Processor Whispers: About prices and placeholders
by Andreas Stiller
With slimmed-down Xeons and even Pentiums, Intel wants to make economically priced entry-level workstations and servers possible. Quantum computing is gradually gaining pace and Samsung is putting its shoulder to the wheel to remain as Apple's processor manufacturer.
The Enterprise Pentiums and the Xeons of the 1400 series are a strange family. Intel hasn't officially released them yet, but they are marked as "launched" in Intel's processor database at ark.intel.com. These chips appeared some time ago in leaked roadmaps and in the compatibility lists of IBM, Dell and other server manufacturers. Some of the new processors in the database (see list of processors below) are not even linked yet; you can only find them if you have a hunch and manage to guess their names. Prices are also still missing and you won't be finding socket specifications either.
We are talking about Sandy Bridge processors for the socket LGA1356, which means three memory channels and one QPI link. However, in this case, it transfers at 0 GT/s because the processors work strictly and stubbornly alone – similar to the E5-1600 family for the socket LGA2011 (four memory channels, two QPI links).
Depending on the perspective, they can be regarded as budget-priced entry-level components or as preliminary placeholders primarily meant to allow the manufacturers to shine with exceedingly low entry-level prices. For instance, Dell is offering the tower version of the PowerEdge R320, equipped with a Pentium 1403, starting at as little as 749 euros (about $940). The slightly newer "super saver" version, the Pentium 1405, wasn't even considered. With its two cores without hyperthreading, with only 1.2 GHz of clock speed and only two memory channels, Intel basically should almost have to give it away for free. After all, the more powerful Pentium 1403, with two cores without HT, manages 2.6 GHz (but without turbo) and has three memory channels, and Dell is asking $120 less for it than for a E5-2403, which takes the price to about $68. But the information about the number of memory channels at ark.intel.com might also be simply wrong, because the Pentium 1405 shouldn't be able to address the indicated 375 GB of memory or reach 25 GB/s of memory bandwidth with only two memory channels.
With the rack server x3530-M3, amongst others, IBM is offering the 1400s as processor option for dual-socket systems, which is more or less like equipping a Porsche with a VW Golf motor. And, well, if the next budget plan provides the money for an upgrade, the system can be equipped properly.
As for the recently presented Puresystems series, it's rather unlikely that IBM is also going to offer such entry-level chips here. On the other hand, it's certain that IBM intends to incorporate the flash storage systems of the recently acquired company Texas Memory Systems into this series as soon as possible. TMS claims that it has – amongst others with the RamSan 820, linked via InfiniBand or 8Gb/s fibre Channel – the world's fastest storage.
IBM doesn't only buy companies, however, but also talented scientists, like the physicist Erik Lucero from the University of California, Santa Barbara. There, not only did he enjoy participating in ultra marathons, but he also conducted intense work, within the scope of his PhD, on the topic of quantum computers with solid-state qubits. They are supposed to provide a better scalability than other technologies such as ion traps, ensembles or photonic systems. However, all of these had already solved the preliminary task of breaking down 15 into its prime factors, now this has successfully been done with solid-state material.
Source: UCSB While the processing power is not exactly impressive yet, Lucero sees, in the not-so-distant future, quantum computers that can handle prime numbers with 600 bits at a speed 15 orders of magnitude faster than traditional computers. And with IBM, he has made a good choice of employer, because Big Blue is dedicating lots of money and enthusiasm to this subject, collaborating with, amongst others, the Juelich Research Center and the RWTH Aachen University. As recently as spring of this year, the "experimental quantum computing" team in Yorktown, headed by Mathias Steffen, announced "groundbreaking" achievements with 3 quantum bits based on silicon. The Karlsruhe Institute of Technology has shown how qubits can be saved for a longer time. It seems like quantum computing might now advance much faster than we believed even a short while ago.
In contrast, Intel is keeping a rather low profile regarding the topic of quantum computing, but it is working together closely with scientists in numerous other areas. For instance, together with medical scientists from Stanford University, it has created a chip that is able to analyse protein interactions and identify severe forms of the autoimmune disease lupus.
Apropos Stanford University: over the last 24 years, toward the end of August, the Hot Chips Conference has always taken place in the University's Memorial Auditorium. But now it's moving to Cupertino, where news about Haswell, Xeon Phi, AMD Jaguar, IBM Power7+, Fujitsu SPARC64-X and so on will soon be announced. And no, in spite of Apple being Cupertino's local superstar, the company is conspicuous by its absence, it's not a sponsor and neither will it send any presenters. And also Apple's processor manufacturer, Samsung, is staying away.
Samsung wants to remain Apple's first choice and is now further accommodating its client: while Samsung is not making any concessions regarding the patent dispute (yet) – a subject that Kwon Oh-hyun, Samsung's new boss since June, and Tim Cook are frequently and diligently discussing via telephone – the company is doing Apple a favour concerning the production site for future iPhone and iPad processors. In Austin, Texas – just where Apple's chip experts are at home and where Samsung already has a large factory for 45nm technology – Samsung now intends to invest another $4 billion. The often-trumpeted move from Samsung to TSMC will most probably remain an empty threat.