Sign up for our newsletter and get the latest HPC news and analysis.

Power to the people

Speaking of the need to push HPC down into the underpinnings of every bit of science and engineering research in this country (which I seem to do every day), The Register is talking about PeakStream

The beta version of PeakStream Workstation for Microsoft Windows serves the same function as its existing Linux counterpart. Developers can use the PeakStream software to craft their own applications for mainstream multi-core chips and the oncoming wave of specialized mutli-core chips such as the Cell processor or GPGPUs (general purpose GPU). PeakStream’s main claim to fame in this field is that it removes some of the coding headaches developers face as they approach more demanding chip hardware.

So far, the company has zeroed in on the high performance computing (HPC) market and is trying to convince the HPC crowd that using the PeakStream Platform is a better option moving forward than trying to write homegrown multi-threaded code.

With this release developers get C/C++ math libraries and a hook into Visual Studio (M$oft’s development tool) all wrapped in a high level API.

Traditional HPC people recognize the phrase “high level API” as code for “you lose all control of the performance gating features of the code and you’re going to be sad.” The company claims

PeakStream does confess to a “couple of per cent overhead performance hit” as a result of code going through its runtime virtual machine.

Even if they are wrong by, say, a factor of 10, I say that products like this are important products for entrants into HPC. Once they have jobs running on supercomputers and see some benefits they can make an educated decision based on real experience with their very own application about whether they need to invest the time in training to get that last “few percent overhead” back.

Trackbacks

  1. [...] You’ll recall that the company makes development software that turns single threaded software into multi-threaded software ready to run on GPGPUs and multicore chips (we’ve covered PeakStream before). [...]

Resource Links: