perspective At the extreme-performance end of HPC (high-performance computing), the supercomputing crowd can be a strange bunch in some ways. As with all HPC users and buyers, they chase goals such as ever-increasing performance per cost and power efficiency.
But their aim is not to streamline their HPC budgets. Give them greener computers, and they'll just see an opportunity to get more of them into the datacentre. Commodity clusters are a way to get more performance for the same capital spend, not to reduce the capital spend.
Indeed, the capital and operating budgets of supercomputer centers seem to be perpetually increasing.
Supercomputer budgets have been increasing as long as I've been involved in the industry. Every procurement cycle, planners demonstrate the need of their users for finer resolution, more complex physics and chemistry, better statistics and more detailed analytics--and write the business case justifying a bigger budget than last time.
The case is usually straightforward. HPC has proved itself too many times in too many industries for there to be any doubt about its value. Businesses can usually point to a reduction in the product-development cycle due to the use of HPC, or to a better product, or a cheaper one.
Research centers can show the improvements in, say, medical science, energy technology or climate modeling that more powerful simulations will bring. In the science arena, there are applications waiting for thousands of times more compute power.
And so the well-justified, increased budget is usually signed off, and the procurement of the next, bigger supercomputer begins. But can these recurring budget increases continue?
First is the challenge of getting useful work out of increasingly big and complex supercomputers--hundreds of thousands of threads of parallelism with hierarchical structure have to be coordinated in a leading-edge simulation. I addressed that, in part, in an earlier article, so no more here.
Second is the power cost--and this is the killer. I mentioned earlier that greener supercomputers just enable the centers to cram more in. But a follow-on effect is starting to occur.
Spending several million pounds capital, followed by £100,000 (US$148,190) per year on electricity, is not going to cause concern. But we are heading towards the point where the center will have to admit to their funding agency or board that, over the course of the life of the latest supercomputer, the largest share of the money in the total cost of ownership will go on electricity.
In the next couple of years, a budget of a few tens of millions of pounds will buy a supercomputer that will consume £10 million (US$14.8 million) per year in electricity. Psychologically, that's a very different sell to whoever is funding the project. Approving a large budget for electricity, with a moderate capital outlay for the supercomputer to consume that electricity, will be problematic, however great the benefits in the business case.
Buying a world-class facility that improves business output makes those who hold the purse strings feel good. But committing to massive electricity requirements in an environmentally aware world, with uncertain energy costs, will make them nervous.
The third threat to budget increases is publicity. Taxpayers or shareholders will probably not notice or worry about a project cost of a few million pounds. But a £100 million (US$148 million) project will cause those shareholders or taxpayers, as well as the budget holders, to ask: "Why?"
Explaining HPC to those intimately involved in the business, such as the board, or to the community, in the shape of the funding agency, is not always easy, even with proven benefits. But trying to explain to the public or to distant shareholders what HPC is and why it is a good investment is a much harder proposition.
However, undertaking such an explanation will reap far greater benefits than just getting the investment through. It will evangelize the contribution of HPC to a much wider audience. It might inspire more to consider the area as a career. It might spread the use of HPC more widely. All those results will help make the costs and use of HPC more viable for the future.
Showing what HPC can do and how it contributes to your business is not a nuisance step in the investment process--it is an opportunity to improve the future of HPC itself.
As vice president of HPC at the Numerical Algorithms Group, Andrew Jones leads the company's HPC services and consulting business, providing expertise in parallel, scalable and robust software development. Jones is well known in the supercomputing community. He is a former head of HPC at the University of Manchester and has more than 10 years' experience in HPC as an end user.