Confirmed: IBM moving Power Systems manufacturing to Mexico
poweritpro.com
poweritpro.com
"The reality is, we constantly evaluate our operations, and we look at the best possible infrastructure to maximize our competitiveness in a global market . . . and this is one of the ways we can do that."
WTF does that mean?
Loved Rochester, MN in spring and fall, you can keep it in Winter or Summer. The city has the Mayo clinic so its not going to fold up if parts of IBM move.
That site is home to their iSeries and the bulk of their level two support and development occurs there.
The long version from my perspective, as a former Power systems sales engineer/programmer/consultant, is that IBM's server sales division is struggling with declining sales. The source is likely virtualization and raw system capacity reducing the number of machines that the overall market needs. IBM moved xSeries (x86 machine) manufacturing/distribution to Mexico more than a decade ago, IIRC. So, this looks like a consolidation of both architectures in a single place to me.
Additional issues besides virt/capacity presenting themselves include not just the availability of cloud computing from companies like Amazon and Rackspace (and in house "clouds" like Google's x86 based white box one), but the design of start-up infrastructure (and even applications) specifically for that model. The emergence and widespread use of SaaS is similar issue.
The core issue with Power though, is that it's a different architecture than x86. If you really want to take advantage of the architecture you have to build your software (and all the underlying libraries and other dependencies) for that architecture. That work was often worth it when x86 chips were single core, 32bit, and workload could max out as many processors as a given system bus could support. But, now days with multi-core true 64bit x86 chips there are very few workloads where you need the Power architecture.
The Power is an amazing and beautiful architecture. But, even in the high end where you need the computational power it can provide it is not usually a compelling choice. For example, if you're doing large amount of complex mathematics (for physics simulations for example) there has been a trend toward using GPUs to perform the calculations due to constrained research and academic budgets. The libraries and code your simulation would run on are likely already optimized for GPU offloading. However, Power optimizations (if they were ever done in the first place) may not be kept up to date (ex: for Power4 vrs Power7) and may not be optimum or even build-able.
As the Power install base has declined, it's become more and more difficult to find people with the Power skill set and experience. While Linux distributions are available and supported for Power. Power was really designed to run AIX. Finding someone with AIX, IBM build tool chain, optimization skills, and Power assembly experience is very difficult.
The reason I left the IBM channel in the early 2000's and took a different career path is because I saw Linux/x86 emerging. I believed that it would eventually become dominant just based off my sales win/loss history then. And, cloud (or grid computing as it was called back then) was starting to emerge.
I don't know if Power is "legacy"/done for. I'm sure there are some specific use cases and applications where it still shines - like I said it's an amazing architecture. But, I think the writing is on the wall for all companies manufacturing large servers - IBM, HP, Dell. After all, there's good reason IBM has been pivoting to become a primarily services company over the last 15 years.