I do lots of scientific computing (radio astronomy signal processing) on consumer-level NVIDIA cards and it is *all* FP32. These kinds of cards (and the 12GB of mem) could be very useful for my work, and others who do DSP-type of *scientific* computing.
Yet another
* back to lurking *
I've read slashdot daily since it started -- it has been a fantastic source of information, ideas, and laughs. Good luck with all your future endeavors.
Scott
All of the energy that we see (as well as the energy we don't see, which is the vast majority of it and which comes out in a relativistic particle wind) comes from the rotation of the neutron star. That means that pulsars are flywheels. And amazingly (even to me, and I study them daily), the most energetic pulsars give off tens of thousands of times more power than the total power output of the Sun. And all from rotation. That's crazy.
Damn the Universe is cool.
Actually, they aren't using the GBT's spectrometer. They are using an instrument that I helped to develop for pulsar research called GUPPI, which uses FPGAs and GPUs to real-time process 800MHz of radio bandwidth.
However, in this case they are using GUPPI's GPU nodes to record 800MHz of Nyquist-sampled band centered at 1.5GHz. Each sample is 2-bits, and with 2 polarizations, that is how they get 800MB/s (or almost a GB/s as it says in the article).
If you want some basic info about GUPPI, you can find it here:
I'm certainly one of them (thanked him for Cosmos in my PhD thesis specifically).
And I agree with you about Demon-Haunted World. I think that should be required reading for all high school students.
The good thing is that the pulsars which glitch are the young ones (hundreds to millions of years old). The pulsars that we are using for NANOGrav are millisecond pulsars which are hundreds of millions or billions of years old, have much smaller magnetic fields than young pulsars, and basically never glitch. They are extremely stable rotators -- much better than normal pulsars.
LaTeX is certainly the standard in physics and astronomy. Of course your point about Unix workstations is correct, as most physics, CS-types, and astronomers use Unix/Linux all the time.
There is no compelling science case for Arecibo that can't be pursued with other telescopes, especially since the frontier of radio astronomy has mostly moved from sensitivity (requiring big apertures) to resolution (requiring long-baseline arrays), or to shorter mm/submm wavelengths that Arecibo can't handle.
Sorry, but that is not true. Radio astronomy needs improvement in a wide variety of areas in order to tackle the tremendously wide variety of science that is done at radio bands. Examples include sensitivity, field-of-view, dynamic range, image fidelity, resolution, and wavelength coverage. But sensitivity is one of the most important. That is why the SKA is on the table to be the world's next generation decameter/centimeter wave radio telescope. The most important thing it provides is sensitivity (i.e. SK = square km = sensitivity). And Arecibo is already a 5-10% SKA.
For my own research (pulsars), Arecibo's sensitivity is what sets it apart. Although, truthfully, the fact that it can't observe any of the southern sky (where most of the pulsars are) is a definite downside.
Finally, you mention surveys and imply that because Arecibo is doing a larger percent of them now that that means it is washed up. However, that also isn't true. Modern astronomy is driven by large surveys (including several of the instruments that you mention, for example, Sloan, PANSTARRS, LSST) as they dramatically increase our discovery space.
Hmm. Doesn't make full sense to me this way.
First part of your explanation is OK - let's say we need to unroll a loop for efficiency. And we decide to unroll it so it has 8 statements inside the loop. So, assuming count>0, we would do something like this (using dots to show indentation because "ecode" eats up leading spaces):
void copy(char *from, char *to, int count) {
..int n = count / 8;
..int i;
..if (n > 0) {
....do {
......*from++ = *to++;
....../* 7 more times (code omitted to avoid slashdot lameness filter) */
....} while(--n > 0);
..}
../* Handle the "leftovers" */
..for (i=0; i<n%8; ++i) {
....*from++ = *to++;
..}
}
That's a standard loop-unrolling for you. Duff's Device just shortens it. How? The switch statement jumps into the loop to a point where we first handle the "leftovers", and then we execute the loop (count/8) times, each time copying 8 elements, thus copying (count/8)*8 elements in the loop (which, because of integer division is not equal to count unless count%8==0, that's why the leftovers).
Thus, the part where you talk about count being excatly divisible by 5, 6, 7, etc., is wrong. The switch statement only checks the remainder of count when divided by 8.
This may be wrong too. In that case, please feel free to correct me.
In case of atomic attack, all work rules will be temporarily suspended.
