Last week, Gaussian Inc. started publicly talking about its GPU port of its Gaussian code. This is as conservative a development company as you will find. I know many other companies with ports (I won’t violate NDAs, which I’ve signed with a number of folks who post/comment here, and who read these … feel free to post a note/link to your accelerated app).
We’ve seen the early adopters come and stay. Mid market adopters come and grow. Late (conservative) adopters test the waters.
This is a good thing. Really a good thing.
I know there is some concern with going CUDA versus something portable across all platforms. I don’t know the uptake of PGI’s tool set or of PathScale’s HMPP tool set. I do think something very much like the PGI tools is the way to go for ease of programming (HMPP is nice, but the PGI approach looks simpler to use … just my opinion right now).
Which leads me to SC11. What will we see there?
First, accelerators are mainstream. Show a cluster w/o an accelerator and you might be in a rapidly decreasing minority.
Second, clouds are in ascendance, though we are early in the process.
What’s interesting is that what I was noting in 2009 seems to be coming to fruition. In some sort of terrible yin-yang scenario, I have a good track record of prediction, and a bad track record of raising capital to build something to take advantage of this. Ugh.
SC11 will be interesting. Listening to what people are working on for next year should be quite enlightening. Not sure if we are going to see anything profound here (last year was light on such things as well). But the stage is set for someone to do something crazy. And having met some really bright folks at Calxeda in Texas last week, yeah, I could see a few … interesting … things come out.
Also, I don’t expect to see as much Microsoft there. HPC has been borged into another division, one with a real P&L requirement, and unless they are generating a noticable amount of revenue, I don’t expect more more than maintenance releases going forward. Microsoft is still trying to figure out what it wants to be. Its biggest challenge is on the client side. This is its greatest threat right now. And its one it needs to address. I don’t see HPC being fundamental to it anymore (as if it ever really was). And this gets to all the change of business points I’ve made in the past. If you build a dependency upon a product that a company isn’t really committed to, and the pull the rug out from under you, where does that leave you? This is an annoying issue for locally installed apps, but survivable. What if the online vendor goes away and takes your data/computing with it? I’d like to see things address this. We need this for HPC.икони