c0t0d0s0.org

We´ve updated the X4450 2U 4 Socket Rack Server and the 4 Socket Blade Server X6450 with the new Xeon 7400 CPU. The 7400 series is the new 6 core version of the Xeon.

It would be interesting to see, how much performance do you get from the two additional cores in realistic workloads, as the CPU still uses the 1066 MT/s Front Side Bus technology, but i assume that much of the issues will be mitigated by the 16 MB cache of the 7400 series (the Xeon series even starts to catch up in cache size with the Itanic). Nevertheless: Sun is able to offer you a x86 24-core system on just two rack units.

In a series of videos David Stewart (an engineering manager at Intel's Open Source Technology Center) describes the work Intel put into Opensolaris:

• Intel and Opensolaris
• Powermanagement
• Performance
• Fault Management
• Virtualisation
• I/O-Accelration
• Dunnington(the upcoming hexacore Xeon)

These short episodes are really worth a view.

There are many people, who don´t like the eLOM of our Intel based x86 servers and want the iLOM of the Opteron servers instead. The cure is near: The eLOM to iLOM migration guide already appeared at docs.sun.com

This is the english translation of an article i wrote to answer an strange blog article in another german blog. While answering it, i found a case of "History Repeating".

In the actual installment of the Prozessorgeflüster (a reoccuring CPU technology article in the german computer magazine c´t) Andreas Stiller discusses the the first facts about the new manycore processor "Larrabee" from Intel. Stiller states in this article, that this processor was discussed to release with with 16 to 24 cores, but will release with 32 cores when it appears in 2009. To the surprise of many experts, the cores are well known. The cores are nothing more than Pentium P54C cores. Well, the P54C was announced 1994. This would be similar to a Niagara T1 on the basis of the SuperSPARC II, which was announced 1994 as well.

Annother interesting fact reported in the article: The Larabee xPU will use roundabout 300 watts. That´s much much more than a UltraSPARC T1 CPU. I know, Larabee is a GPU at start, but do you really believe that a manycore GPU with x86 compatible commands will stay on the graphics card for long time?

There is a small irony at this story. Sun learned many things about designing multicores when Sun developed the MAJC-5200 CPU and used it on the XVR1000 and XVR4000. The XVR4000 was the graphic card, that didn´t used something earthly like a PCI-bus, you plugged it directly onto the Fireplane Interconnect of the V880 instead of the CPU. Sun learned so much about multicore that some of the findings will reappear only in future incarnations like the Rock CPU (speculative multithreading e.g.)

The funny (and "history repeating" part): Intel start s to use it´s first manycore design on a graphics card as well. Like we did ... in 2002.

Der werte Herr Siering hat ja in seinem Blog angemerkt, das ich wieder mal Pseudoargumente fuer CMT finde. Ich habe mich dort in einem kurzen Kommentar ausgelassen, nichts ahnend welche Perle als Gegenargument mir Herr Stiller vom Prozessorgefluester einige Tage spaeter zuspielen sollte. Herr Siering ist ja der Meinung "CMT? Klar. Nur nicht auf dem Nieveau von 2001!" was in seiner Argumentation schon daher nicht treffend ist,, das das Gerücht die Cores der T1/T2 CPU seien nur UltraSPARC II Cores eben nur ein solches ist, ein Gerücht. Was zugegebenermassen von unseren verehrten Marktbegleitern oefters gestreut wird und in wenig Sun-nahen Kreisen leider auch zu oft geglaubt wird ...

Im aktuellen Prozessorgefluester werden die ersten Daten des ersten Manycoreprozessors von Intel mit dem Namen Larrabee diskutiert:

[..]ist Larrabee gedacht, ein Prozessor, der bisher mit 16 bis 24 Kernen gehandelt wurde, wahrscheinlich aber gleich mit 32 Kernen im nächsten Jahr debütieren wird – und zwar wie inzwischen durchdrang zur allgemeinen Überraschung wohl mit genau den gut bekannten Pentium-Kernen: dem Pentium P54C.

