Processor Whispers: About Old and New Friends
by Andreas Stiller
Yes, we can: With a little help from a friend, AMD was finally, after 12 negative quarters, able to show a profit of 1.18 billion dollars. Things are starting to look brighter for the scene in general, although a few are still wrestling with problems.
Giddy with joy over the good results, AMD rolled out five new desktop CPUs shortly after announcing the new numbers. It’s a pity for AMD that the 1.25 billion from Intel - which went into that last quarter - was a lump sum and won’t improve the budget on a regular basis. With a net income of 2.3 billion dollars and sales of 10.6 billion dollars, Intel’s financial statement was apparently hardly affected by the small settlement payment. As an indicator for the whole sector Intel therefore signals the end of the crisis. Apple, IBM, Google, Texas Instruments and other protagonists confirm this trend by showing good results, too.
The numbers presented by the market research company IDC point in the same direction: never before had as many microprocessors been sold as in last year’s fourth quarter - a 31.1 per cent increase in comparison to 2008. This development has been accompanied by Intel loosing a bit of its market share to its competitor AMD again, during the fourth quarter; Intel dropped from 81.1 to 80.5 per cent and AMD went up from 18.7 to 19.4 per cent. Meanwhile, VIA processors have become almost completely irrelevant – their share dropped to 0.2 per cent.
Amongst the winners is also the Korean DRAM manufacturer Hynix, which was able to increase its sales by 32 per cent and present a profit of more than 400 million US dollars. Obviously, the DRAM business is reviving. Maybe it would be wise to reopen the Qimonda facilities in Dresden, Germany, instead of getting rid of the rest of them; three quarters of the equipment still remains there. In any case, neighbor Globalfoundries is rigging up. CEO Grose plans to invest 900 million euros, exactly the amount AMD got from Intel, into Globalfoundries’ facilities in Dresden this year. Main investor Mubadala from Abu Dhabi has also merged the acquired manufacturing sites in Singapore with Globalfoundries. Mubadala’s subsidiary ATIC, which now owns about 65 per cent of Globalfoundries, is supposed to take over the facilities in Dresden completely after coming to an agreement with Intel – the application is already in the hands of the Federal Cartel Office.
Meanwhile, the European Commission for Competition has given Oracle the permission to acquire Sun. “So long, old friend” – it was with these words beneath the image of a Sun tombstone that Java creator James Gosling commented the final goodbye.
At the time of writing this column, it was still unclear what Oracle has in mind for the server branch and for the future of Jonathan Schwartz and his place on the Sun King’s throne. According to IDC, Sun – with a market share of 9.5 per cent – still holds the fourth position behind IBM and HP (both with around 30 per cent), and Dell with 13.5 per cent.
However, Sun’s competition isn’t idle: IBM is going to officially release the server processor Power 7 at the Solid-State Circuit Conference (ISSCC) in February. Intel and AMD plan to go strong with their many cores Nehalem-EX and Magny-Cours respectively a bit later, in March. And also the Tukwila-Itanium is supposed to get into the server act someday – though, most likely, much too late; its environment is getting shakier by the day. Although Itanium servers are about head to head with SPARC sales, CERN, one of the flagship instances for the Itanium architecture, is discontinuing Scientific Linux CERN ia64 due to lack of interest at the end of March.
An old architecture for large servers or supercomputers might be making a comeback: MIPS. Even though SiCortex failed with this idea only last year and SGI now belongs to Rackable, the Institute of Computing Technology (ICT) of the Chinese Academy of Sciences (CAS) intends to have a go at it with the Loongson-3 alias Godson-3 featuring up to 16 MIPS64 cores. The 65-nm chip – developed in cooperation with STMicroelectronics as manufacturer – is supposed to push the next Dawning-6000 supercomputer over the petaflop mark. The Loongson-3 can also process x86 code and a co-processor is meant to boost the performance particularly in the Linpack benchmark.
In the last days, there has been a lot of news on the patent front. The list of new accusations seems almost endless: Kodak against Apple, Motorola against Blackberry, Microsoft against TiVo, Infineon against Volterra, Richtek against AMD, Toshiba against Wistron ...
And the dispute between Nokia and Apple is escalating further – Apple now wants to see a US import ban on Nokia phones.
Some new Rambus victories stand out among the rest of the patent quarrelling. The small technology licensing company with the large legal department has come to an arrangement with the second largest global semiconductor manufacturer Samsung and is pocketing 900 million U.S. dollars for an indefinite DRAM patent license. In the action against Nvidia, before the U.S. International Trade Commission, two of the five Rambus patents have been determined as invalid, but the other three have been judged as violations. Now Nvidia is facing a possible import ban on its graphic cards. Obviously, Nvidia won’t settle with that, wants to appeal and, among other things, points out that an investigation by the U.S. Patent and Trademark Office is still under-way.
However, Nvidia is not only under legal pressure. While the launch date for the DirectX 11-compatible GeForce chip is pushed back again and again and its power consumption reportedly rises, AMD is mercilessly launching one DirectX 11 GPU after another.
The supply bottlenecks are history, promises the boss of the chip manufacturer TSMC. The 40-nm process, which is also used for the manufacture of Radeon chips, has overcome the initial problems, he says. Also Nvidia wants to use the new TSMC technology, for future Fermi chips. But Nvidias’ “favourite” journalist Charlie Demerjian, from semiaccurate.com, reports that the most recent Fermi A3 stepping is not holding up to expectations and is running on lower clock speeds than was planned.
He adds that, according to his information, the smaller and much cheaper Radeon chips have hardware tessellation, while Nvidias GF100 mostly has to get this job – required by DirectX 11 – done by software, via the shaders. Hard times for Nvidia – and a CPU (apart from the Tegra) for the CPU/GPU integration, like with AMD’s Fusion, is also missing, at least for now.