In the big picture HPC hardware is a small part of the economic equation. What are the big parts?
When I’m talking to people about HPC I like to make what I call “provocative over generalizations.” My intention is to stimulate discussion and maybe get people to look at something from a different angle. One of my favorites line sounds like this, “You know in a sense, HPC hardware is now free.” Some curious looks, then someone says, “your crazy.” Then it is my turn. I qualify my statement with the fact that if an organization spends five or even six figures on a cluster, that is cheap compared to what a supercomputer cost in the past. And, it is cheap compared to the cost of the people required program and run the thing – which is the point of my argument. Some discussion of “What do you mean by cost?” usually follows. I have of course neglected infrastructure cost, but by this time they have taken the bait and we are talking about how you can put a large amount of computing power into someones hands real cheap these days.
Indeed, with the advent of the GP-GPU you may already have enough GFLOPS in your desktops to design your own jet. Even if you do not have have a pile of GFLOPS in your PC or basement, if you wanted them, you could get them cheap. (I am not going to mention any names, but there are those who have a fully populated 19 inch equipment rack in their basement or garage. Mine is in the basement.)
The point I am trying to make is that if HPC FLOPS are so cheap, then why isn’t everybody running out to buy hardware and solve important problems? In the past, I have bemoaned the software issue and made other over generalized statements about parallel computing and multi-core. I often state things like “what was hard just got harder.” Yes it is hard, but not impossible. There is then something else that keeps that “free hardware” away from the masses.
If you said “software applications” you are half right. Applications drive “free hardware” sales and if you do your homework, you may even create an appliance and retire early. I believe the other issue is a market disconnect of some type. Or put another way, there are many places HPC is valuable, but the practitioners have no idea what is possible or how much it will cost.
The largest headlines for HPC are often in the educational and government sectors. These are the pioneers who have shown us the way forward. One other area that is ripe for HPC is the commercial sector. Of course, you have your large R&D clusters at many organizations, but I am talking about those HPC efforts aimed at making or saving money. And, I am not talking about Wall Street I am talking about manufacturing. Procter & Gamble has shown that HPC contributes to the best part of waken’ up ™. Want more proof, then take a look at the Council on Competitiveness HPC pages.
The world of manufacturing is replete with opportunities for HPC. The range of opportunities is immense, from better products to the lowering of manufacturing costs. Many large companies have identified HPC as a strategic necessity. The large companies certainly have the incentive to lower costs given the magnitude of savings that can be achieved. What about the small and mid-size companies? How can the take advantage of the “free HPC hardware?”
The answer of course is education and cough, cough, marketing. Yes, there is a message that needs to be developed and broadcast far and wide.
“Hey small and medium sized business. Do you want to be more competitive and have access to the secret weapon the big guys have? How about saving money, going green, and building better products. We have just what you need. And the hardware is free (sort of).”
As I mentioned, the big cost in this proposition is people. There are all kinds of companies that will gladly take money from other companies for the sole purpose of telling them how to run their company better. I always wonder if the the companies who make a living telling companies what to do, hire companies to tell them what to do. Oh, how I love recursive to the absurd humor. In any case, the results promised by these companies, I would assume, are based the bottom line (or at least they should be). The great thing about HPC is the results are easily measured and quantified. There are business opportunities across the market. Those companies that can deliver the middle ground, that space between the production line and the cluster or GP-GPU, will be well rewarded.
Indeed, what if there was product that would allow a detailed HPC simulation of an entire business structure? What if such simulations could be used to aid in decisions about a companies future. I’ll even stake out my own trademark Business Matrix™. I not advocating a large spread sheet, although the interface may look like one, but rather a detailed simulation that includes information flow and decision processes within a company and even tracks the effects of current events on productivity (i.e. everything from weather to productivity on the day after the Superbowl) HPC has already been employed to help understand Pandemic flu outbreak in the US, why not use a similar technology to help companies get some insight into the “What if” questions. Of course, I am not advocating blind allegiance to some new business simulation overlord, nor am I suggesting some kind of Wall Street “Value at Risk” dodge, but rather developing working models of real businesses. Sounds complicated and hard, like parallel programming. The pay-off would be huge, however. With a big payoff may come big innovation — like better ways to program all that “free hardware.”
Douglas Eadline is the Senior HPC Editor for Linux Magazine.