Some conveniently late news (after SC09) for the HPC market
I may write more about SC09 in the next few weeks, but for now the video is starting to show up on at Linux Magazine and I already posted a SC09: Three Trends Worth Watching over at HPC Community. You can get a good feel for things from those two articles. I'll also be posting more video on Linux Magazine over the next several weeks.
The weeks after SC are usually quiet as the holidays approach. This year seems to be a little different. There have been more than a few interesting annoucements about the HPC market. I find it kid of interesting how all these show up after SC09. Perhaps these are the kind of announcements you don't want gobbled up by the press. In any case, I have collected the stories and added a little commentary to each. An interesting time in HPC to say the least.
There were two news items from Intel. First, they have announced a 48-core concept chip. Nothing excites me more than new hardware and this processor seems to be plowing some new ground. Note, however, that is a research processor and they only expect to seed 100 or so out to researchers, to which I need to point out I do research. Please disregard any references to "cloud on a chip" circulating in some of the Intel press releases. Attention Intel PR people, I do cloud research.
In other Intel news, there was the cancellation of the Larrabee retail launch. As I understand it, as a commercial product, Larrabee is done. Forget all the "HPC Chip" chatter as well. If a processor is going to make any in-roads in HPC, it must be produced in mass quantity otherwise the price will be too high. In my opinion if there is not a consumer (e.g. gamer) product, there is no hope for an new HPC processor -- other than R&D as mentioned in the above paragraph. I wish it were otherwise, but such are the economic realities.
While we are talking about canceled chips, IBM has announced no future development for HPC PowerXCell. This announcement does not mean the Cell is being canceled, just the use of it IBM HPC (read Deep Computing) systems. Rumors are AMD/ATI chip-sets are going to be used. When ever I read announcements like these, I always think, what about all the people IBM encouraged to write software for the Cell? There are a few good reasons x86 processors took over HPC. One of them is that the architecture is not going anywhere. The HPC market is replete with stories of latest-and-greatest to orphan status within three years. The x86 architecture, warts and all, has one thing going for it, namely, it is "too big to fail."
As you can imagine, the hybrid HPC competition (GP-GPUs) is now a two pony race -- NVidia and AMD/ATI. In a way that is comforting to developers. However, there is still the software issue. More on that at another time.
Enough hardware news. I would like to finish up with some exciting software news. The Open MPI team has announced The Open Resilient Cluster Manager (ORCM, or OpenRCM). The project is open-source and focuses on development of an "always on" resource manager for high-performance computing systems of any size. The objectives for the system are:
- Maintain operation of running applications in the face of single or multiple failures of any given process within that application.
- Proactively detect incipient failures (hardware and/or software) and respond appropriately to maintain overall system operation.
- Support both MPI and non-MPI applications.
- Provide a research platform for exploring new concepts and methods in resilient systems.
Good idea because I have been told things break.