About Vuvuzelas and Other Noisemakers

Friendly e-mails between the bosses of big IT companies show how they do business. The chip manufacturers are again investing billions into new fabs and the server racks and blades are hastily being equipped with PCIe-16 slots to make them ready for the hot GPGPU cards.

The vuvuzelas are potent but so are the inventors of the vuvuzela filters. What an invention: Tones of about 233 hertz – including overtones – get eliminated. A task every well designed digital filter software or analogue band-rejection filter can handle – the latter might even do a better job and would be more energy-efficient in any case. But we are all digital nowadays: the ancient software I use for this purpose is from the early Delphi era and employs Intel’s DLL for Native Signal Processing (NSP.DLL) for the necessary Fourier transforms.

Something younger readers might not know is that NSP almost caused a serious breach between Microsoft and Intel in the mid-90s. Intel had noisily exclaimed that certain tasks, like audio and video processing, could be handled by an NSP interface (via the – back then – recently introduced MMX) instead of the operating system. Microsoft was indignant and felt that Intel was poaching in its software waters and endangering the success of its new OS Windows 95. In the summer of 1995, Microsoft boss Bill Gates and Intel boss Andy Groves had such an “animated” discussion about this matter at the Intel headquarters that the whole Robert Noyce Building overheard them – must have been much worse than the current vuvuzela buzzing.

  The cores of the Apple A4 (on the left) and the Samsung S5PC110A01 are as alike as two peas in a pod. (Source: PR Newswire/UBM TechInsights)

Eventually, Intel gave in by giving up on its NSP concept – or rather by moving it to compiler libraries (IPP) – and promised to support DirectX in the future. To what extend the gentle pressure from Microsoft might be judged as a distortion of competition became a topic later on, during the anti-trust trial against Microsoft – which was mostly concerned with the Internet Explorer bundling issue. Intel’s vice president Steven McGeady was heard as a witness, at which also some friendly e-mails exchanged between Grove and Gates in that time came to light.

Best Friends

And now, a new set of e-mails has been unearthed, e-mails that show how Intel maintains the commitment of its partners. Apparently, Dell has come to an agreement with the U.S. Securities and Exchange Commission (SEC) about how the once received “contributions” from Intel should affect the balance sheet and is willing to settle the matter – at a cost of 100 million U.S. dollars. “Best friend money can buy” – that’s what Intel boss Otellini is supposed to have written about Dell in an e-mail.

The questionable payments also play an important role in the trial against Intel that was initiated by the ambitious New York State Attorney General Andrew Cuomo at the end of last year. According to the e-mails, Michael Dell complained to Otellini – in 2005 – about having lost deals to rivals that had used AMD chips. Otellini replied that Intel had paid Dell more than one billion dollars in the year before and that this should more than make up for the problems with the competition.

Some assume that there might be a special motive behind this trial against Intel. After all, the State of New York subsidizes AMD’s manufacturing company Globalfoundries with a billion dollars. However, Cuomo – who aspires to become New York’s governor in the future – will most probably disclaim that this played a role.

In any case, Globalfoundries has meanwhile settled things with Intel and business is apparently running better, too. The company, which is now mostly financed by Abu Dhabi, intents to invest an additional three billion dollars to expand its production capabilities in Dresden and Singapore and to increase the number of monthly wafer starts from 42 000 to 60 000 in Malta (New York).

The expensive processor development isn’t run by Globalfoundries alone, but in cooperation with IBM, Samsung, ARM and Synopsis in the context of a “Common Platform Alliance”. This alliance has now presented the first “vertically optimized” 32/28nm design platform at the Design Automation Conference in New Orleans.

Also Samsung is expanding and is spending 3.6 billion dollars on a new factory for smartphone chips and chips for other mobile devices in Austin/Texas, where it already operates a big manufacturing facility for flash chips and where – coincidentally – Apple is represented, too – by its recent acquisition, the small chip design company Intrinsity. Chip analysts have confirmed that the Cortex A8 design in Samsung’s Android smartphone Wave s8500 (S5PC110A01) is 100 percent identical to the A4 design in Apple’s iPad and iPhone as far as the core is concerned – so it all makes sense.

Optimism for everyone it seems, with big projected growth rates of up to 57 percent per year for iPads and tablet PCs, according to the market research institution IDC. For the desktop PC and server sector, IDC announced a strong growth of 23 percent over the last quarter. Overall, the sales on the server market have increased by 4.6 percent, while Dell improved most with a sales increase of more than 50 percent.

Tesla Frenzy

For high performance computing, the server companies now heavily rely on new racks and blades with PCIe-16 slots for GPGPU cards. IBM could even show off some Tesla-M2050 cards for the iDataPlex at the Super Computer Conference ISC10, but they only were mechanical samples. Some competitors even plan to equip a single node with two GPGPU cards, like Dell with its new blade PowerEdge M610x or NEC with its HPC server 24G2Rc-2.

HP still tops this: together with the Tokio Institute of Technology, HP is developing a large node for the SL line (SL390s G7) with two Intel Westmere processors, two Infiniband QDR interfaces and three Tesla-M2050 cards.

AMD is being noticeably shy about its new FireStream cards, the Tesla competition, but as far as its Magny-Cours processors are concerned things look promising in the high performance scene. Apart from NEC, with the mentioned HPC server 24G2Rc-2, also many other first tier players offer corresponding platforms. Among the most powerful: the new Cray XE6, which links its two Magny-Cours processors per core via the Gemini Interconnect – with 160GB/s of internal switch performance.

Supermicro and its partners, like Boston Inc., are already delivering four-way systems in a 1U from factor while most others intent to follow up with 2U slots soon, like Dell with the PowerEdge R815 and HP with the DL585 G7.

Relatively new to support AMD, SGI is offering the Cloudrack X2, the blades IP-106 and IP-110 and soon also the Altix-ICE-8400 line with the Dual Opteron.