China to go ahead with ‘National Processor Architecture’, but which one?
Reports say China is going ahead with its own nationwide instruction set architecture, something that will surely not be X86. This is a massive endeavor – what may really happen?
We've been reporting quite a bit about China's multiple CPU development and production efforts across many instruction sets, even those long ignored in the West like the Alpha, for instance. Now, the news, also referred to in the US press (source http://www.eetimes.com/electro
This is a big one, because it means that, among at least five instruction sets to choose that are locally further developed, one would become the preferred one. Basically, we got MIPS in Loongson and Ingeniq, covering all from smartphone to supercomputer; then 'Shenwei' Alpha, mostly for military-linked workstation, servers, supercomputers and such; 'Fengtian' SPARC in the same fields as Alpha; Icube UPU integrated CPU-GPU for mobile and microserver markets, and over a dozen ARM licensees.
I made some calls here, and the take is this: there is concern about MIPS' possible sale of the company and the associated IP and such, but MIPS is trying to sell itself for over a decade now, without results. Plus, Loongson seemingly has an architectural license of the architecture, enabling them to modify and evolve it as needed – no matter what happens to the original license provider. After all, tens of millions of mini PCs with Loongson are there for Chinese schools here. And, the Alpha core in Shenwei is a derivative of a 1995 design, for which the IP challenge would be all but gone. Therefore, these two architectures, by themselves having been acknowledged as the most elegant and simple that the West had by the end of last century and so easy to develop further, are good candidates. Then, the Icube's own, existing ISA, that has both CPU and GPU integrated at the register set and execution level, comes in for consideration too.
Beyond this, we're pretty sure if will not be another instruction set to be defined from scratch, especially not by a committee. Defining a new ISA is a colossal task, and something that, in the West, hasn't really been done since the Alpha over two decades ago. We all still seem to be enslaved by the archaic X86, and no, don't blame Intel for it. It's is IBM that made the far reaching bad decision to propel X86 into the PC realm 30 years ago, just to save a bit of money compared to the vastly superior Motorola 68000, the one used by Apple Mac soon later – its power enabled the first GUI, after all.
China needs to keep a choice of architectures, it's a large country and big market unto itself, and a bit of competitive spirit would help those architectures develop faster and keep prices in check.
After all, a single chosen ISA will not be pressed to innovate and advance, if being granted the market dominance and government support. Beyond that, the whole thing would need a big ecosystem, with design and simulation tools, compilers, libraries, and so many other things.
Let's see what happens, based on what I was told this morning, the decision is likely by early next year, and there could be more than one 'winner' at the end.