Tell me more ×
Unix & Linux Stack Exchange is a question and answer site for users of Linux, FreeBSD and other Un*x-like operating systems.. It's 100% free, no registration required.
up vote 0 down vote favorite

Here is a bundle of questions I've been conjuring up during my stay at the SC11 Supercomputing conference this year:

  1. Why is there a saying that "we won't be able to enter the era of exascale computing just by adding more and more Xeons"? Why do we need the GPU/Accelerator architecture to achieve what we want?

  2. Why can't compilers automatically map high-level code to new architectures? For instance, instead of having to learn CUDA to reprogram all your applications to run in massive parallel on nvidia GPUs, why not just write in C, and let the compiler researchers do the job of developing a compiler that will conveniently parallelize the program for you? Maybe that was a bad example, and if anyone wants to provide their own, that's certainly welcome :)

  3. What makes a good compiler a good compiler?

  4. How does High Performance Computing affect the economy, and why is there such a big drive (especially between the three countries China, Japan, and the USA) to create marginally faster supercomputers over each other year after year?

  5. What draws the line between a 'supercomputer' and a 'cluster', or is there not really a distinction?

Any other personal thoughts or ruminations to add on top of these are welcome.

share|improve this question
I don't see the connection with unix. There is possibly a question for Programmers struggling to get out here, but do not repost this in its present form. Your bullet points 1 and 2 might make good questions if you post them separately and polish them. 3 is far too vague, I don't think 4 would be on-topic (not technical enough), and 5 is a matter of terminology. – Gilles Nov 18 '11 at 10:24
2  
P.S. For 1, there's a marketing element: these are people with an interest in GPUs. But there's also a technical point, that GPUs are geared towards parallel computations (but can't easily cope with high memory requirements) while current CPUs are good at memory management but don't scale much on parallelization. For 2), automatic parallelization is not possible except in very simple cases, that's what decades of research shows. If you “just write in C”, you get an intrinsically sequential program. – Gilles Nov 18 '11 at 10:29
Yes, but shouldn't the point of an automatic compiler be to change an "intrinsically" sequential program into something that's parallel? – Dark Templar Nov 22 '11 at 2:19
2  
Yes, and the point of a fusion power plant is to generate cheap, clean energy. We are far from knowing how to make either. – Gilles Nov 22 '11 at 8:11

closed as off topic by cjm, Mat, Gilles, Michael Mrozek♦ Nov 19 '11 at 0:32

Questions on Unix & Linux Stack Exchange are expected to relate to Unix or Linux within the scope defined in the FAQ. Consider editing the question or leaving comments for improvement if you believe the question can be reworded to fit within the scope. Read more about closed questions here.