Looking at Old HPC IDC Predictions (2010)

It is  more interesting what was predicted by top guru analysts for 2010, now that we are in 2013

Idc Hpc Web Conf Predictions 2010 Final 

IDC predicted Infiniband will be king, on the "evolutionary" supercomputers of 2010 (meaning old technology, stalled thinking) as described on slide 9
  • Most of the biggest, baddest supercomputers are architectural clusters or x86 MPPs, with bulked-up interconnects and support for MPI or PGAS languages
  • Revolutionary change may still be on the horizon, because evolutionary change is not addressing fundamental issues, such as how to program machines that will soon have 1M or more cores
This blog covered the first bullet in the article about Higgs Boson calculations at CERN. Large Hadron Collider people assembled a grid of 150 data centers in 40 countries using opportunistic cycles and resources. delivering the power of High Throughput Computing. There is no supercomputer yet built  using the "evolutionary" thinking (like TOP500 thinking) to deliver this power, never mind the prohibitive costs.

The second bullet, on how to program 1 million cores or more, is dealt in a few blog entries here, like the Interview with David Ungar  He discovered the  genial  "heresy" of not communicating between nodes, an Aha!  similar to the non-Euclidean geometry. We must decide what precision we need, because there are  not absolute single answers in nature. And the precision we need is directly proportional with the cost to produce such calculation.

This is why I am impressed by IDC predictions

Post a Comment

Popular Posts