Senior Fellow, Chief Product Architect at AMD
Now that he’s back in the game and re-focused on his passion for graphics and visualization technologies, what’s in store from AMD this year that will level the play field in today’s fiercely competitive graphics and visualization landscape?
Gustafson has actually been a catalyst for introducing disruptive ideas into HPC since the early 1980s. Early on while at Floating Point Systems, he pointed out the need for level 3 BLAS and showed the LINPACK crowd how to make the benchmark scalable; the result is the basis for today’s TOP500 ranking. At a time when Cray was defining supercomputing with vector mainframes, John promoted the use of commodity cluster VLSI parts that would fill an entire building and use massive amounts of message-passing parallelism.
While at Sandia National Labs, he formulated the infamous counter-argument to Amdahl’s law that now bears his name, proving that the serial fraction of a problem can shrink as a problem is scaled. As a result of his efforts, his work won Gustafson the inaugural Gordon Bell Award, and marked a watershed in persuading computer manufacturers to embrace large-scale parallel architectures.
As Senior Fellow, Chief Product Architect for AMD Gustafson’s efforts are once again focused upon his passion for accelerator technology, with an eye to maximizing performance-per-watt, in addition to looking for radical fundamental improvements to the way we use floating-point arithmetic.
It’s interesting to note that Gustafson’s fascination with energy-efficiency and low carbon footprint isn’t limited to just supercomputer design; he drives a Tesla Roadster running on solar panels, and spearheaded a research effort to develop novel electrical energy storage devices that could displace batteries.
On the more personal side, Gustafson is a former trampoline gymnast and an accomplished keyboard musician, both on piano and harpsichord. (he built his first harpsichord at age 16 with his own hands!) An avid self-confessed video gamer, Gustafson also appreciates good wines, the arts, and enjoys raising exotic orchids.
John’s Top 5 HPC initiatives or technologies to watch in 2013:
- Education, particularly in how we teach parallel computing – 2013 may be the watershed year when we stop relegating parallel programming to graduate-level courses and universities treat it as mainstream approach that needs no apology.
- Renewed attention to answer quality, driven by the challenges of exascale – We are attempting ever more ambitious calculations with ever less attention to numerical analysis, and those two trends are on a collision course.
- Heterogeneous systems – (like those that use graphics accelerators) will spread throughout the TOP500 and influence the GREEN500 rankings. The result will be interesting battles regarding the best programming environment.
- What’s after MPI? – Now that systems are approaching a million processor cores, MPI is starting to look a little long in the tooth. Next to energy-efficiency, the biggest barrier to exascale is finding the programming environment appropriate for such a massive number of cores.
- Memory technology – Watch for truly disruptive things to happen to the entrenched DRAM market, and to flash memory. Several promising technologies are about to go mainstream after they solve problems that have prevented adoption in HPC.