Processor Whispers: About old annoyances and new challenges
by Andreas Stiller
It is hard to believe, but with the Remote Management Module 4, for servers with the new Xeon E5 processors, Intel gets rid of a long-standing annoyance. AMD sells its remaining stake in Globalfoundries and might be equipping the next Sony PlayStation.
Intel's PR in Germany happily moved its press conference from its traditional time in the morning of the first day of CeBIT to the late afternoon in order to celebrate the launch of an important processor – the Xeon E5, code-named Sandy Bridge-EP – at a time more suitable for the US. Almost more than twice as fast as its predecessor, the Westmere-EP, the Xeon E5 unexpectedly brings with it another improvement: 28 years after the AT, Intel has finally managed to eliminate an age-old annoyance. No, it's not the A20 gate, the cute, small gate that ensures the important compatibility to the address space of the late 8088 processor from the 70s and which continues to be present in the Sandy Bridge processor's integrated I/O (PII). This is about the exasperating key press required for entering the BIOS setup. In the Remote Management Module (RMM4), there now is the menu option "Reset Server, Force-enter BIOS Setup" on the "Power Control and Status" page – fanfare, please!
Hundreds of man- and woman-years have been wasted, only because it's so easy to miss the short period of time, which feels like one-hundredth of a second, in which you have to press the correct key – which one was it again? – to enter the BIOS setup. And then, of course, Windows Server will boot, won't even let you stop it with Ctrl+Alt+Del and will eventually insist that you log in. After what feels like an eternity, you can finally give it another try. In earlier times, if you had a USB keyboard, you had no chance at all, because they were only configured after the setup phase. And, in even earlier times, other hurdles had to be overcome to reach the setup, for instance, with PS/2, a specially formatted reference floppy disc was required.
With the ancient IBM/AT you even had to provoke a hardware failure – by removing the cable from the floppy disc drive – but at least it offered the generally unknown possibility of uploading and running a debug or setup program to 0500h through the keyboard connector before the boot process continued. So basically it already featured the early loading of drivers long before EFI/UEFI, the extended firmware interface that's been trying to replace the totally antiquated PC BIOS for more than a decade. Linus Torvalds might consider EFI to be fundamentally broken, but Apple has established it well and now Microsoft is giving it another try with Windows 8. Nonetheless, the setup challenge of pressing the correct key quickly enough will probably remain the same for many PCs.
Parallel to the launch of the long expected Xeon E5 family, AMD rolled out faster Interlagos processors, which offer up to 2.8GHz of nominal clock rate, as a "greeting". In terms of pure performance, these new versions are probably unable to compete with Intel's new high-end racers (first of all the Xeon E5 2960), but the Interlagos processors are significantly cheaper, so they are probably a good buy nonetheless.
That's most likely why Germany's fastest supercomputer, Hermit, at the High Performance Computing Centre Stuttgart (HLRS), is equipped with the reasonably priced Opteron 6276. Altogether, Hermit's Cray XR5 HE racks house 113,644 cores, which enabled it to score 12th place on the Top500 list of supercomputers with 831 teraflops – the giant data centres of Google, Amazon, Microsoft and the NSA are not listed, though. At the end of February, Hermit was officially inaugurated by the Federal Minister of Education and Research, Annette Schavan, and the governor of Baden-Württemberg, Kretschmann. Its main purpose will be to compute data for scientific projects and industrial contracts within the scope of the Partnership for Advanced Computing in Europe (PRACE).
However, Hermit won't keep the top position in Germany for long, because IBM is eagerly working on the SuperMUC with Xeon E5 processors in the Leibniz Computing Centre, located in Garching close to Munich. With 112,896 cores it will have a few cores less than Hermit, but it's supposed to reach (a projected) 2.21 petaflops in the Linpack benchmark this summer, more than 2.5 times as much as Hermit.
AMD got rid of its remaining stake in Globalfoundries, which is now wholly owned by ATIC. At the same time and in connection with this deal, the wafer supply agreement between Globalfoundries and AMD has been renegotiated. There won't be any more quarterly payments in 2012, only payments on a per-wafer basis. Additionally, the existing exclusivity agreement for the production of certain, not further specified 28nm APUs has been annulled; so AMD can now also contract TSMC to manufacture these chips.
Whichever manufacturer AMD chooses, with its heterogeneous systems architecture (HSA) it's facing interesting new challenges. The signs indicating that Sony intends to switch to x86 processors and AMD GPUs for its next PlayStation generation (2013/2014) are mounting. That Cell is not an option for Sony any more became largely clear when Cell partner IBM pulled out. With AMD's HSA concept, Sony could even fit the chip with its own extensions (FPGAs, media processors, DSPs and so on). Microsoft is also supposed to be highly interested in an extended cooperation with AMD, for its next Xbox generation. So why not leave the high-end sector to Intel if one is conquering other business branches?
And Something Else
It's much more likely that the pope becomes the federal chancellor – said Prof Karsten Danzmann from the Albert Einstein Institute in Hannover some time ago when faced with the question of whether neutrinos could move faster than light. The 15 scientists who also considered the paper about the faster-than-light neutrinos to be premature and decided not to support it can now rejoice because it seems very likely that a simple wiring fault in the plug of the optical fibre connecting to the GPS station caused the "60 nanoseconds faster than light" result. Well, at least everyone knows about the OPERA project now, and we could all enjoy lots of neutrino jokes.