The CHPC National conference took place in the first week of December, and it had some interesting talks. All the Videos are viewable on YouTube
The last keynote was by Prof Thomas Sterling, of Indiana University, where in a wide-ranging retrospective of HPC technology, he talked at the end about :-
- “A brief discussion of early attempts of memory-centric computing such as SIMD and PIM will motivate revolutionary concepts of the future. Questions from the audience will be welcome assuming remote communication technology permits.”
You can watch his presentation here.
Well, just right up my street.
Earlier that day, I gave a talk about my experiments with a SIMD computer of my own design, and demonstrated a program running on it, using the available parallelism.
You can see my 20 minute talk in the HPC technology session.
I also have similar high hopes to Prof Sterling, but it will require new processing algorithms - probably dramatically simpler - to utilise this massively parallel technology.
When the parallel processing comes for free, the algorithms must be adjusted to the machine - using straight-line code with no conditionals - switching program paths by conditionally enabling a subset of the total processors.
The bandwidth and power advantages means that this approach must be investigated.
There is history of previous attempts at this golden calf, but todays technology dramatically improves the odds of it succeeding.
Though the talk is about the algorithm, the underlying machine I hope can be seen underneath.
I must also give a shoutout to the other presenter in my session, Yaniv Swiel (Wits)
who had a project to use a C to HDL tool to dramatically improve his bioinformatics codebase by using the FPGA for permutation testing.