Hmmm ...P54C ... hmmm .... vorgestellt wurde dieser 1994. Das wäre ungefähr vergleichbar, wenn wir Niagara T2 auf Basis von SuperSPARC II (Super, nicht Ultra) gebau haetten. Der Core wurde naemlich im selben Jahre vorgestellt

Vermutlich bezieht sich der Artikel im Intel@Research blog, den ich vor einigen Tagen verlinkt habe auf diese xPU, die Eigenschaften eines solchen Prozessors wuerden zu jenem dort geforderten Herangehen an Softwareentwicklung passen.

Interessant: Insgesamt soll diese xPU eine Gesamtstromaufnahme von 300 Watt haben (mithin das 3 bis 2,5 fache einer T2-CPU und fast das 4 fache einer T1-CPU). Ich weiss Larabee ist erstmal eine GPU ... aber glaubt jemand ernsthaft, das eine GPU mit x86 kompatiblen Befehlssatz lange auf der Graphikkarte bleibt?

Es gibt hier uebrigens noch eine kleine Ironie der Geschichte. Sun hat damaligerzeit viel über Multicores gelernt, als wir die MAJC-5200 CPU gebaut haben und dann auf den XVR-1000 und XVR4000 eingesetzt haben. Die XVR4000 war jene Graphikkarte, die sich nicht mit sowas laeppischen wie einem PCI-Bus abgegeben hat, sondern direkt anstatt eines Prozessorsboards auf den Fireplane-Interconnect der V880 gestoepselt wurde. Sun lernte damals soviel, das manche Dinge erst mit Rock wieder in CPU fliessen werden (Speculative Multithreading beispielsweise). Interessant ist nun, das Intel ein solches Design jetzt auch zuerst auf einer Graphikkarte einsetzt. Nur: MAJC und die XVR4000 wurden September 2002 announced

Somit verbleibe ich mit einem Gruss an Herrn Sierung und moechte mit einer genauso reisserischen Schlussformel enden: "CMT? Klar. Nur nicht auf dem Niveau von 1994!"

The free lunch of the ever increasing per-core performance will come to an end. Ars Technica writes in Intel: an expensive many-core future is ahead of us:

Intel has bad news for software developers. It's been hinted at already, but now the company has stated explicitly: it's not enough for software developers to be targeting dual, quad, or eight cores. No, the future holds tens, hundreds, or thousands of cores, and developers are going to have to bite the bullet and write programs that will scale to such systems.

With their article they refer to a blog post on the Research@Sun blog: Unwelcome Advice. So ... please no more questions if CMT is the right solution. It´s the only solution when even "throw-tremendous-amount-of-money-to-frequency-increases" Intel doesn´t think that the other way is future-proof.

Interesting rumour at EEtimes - Updated: Intel slips in 'Nehalem' launch, says analyst :

Our checks suggest Intel’s Nehalem server parts will initially launch in single-socket only, which we believe will be viewed by customers and investors as a delayed ramp of the company’s second generation 45nm server architecture due out in late 2H ’08.

Intel denies this delay, but i understand the logic behind the analysts opinion. When a server proc isn´t available for multi-proc (unlike it´s predecessors) this looks somewhat like a half-launch.

x86 got 30 years old now. This is interesting as Mom and Dad tried to kill this baby several times. The kudos for keeping this architecture afloat must be credited to a completly different company: AMD. Imagine the market, if AMD didn´t introduced 64 bit to the x86 world? Okay, SPARC would have it much easier. But this is a different story. Enterprise computing by Intel would be Itanium. Not a desktop processor kept afloat by heavy wizardry. And they tried to kill off x86 as early 1982 with iAPX 432 (does anybody remember this proc?

Albeit it´s a cash cow for Intel, they are not really responsible for the success. I really thing x86 is the predominant architecture by pure accident. How would the market look, if Linus bought a Amiga with 68040 or an early Mac with PowerPC instead of the 386? What would have happened to x86 without AMD decision do develop the Sledgehammer architecture? Back in 1992 nobody thought of x86 as the dominant gaming plattform. Today it´s normal to use an x86 compatible processes as a data feeding device for graphic cards.

