Huawei moves into supercomputing (too)
Huawei network switch? Then, Huawei smartphone? Well, get used to Huawei supercomputers soon.
The Shenzhen giant company which, as its proponents will repeat, started without any government help – although, of course, it is well linked with it today – is, in line with their 'next IBM' strategy we unveiled here some months ago, widening its reach.
Geographically, they pretty much covered the Earth, and their componentry are bound to go beyond this planet too, in Chinese space missions. Target market wise, the telecoms (both consumer and enterprise related) markets are covered as well, with CPUs and systems markets now under the assault.
The newest arena Huawei is entering is at the very top: supercomputers. Why bother, when it rarely makes money? Well, Huawei management, according to our sources, thinks high performance computing can make money, and not just in China. There is enough differentiation to make the designs blades and other nodes more HPC optimised, both CPU and memory performance wise, and then use Huawei network and interconnect expertise to improve on the inter-system connection efficiency, where even the Infiniband leaves a lot to be wanted.
Huawei, according to our high level sources, already got a substantial system portion of some of the upcoming ultra large systems in China, including what would likely be the world's first 100 Petaflop supercomputer in a year's time, at least five times as fast as the just unveiled fastest US system two weeks ago. This includes the node board design and rack infrastructure, among others. Of course, more of the same will follow, both above and below that performance threshold – some of it likely with Chinese – even Huawei's own – CPUs.
Since supercomputers are a costly and tough to earn money market with even tougher customer support for most generic vendors like HP or Dell, it'll be interesting how Huawei will handle the challenges of high performance computing marketing, sales, integration and support. If the past is anything to go by, looking how Cisco fared trying to compete with Huawei, the Western SI vendors better beware. After all, they themselves killed off their only advantage – their own CPUs like Alpha or MIPS or HP-PA – to be left essentially as large scale Intel or AMD system integrators. At that point, price and little tweaks, plus support, become the only differentiators, and the Chinese can do it just as well, if not better. After all, for over a decade, even IBM's main PC server server design solution, integration and manufacturing facility for Asia Pacific was right in Shenzhen. The staff must have learned a lot during all those years, before many of them moved to the local big vendors, including Huawei…