Google and PeakStream
Recent reports in the Press have Google acquiring PeakStream Inc. According to some sources PeakStreams products will no longer be available.
PeakStream are, or maybe were, a Redwood, California-based company specialising in software tools to support programmers developing for multicore technologies. Target applications were all in the technical arena and didn’t include mainstream commercial areas. The hardware platforms that they aimed at were varied and covered the x86 derivatives, IBM’s Cell processor and GPUs (Graphics Processing Units) among others with a Linux-based toolkit. A Windows-based beta version of their tools became available earlier this year, which firmly carried them into the wider commercial marketplace.
In a much-quoted article in HPC Wire earlier this year, PeakStream’s founder Matthew Papakipos, talked of the convergence of various architectures that lay behind his company’s vision and looked forward ten years or more to heterogeneous many-core HPF systems. He implied that this was the direction in which he was seeking to take his company. With a background in GPUs and looking towards general purpose GPUs (GPGPUs) PeakStream certainly had the architectural experience to address those areas.
Is this just another recent of Google’s recent buying spree or is there a deeper strategic purpose here?
Maybe. Google is allegedly one of the largest users of compute power in the world and is expanding rapidly into applications areas that were once well outside its initial remit. This of course begs the question as to why on earth Google should acquire PeakStream in particular. Much has been made in by analysts of the convergence between Google’s interests in image analysis and PeakStream’s graphics heritage. In particular they have focussed on the ability that Google would liker to have of being able to recognise that an image contains an instance of a particular target object, even though the object is not itself identified in the description of the image. Of course that fits well with Papakipos’ previous history and may well be at least part of Google’s reason.
Google is also one of the largest purchasers of compute-power world-wide. It also deploys that power in ever more complex ways. That may also hold the reasons why. In recent years the team at Google have developed several new classes of algorithm. Although its architectures aren’t public domain, it is generally believed that Google is a fairly intensive user of parallel and grid-style architectures. This would imply that they are likely to be looking to large-scale multicore systems in the not too-distant future. It would not seem unreasonable then for Google to acquire PeakStream for two reasons. Firstly to enhance its own existing teams and secondly to build the necessary tools to program any future HPC/HPF systems.
It’s interesting to note that the Google has a record of turning its in-house research projects and tools into marketable products. Will it simply integrate PeakStream’s team with other projects to help it build up its in-house expertise? PeakStream itself had a track record of producing commercial software products. Now, if Google wanted to dominate MCP software would a company like PeakStream be the sort of company it might buy up?