Oh So Boring…

Not the Clash song, per se, but the question itself. Should I stay or should I go…from the ACS and APS.

For many moons, I’ve been a member of both the ACS (American Chemical Society) and APS (American Physical Society), the former for ten or more, the latter for a few.  As a theoretical chemist, having a membership to these societies was quite useful. You get a subscription to a very good “popular” magazine, Chemical and Engineering News and Physics Today, discounted registration at conferences, and, perhaps most importantly, access to the respective job banks.

All this, though, comes at a cost. When you are a grad student and recent postdoc, the membership cost isn’t too expensive and the benefits well outweigh the costs.  But once you are a “professional member” the costs double–I think–at least to $140 and $118 per annum.

Okay, that’s not bad, not great, but not horrible. *But*, I am no longer a practicing chemist/physicist. I’m a code monkey. A membership to IEEE or ACM would make more sense, really. Heck, I’m kind of attending SC09, the supercomputing conference, next weekend.

But, but, but. I am a chemist. I am a physicist (according to Feynman, no less). I still think of myself as such and I suppose a part of me thinks one day I might be a practicing chemist/physicist again.

I’m not sure what to do.

So, a little over a month ago I started my new job at SAIC, at NASA.  How’s it been?  It’s been up and it’s been down.

The first week was tough. That much I won’t argue. Not only was there the stress of a new job, there was the stress of torrential rains and a new, long commute. So…yeah, but that week is done.

After that, well, I started getting into a flow.  Much of the first days were spent just getting used to Goddard, the computational facilites, heck, just getting an email address. Add to that the getting used to SAIC and that culture.  Timesheets for the first time!

But, in the last week or so, things seem to be looking up. I have a plan for my task. I know what I’m (supposed to be) doing. In short, I’m looking into the use of NVIDIA’s CUDA to speed up calculations of a global atmospheric model.  Sounds fancy, eh?  Right now it’s mainly me trying to get our Tesla testbed able to compile things. And learn how to use video cards to speed up calculations. Not easy, quite a challenge in fact.  Having to rewire my brain to think in a different way.

But, in my quest to learn CUDA, stream processing, uber-multiprocessing, etc., I did have one success this week.  One that I’m proud of.  This week, in order to test out PGI’s new Accelerator technology (think easy-CUDA: add a couple of lines and speed up your code by 50-60x).  So, I followed some tutorials online and made up some simple matrix multiplication programs.  The simplest did about 1 GFLOPs, using OpenMP to use 4 processors got me up to about 4 or 5.  Then I used PGI’s Accelerators and got 80 GFLOP.  I mean, eighty billion floating point operations aren’t bad.  Especially as it was just two extra lines.

But, I knew there was a way to get more speed.  What my routines were doing were, in BLAS parlance, SGEMM. Single-precision GEneral Matrix Multiply. And NVIDIA has a GPU-optimized library called CUBLAS that can get much, much better performance.  The problems? Well, this library is mainly used in C and there aren’t many Fortran examples online.  And the ones that are, I learned, are *not* good.

So, I spent a few days, had some struggles, and…yay! Got it!  With CUBLAS I was able to get up 350 GFLOPs with just one video card.  That’s a nice number to see.  And, well, I’m kinda happy I figured that out.  So work ain’t all bad.

This week at least.

Of course, folks, this is just work.  My personal life? In many ways, work has been as joyful as tree full of monkeys on nitrous oxide compared to that. But the paean to loneliness and malaise that will be a post on personal life…that’s for another time.