OpenWorld 2012 This time last year at Oracle's OpenWorld extravaganza, John Fowler, Oracle's executive vice president of hardware, revealed a faster-paced roadmap for Sparc processors that showed the Sparc T5 chip being pulled into late 2012 from its original early 2013 delivery date. It now looks like Oracle needs a little more time to put the spit and polish on the Sparc T5.
Considering that the sixteen-core Sparc T5 was running ahead of schedule this time last year, as Fowler showed in last year's roadmap, it is not fair to call the shipments of Sparc T5 chips in 2013 late. The dates on the December 2010 Sparc roadmap were always a bit vague, so it is hard to pin down when Oracle meant to do the launches.
This is intentionally so. Date obfuscation is similarly practiced by other chip makers – if they even have public roadmaps – because designing, etching, and testing a complex processor with billions of transistors is one of the hardest things in the world to do.
That said, if you line up the 2010 and 2011 roadmaps, there was a chance that the Sparc T5 might have launched before the end of this year. And those of us who like shiny new iron are always happy to see it sooner than later.
With Oracle's techies talking up the Sparc T5 chip at the Hot Chips 24 conference back at the end of August, it was natural enough to expect for Oracle to make the forthcoming chips one of the highlights of this year's OpenWorld event.
This was precisely the pattern for the Sparc T4s that are currently shipping. The chips were revealed at Hot Chips 23 in August 2011and the Sparc T4 systems were announced a few days ahead of OpenWorld 2011, so it was logical to think the pattern might repeat.
I know I certainly thought it would, particularly with the Sparc T5 in test back in October last year. Chip happens.
But when Fowler gave his cloud systems keynote at OpenWorld this morning, alas, there were no new Sparc T series systems. Solaris shops are going to have to wait a bit longer to get their hands on these chips, which are most certainly going to give IBM's Power7+ and Intel's "Poulson" Itanium 9500s a run for their money in the RISC/Unix server racket.
Oracle's latest Sparc T series roadmap
Given that the chip is still on its original target – again, as best as you can tell with the X axis on any of these roadmaps – it now looks like we can expect the Sparc T5 chip a few months into 2013.
As is often the case among server chip makers, optimism sometimes gets ahead of deliveries, and that doesn't change because you don't have your own fab. Taiwan Semiconductor Manufacturing Corp, the fab Oracle uses to etch its Sparc T series of processors, is having issues with the ramp of its 28 nanometer processes.
It stands to reason that any subtle revision in the Sparc T series roadmap might have nothing at all to do with the chip and everything to do with where Oracle is in line to get 28 nanometer capacity, and at what cost. TSMC is using the same processes to etch GPUs for Nvidia, as well as GPUs and some CPUs for Advanced Micro Devices and a number of ARM processor upstarts.
The important thing for Sparc customers is that Oracle remains committed to Sparc and is spending money on chips and systems. The important thing for Oracle is that Sparc T series machines are making money.
Fowler said that three Sparc processors were currently in test right now in the Oracle labs, including the Sparc T5 as well as two other processors known as the M4 and the M5. It is still not entirely clear if the M4 chip is another name for a future Sparc64 processor from Fujitsu, as most of us expect.
In fact, it is highly likely that the M4 chip is none other than the 16-core Sparc64-X that Fujitsu was showing off at Hot Chips 24. That chip was revealed to be code-named "Athena" and is part of a new line of 64-socket boxes with that same name that Fujitsu announced earlier this week at OpenWorld. Fowler did not say that Oracle would be reselling these Athena boxes, and neither has CEO Larry Ellison, who spoke after Fujitsu on Sunday night.
The Sparc M4 and M5 were not on the roadmap shown by Fowler today, and neither was the Sparc T6. But that does not mean there are any changes between now and 2016 on the Sparc roadmap.
What Fowler is focused on is what he called Larry's Law, which is a corollary of the famous Moore's Law, which says transistor count on a die goes up every 18 months or so (although we're stretching it out to 24 months these days). Larry's Law is simple: "Just double it." Fowler said that when he explains to Ellison how difficult this is, that Moore's Law doesn't work that way, Ellison just stares at him blankly and says, "Just double it."
It must be a great deal of fun to be a 30-billionaire, even if you can't be Tony Stark.
To that end, Oracle's chip engineers keep adding in accelerators and other features to the Sparc chips that are designed to run specific database routines and other functions in the Oracle stack in need of electronic assistance, and that is how you can "just double it" despite Moore's Law. IBM has added similar accelerators to the Power7+ just announced today and Fujitsu is doing the same with the Sparc64-X. Coprocessing is the wave of the future.
Oracle's edge will still be that it owns the whole stack. "We are the only company in the world that can take application knowledge down to the silicon," Fowler said as he ended his keynote. Until Microsoft or SAP get into the hardware biz, or IBM or Hewlett-Packard buys SAP, he has a good point.
Now, as far as sales of Sparc T4 systems go, Fowler said that the machines were "the fastest ramping server product in Sun's history, and notice I said in Sun's history." This is a remarkable statement, considering how popular the Enterprise 10000, or E10K, servers were during the dot-com boom, but that is what Fowler said.
Fowler's comment comes on the heels of co-president Mark Hurd telling Wall Street two weeks ago that the Sparc T series machines had double-digit revenue growth and that Oracle expected this to continue through the rest of its fiscal year, which ends in May 2013.