incorrectly again....." As I have previously reminded you many times, the moniker "Slowaris" was given to Solaris by Sun's own customers due to our unhappiness with SPARC-Solaris's performance. Pretend all you like, that's a simple fact.

"....at least appear to be credible...." Given your previous and hilarious outbursts of Sunshine (e.g., "Rock will not be killed", "Sun is not up for sale"), I'm sure there is no way I can appear credible to such a self-certified Sunshiner! You carry on keeping The Faith, don't let that little old Sunset event or the loss of Prophet Ponytail deflect you, just don't expect the rest of us to suffer from such delusions.

"....It seems Oracle has released several refreshed several lines of their own servers since the Sun acquisition....." Hmmmm, you mentioned a spell-checker, ever think of using the grammar-checker? All Oracle have done is slap new badges on old Sun designs that were already available or already in the works. Oracle has released no server truly of Oracle's own making yet, and it looks like the next gen top-end servers are just going to be more badged Fudgeitso kit.

"....after dumping former HP boxes in some integrated system lines...." Well, Larry did inherit a whole stack of servers no-one wanted to buy, I'm not surprised he'd want to burn through that old Sun stock, especially after hp had done all the tricky engineering and integration for him. It obviously didn't register with you that hp shrugged and carried on being the leading server vendor, whilst Snoreacle is still third tier at best.

".......The Oracle T3 platform over doubles performance over Sun T2 platforms...." Just not in the way customers want. Stop talking the "throughput" drivel, that only matters to webserving, it's irrellevant to the single-threaded apps most customers want to run. A three-year old Xeon server can thrash a T3 server with real applications and real data.

".....suggesting otherwise logically suggests HP has no "top-end offering", because HP stopped making systems with Alpha processors...." Oh dear, just showing your lack of knowledge again! You forgot PA-RISC, which was caning UltraSPANKed for years before Sun finally went to SPARC64. It had already been agreed between Compaq and Intel that Itanium would replace Alpha before hp bought Compaq. Unlike Sun, Compaq and hp saw the fact that traditional RISC CPUs were reaching their design limits and both moved on before they destroyed themselves. Sun presisted with UltraSPARC, failed with Rock, stumbled with Niagara, and saw their company value crash from a 2000 peak market cap of $200bn to less than $4bn. If Oracle hadn't saved them (after IBM and hp both turned down the opportunity), then that would have been the end of SPARC right then. Even then, even Larry admitted his original intent was to buy just Sun's software business but he got stuck with the hardware failure as well.

".....The T4, which is already in the lab...." So, more unproven vapourware arguments. Hey, didn't the Sunshiners assure us Rock was "taped out" right up until it got killed? And if all T4 does is place itself "between the former CoolThreads systems and M systems" then I'm not impressed, it will be slower than current Power or Itanium and probably x64 with the apps us customers actually might want to run on it. It may be finally suitable for replacing the old UltarSPARC IIIi kit still out there on a like-for-like basis, but seeing as I can already consolidate old UltraSPARC onto Lintel blades, at a better than one-to-one ratio and for a lot less money, I'm just not impressed.

/SP&L

RE: @Matt Bryant - chuckle

"......wasn't HP just a printer company until it bough Compaq?...." I suggest you remove the Sunshiner Blinkers (TM) and read a little industry history. There were server and storage products from hp long before the Compaq buy, using Intel's and AMD's x86 chips and hp's own PA-RISC CPUs. They also had a storage line before Sun, management software (in fact, Sun had to resell hp Openview as they didn't have their own management software), PCs, networking products, and of course the printers biz (which continues to make more profit in a quarter than Snoreacle makes from all server products in a year!).

But, this forum is about the refresh of Oracle's server roadmap, do you have anything even mildly related and vaguely technical to post? Go on, have a try, you may surprise yourself.

/SP&L