My experience with ScaleMP is

not very good. I tested software which ran beautifully on a 24 core (6 quad-core) opterons (18-22 x speed-up) crunching through multi-scale analysis of a 1.2 gigapixel image in 60 seconds. On a 64 core (4x 16 cores if I am right) ScaleMP box performance was DISMAL. As more threads are added, the performance tends to drop severely. On a single thread I would get a timing of say 60 seconds for a smallish data set, on 2 threads it took anything up to 5 minutes. The scheduler NEVER puts two threads of the same program on a single board, but scatters them far and wide. You can only gain speed up on these boxes if you have many light-weight processes which do not need to share much memory. Did we not have clusters for that?

Actually, in most embedded systems

you learn to program with energy efficiency in mind. Talk to guys doing sensor-network coding, they design communication protocols specifically with ULTRA low energy consumption in mind.

Other than that, it is just a case of efficient coding. If I can reduce the number of instructions by an order, I save energy by the same order. Lazy programming and reliance of Moore's law to do YOUR work of getting things running quickly causes these problems.

After all, we now assume you need the computing power of a Cray Y-MP to edit letters or do some spreadsheet work effectively.

A patent on placement of objects on a screen?

only in the US

Nice one Neelie

Hit them where it hurts

Ah, the mythology of data mining

"If we have enough data, the problem will be solved"

You want information, not more data. As more data are added, the problems of finding what you need get harder and harder. It's like Pratchett's omniscope: becase it can show you everything, it is practically useless at showing anything in particular.

THINK before you add more sensors and more data.

Pedant alert

The REST mass of the protons was way smaller than the B+ (disappointing grade to some ;-) )

The sum of the masses they had in the lab frame of reference was as high or higher than the B+.

Great fun nonetheless

Back in charge

KZEEEEERT!

we need a cattle-prod icon

Ah, it's Friday

Ye cannae break the laws of physics capt'n II

No Nokia, you cannot take deep-space photos with your camera phone, and you cannot replace a 400 mm F5.6 APO telephoto with a lens that fits into a thimble.

The guy probably believes the image enhancement effects in CSI.

I get so tired of this sort of idiots (highly paid at that). For those still doubting:

The first issue is all about photon counts. The noise in an image is determined (apart from detector noise) most fundamentally by photon noise. Because emission and detection of photons is a random process, the noise is equal to the square root of the number of photons detected. So if a pixel captures 100 photons, the expected noise is 10. If you capture 10000 photons the noise is 100, but the signal-to-noise ratio (S/N) is 10 in the first case and 100 in the latter. S/N is very important in image quality. This is why astronomers want BIG scopes, because doubling the diameter quadruples the amount of light, and doubles S/N.

No matter of post-processing can alter these facts. I teach computer vision and image processing at the University of Groningen, and have worked quite a bit on developing ways to counter the effects of noise.

The second issue is resolution: the limit of resolution is determined by the ratio of wavelength to aperture. This is why we are building a synthetic aperture telescope HUNDREDS OF MILES ACROSS for long radio wavelength. There used to be al sorts of wild claims on what deconvolution methods could do, but in the field it is now accepted that whilst you can enhance details that are faint (at the risk of increasing noise) you cannot reconstruct information that has simply been lost at the aperture of the lens.

Ig Nobel prize!!

worthy at least

when will they start

colliding aircraft carriers?

Lends a whole new meaning to

having them by the short and curlies