Sometimes the accident is the most powerful tool to shape the market.

When you want a good itanic laugh? In this case i recommend the "by the way"-announcement regarding Intel® Itanium® Processor 9100 series processors. It looks like Intel is so desperate about the software support that they aren´t afraid of really far fetched arguments:

Unlike products from the remaining RISC vendors, the 9100 series continues to offer end-user freedom through a broad choice of software with more than 12,000 applications in production, and flexibility to support multiple operating systems, including Linux, Windows, HP-UX, HP NonStop, HP OpenVMS, z/OS and Solaris/SPARC.

Okay, z/OS is subject of a legal battle at the moment: IBM sued PSI (the manufacturer of an itanium based Mainframe clone) and PSI sued IBM. I assume this will end nowhere soon, but as mainframe buyers tend to be the most conserative and risk avoidant people, this isn´t a good business card for going into real deployment. And the Solaris/SPARC stuff. They use this Transitive stuff for this. I wrote about the problems of such a solution not long ago. Before trying to run an application for Solaris on SPARC on Transitive on Suse Linux on VMware on x86 you should try to run the application on Solaris 10 on Sparc. When when you have some problems by hardcoded dependencies(shame on the programmers) , you should try the Solaris 8 Containers for Solaris 10. From my view, Transitive is a solution for a non-existant problem, you have the binary compatibility under Solaris and for hard nuts you have Project Etude or it´s downloadable variant.

So, dear Intel, is there nothing more convincing for buying Itanium. Oh, yeah, i forgot ... next year you will show us the Itanium variant leaving all others processor in the dust ... as every year ... really ...

Brian Dipert writes in The Intel Developer Forum: Where In The World Is Itanium?

I smell death. Or, if you prefer, I see dead processors.

Scientia writes in his blog about a little-reported fact: Albeit announced in 2006, a Core2Duo with more than 3 GHz is still not available.

The Register reports about announcement of quad and dual socket Intel Blades for our Modular Server chassis.

The four-socket blade server will fit into Sun's existing Blade 8000 chassis and should ship in the second half of this year. The system will be based on the four-core "Tigerton" version of Xeon from Intel and will support up to 128GB of memory.

I would prefer the Opteron Blades (an surely when Barcelona is out) out of several reasons like nested pages, better overall architecture and stuff like that, but your milage with your respective workload may vary. It´s certainly a good thing that we offer the Intel Blades for such workloads.

Sometimes journalists are really narrowminded: Hypeing an technology , that won´t deliver products the next several years. Of course, massive multicores are the future. But not in the touted 5 years. Last year, this year, next year. Niagara, Niagara II, Rock, Victoria Falls. You can speak to your friendly Sun Sales rep and order a Niagara-based system. Virtualisation? Already there. In Solaris.

Dear Mr. Windley, look over the edge of your plate before sounding the fanfares for a useless technology demonstrator when others have already buyable products out there. And don´t try to tell your readers, that Intel now solved the problems of the datacenter owners.

Nice comment about 80 core Intel at the register:

Terry Shannon will be turning in his grave. All X86 processors are really Alpha chips now. Pity it took AMD and Intel 10 years to catch up. And it seems DEC was quite right when it introduced the Alpha processor and said it would be good until the year 2025.

In my opinion, the Alphacide was the baddest thing that happened to the computer industry. The computing industry would look a whole different now with it.

PS: Now it´s up to Sun with Rock/Niagara/NiagaraII to bring some new concepts into computing.

Interesting article about a different perspective to the Suntel-Deal: The forces that forged the Suntel alliance:

So the SIs – the firms with both specific market knowledge and increasingly trusted brands – are becoming the front line for hardware and software vendors like Intel and Sun, because they are the companies that users turn to first.

In my personal opinion the x86 business has to be an business for system integrators or as an part of a project, because the low margins at 1 or 2 processor systems doesn´t let much room for presales. SI can cover the presales costs in bigger solutions but when you only sell the box, you must keep human work out of the equation as long as possible. As Intel has a huge influence to the SI ecosphere, it has a positive effect for Sun to have Intel in the boat for sure.