After years of focusing on building infrastructure, one of priorities for the Chinese government is focusing on building the knowledge infrastructure as well. In terms of computers, this is a real battlefield. While the western media tends to go by Blue's company line about GPUs being a "thing of the past" and frantically releasing the Larrabee 2 as Knights' Corner (and pricing it to fight NVIDIA Tesla), China has a GPU attach rate of over 80%, and the amount of PC gamers in the country is beyond that of all other continents... combined.
Thus, being in the Chinese market is a paramount. Keep in mind that the craziest GPU designs now come from Chinese Colorful and Zotac, with models that are not available on the global market. On the homefront, China is not looking bad either. The country is advancing its own high end CPU agenda at a rapid pace, a pace unseen in recent history. Just like they have the world's fastest commercial trains, with the 430 km/h, 267 mph Shanghai Maglev, China may - in a year or two - have world's fastest general purpose CPUs, whether their semiconductor process matches that of the West or not.
The Chinese have their own Alpha, MIPS, Sparc and many ARM multi-core implementations, covering everything from smartphone to supercomputer. The Loongson MIPS-based team from Beijing, supported by the Chinese Academy of Science and the Central Government, leads in commercial acceptance and has a multi-million user base. China even has their own instruction set architectures e.g. ISA, like the Icube from Shenzhen, the fully fused CPU and GPU on the register and core level, named UPU (Unified Processor Unit), a complete fusion. The architects of this little wonder are some of the creators of high-end MIPS R12K/R14K 64-bit CPUs, the PowerPC 620 (the only 64-bit PowerPC). More importantly, the chief architects of UPU are former NVIDIA employees. They created the very first NVIDIA GPU compute shader, an architecture we know today as the G80. As you can imagine, there is a lot of positive sentiment for the green-chrome company since they do produce world's most complex silicon.
What Went Wrong? Recently, Linus Torvalds, the father of Linux went on a tangent during a talk, gave NVIDIA the middle finger and said "**** You, NVIDIA."
While NVIDIA attempted to mitigate the damage by releasing a story, there was a 'behind the curtain' reason why Linus was so frustrated with the company. It did not have to do with open source as much as it did with China.
A rumor appeared from the heart of Beijing that due to the performance of its GPU architecture and its Linux drivers, NVIDIA was approached by one of the leading Chinese CPU teams to use an NV GPU in a pilot school PC project. The Linux would run on the Chinese CPU, while GeForce GPU would provide the graphics power. 'Pilot project' in this case means over 10 million PCs in one order, broken down - 100,000 schools with 100-150 PCs each. The problem was two-fold; NVIDIA never releases source code for its Linux drivers, and the binaries are only X86. Incentivized by the Chinese government, the Chinese CPU team called NVIDIA to come to China and work with them.
To cut the story short, the NV team appeared there, and in very arrogant manner told the Chinese side that they are a large US corporation, and that recompiling the Linux drivers would cost the Chinese a lot of money. The money that Chinese CPU team and the Academy of Science were supposed to fork out was to the tune of several million dollars in incentive that are typically referred to as NRE - Non-recurring Engineering.
Our sources close to the heart of the matter said that was the end of the meeting and of the relationship. While we cannot foresee the consequences of that meeting, bear in mind that back at the day, Intel supplied Chinese government with an Itanium-based cluster that failed miserably, and the Chinese forced Intel to invest heavily in China. To this date, this was one of smartest moves Intel pulled, as they enjoy a very fruitful relationship with the Chinese government.
Epilogue - AMD Wins
With NVIDIA back in Santa Clara, California and Southern China, there was no doubt as to who the Chinese would call next. The other GPU vendor, while having mediocre Linux drivers, at least did not make any fuss over source code access. This ended up in being part of a 10-15 million PC project. Even if selling its cheapest DirectX 11 capable GPUs, this is a revenue opportunity of at least 250-350 million dollars. Thus, if the Beijing rumor turns out to be true, this means NVIDIA destroyed a relationship with a potential long term partner, which in turn began working with AMD. All of this was caused by a couple of million dollars in NRE's on GPUs which were already paid out (let's say NVIDIA would end up selling 10 million GeForce GT 520 boards), and for a project that required drivers that NVIDIA needs to develop anyways (after all, CARMA toolkit is consisted out of ARM-powered Tegra 3 processor and a 96-core Quadro GPU)?
The real question remains, though - when will the company stockholders start demanding responsibility for a lost deal of at least quarter a billion dollars, a deal that went to NVIDIA's direct competitor - which is rapidly gaining market share?
How many similar deals has the company lost by being arrogant and demanding NRE (Non-Return Engineering) from companies that would buy anywhere from tens of thousands of Tegras, GeForces, Quadros and Teslas - to tens of millions? After all, NVIDIA proudly pronounced that the company took Intel's spot with BMW Group, while the deal for next generation of embedded systems in cars went to Freescale as the SoC and Vivante as the GPU provider.
Bear in mind that this school PC project was just a pilot, while the full fledged program aims to place a computer on every desk in China's schools and replace as much paper as possible. While the kids in the West are stuck carrying their books around, the goal in China is to have children carrying no more than one or two paper notebooks, to be replaced with tablets in near future. What is considered science fiction in the West, due to numerous interest groups - is considered a plan in China.
Many high end Chinese CPUs use AMD-designed HyperTransport to connect their processors to SMP setups, which provides more opportunity for the red team, and we don't mean red as in RED 4K/5K camera, although China is very interested in that as well...