Group items tagged

Finally a massively multi-core general-purpose architecture.

Well, nice, but I wonder how much cache per core will be available... With 48 cores a single memory bus becomes nothing more than one big (small? :) ) bottleneck.

Apparently they have separated L2 cache per-tile (i.e., every two processors) and a high speed bus connecting the tiles. As usual, whether it will be fast enough will depend from the specific applications (which BTW is also true for other current multi-core architectures). The nice thing is of course that porting software to this architecture will be one order of magnitude less difficult than going to Tesla/Fermi/CELL architectures. Also, this architecture will also be suitable for other tasks than floating point computations (damn engineers polluting computer science :P) and it has the potential to be more future-proof than other solutions.

who wants 100 cores ... future of PAGMO?

The problem with this flurry of activity today regarding multicore architectures is that it is really unclear which one will be the winner in the long run. Never understimate the power of inertia, especially in the software industry (after all, people are still programming in COBOL and Fortran today). For instance, NVIDIA gives you the Teslas with 10000 cores, but then you have to rewrite extensive parts of your code in order to take advantage of this. Is this an investment worth undertaking? Difficult to say, it would certainly be if the whole software world moves into that direction (which is not happening - yet?). But then you have other approaches coming out, suche as the Cell processor by IBM (the one on the PS3) which has really impressive floating point performance and, of course, a completely different programming model. The nice thing about this Tilera processor seems to be that it is a general-purpose processor, which may not require extensive re-engineering of existing code (but I'm really hypothesizing, since the thechincal details are not very abundant on their website).

Moreover PaGMO computation model is more towards systems with distributed memory, and not with shared memory (i.e. multi-core). In the latter, at certain point the memory access becomes the bottleneck.

As the researchers desribe, the aquifer found is defined as a basin filled with aged snow (firn) that was saturated with water trapped within its porous structure. More ice cores to study! But in that case the time sequence along the length of the cores - to be extracted- would not be monotonic, as was the case in the dry cores we analysed. That's because there is constant input of surface water percolating through the ice-sheet to reach the depth of the aquifer, and since it flows downward its temperature could affect its partial pressure => its vertical position along the core

"Her goal was to design and synthesise a super capacitor with increased energy density while maintaining power density and long cycle life. She designed, synthesised and characterised a novel core-shell nanorod electrode with hydrogemated TiO2(H-TiO2) core and polyaniline shell. H-TiO2 acts as the double layer electrostatic core. Good conductivity of H-TiO2 combined with the high pseudo capacitance of polyaniline results in significantly higher overall capacitance and energy density while retaining good power density and cycle life. This new electrode was fabricated into a flexible solid-state device to light an LED to test it in a practical application. Khare then evaluated the structural and electrochemical properties of the new electrode. It demonstrated high capacitance of 203.3 mF/cm2 (238.5 F/g) compared to the next best alternative super capacitor in previous research of 80 F/g, due to the design of the core-shell structure. This resulted in excellent energy density of 20.1 Wh/kg, comparable to batteries, while maintaining a high power density of 20540 W/kg. It also demonstrated a much higher cycle life compared to batteries, with a low 32.5% capacitance loss over 10,000 cycles at a high scan rate of 200 mV/s."

Apparently the core does not only produce a magnetic field - it is also an electric motor!

38 degree core body temp in microgravity stabilised after 2 months - due to reduced conv. heat transfer+evaporation https://www.nature.com/articles/s41598-017-15560-w

so that's hypopyrexia (augmented concentrations of interleukin-1 receptor antagonist) AND hypothermia (convection/evaportation)? what puzzles me is that temperatures take so long to return to baseline after astronauts return to earth.

once more ... the Romans did it already 2000 years ago! this time: burning so much wood that they increased the methane level in a way we can still measure it in ice cores ...

NASA has it ... I WANT IT TOO!!!! 240 threads on 60 cores ... Imagine the possibilities of this new toy!! Francesco also has it in his new "kill the seals" job

towards a better understanding about the fusion processes in the last moments of a star's life.

Nice opinion piece by the creator of C++ Bjarne Stroustrup. Substitute "industry" with "science" and many considerations still apply :)

"for many, "programming" has become a strange combination of unprincipled hacking and invoking other people's libraries (with only the vaguest idea of what's going on). The notions of "maintenance" and "code quality" are typically forgotten or poorly understood. " ... seen so many of those students :( and ad "My suggestion is to define a structure of CS education based on a core plus specializations and application areas", I am not saying the austrian university system is good, but e.g. the CS degrees in Vienna are done like this, there is a core which is the same for everybody 4-5 semester, and then you specialise in e.g. software engineering or computational mgmt and so forth, and then after 2 semester you specialize again into one of I think 7 or 8 master degrees ... It does not make it easy for industry to hire people, as I have noticed, they sometimes really have no clue what the difference between Software Engineering is compared to Computational Intelligence, at least in HR :/

Graphic cards in the core of a supercomputer.

Yet it does all this with a thermal design power of around 2W - incredibly, less than three per cent that of an everyday Core 2 Duo. Average power consumption is promised to be in the milliwatt range, with idle draw as low as 30mW.

The portable nuclear reactor is the size of a hot tub. It's shaped like a sake cup, filled with a uranium hydride core and surrounded by a hydrogen atmosphere.

In a work published on Nature (http://www.nature.com/nature/journal/v517/n7536/full/nature14090.html#affil-auth) a new DFT based simulation of convection in Earth's Core iron shows that electron-electron scattering has a similar contribution to electron's thermal vibration. The outcome is that using the old dynamo theory the simulation matches the Earth magnetic field experimental results, solving an 80 years old puzzle.

Yay to science! I'm always intrigued by related experiments that try to measure material properties at the GPa range. Especially, the efforts of reaching 'metallic hydrogen' (http://en.wikipedia.org/wiki/Metallic_hydrogen), requiring pressures above 25GPa at which hydrogen becomes conductive. It is thought that gas giant planets could have such a core, but no-one has been able to produce/verify this theory as off yet.

I got a fever. And the only prescription is more cat faces! ...../\_/\ ...(=^_^) ..\\(___) The article sounds quite interesting, though. I think the idea of a "fake" agent that tries to trick the classifier while both co-evolve is nice as it allows the classifier to first cope with the lower order complexity of the problem. As the fake agent mimics the real agent better and better the classifier has time to add complexity to itself instead of trying to do it all at once. It would be interesting if this is later reflected in the neural nets structure, i.e. having core regions that deal with lower order approximation / classification and peripheral regions (added at a later stage) that deal with nuances as they become apparent. Also this approach will develop not just a classifier for agent behavior but at the same time a model of the same. The later may be useful in itself and might in same cases be the actual goal of the "researcher". I suspect, however, that the problem of producing / evolving the "fake agent" model might in most case be at least as hard as producing a working classifier...

This paper from 2014 seems discribe something pretty similar (except for not using physical robots, etc...): https://papers.nips.cc/paper/5423-generative-adversarial-nets.pdf

Yes, this IS basically adversarial learning. Except the generator part instead of being a neural net is some kind of swarm parametrization. I just love how they rebranded it, though. :))

interesting workshop on language here in Leiden ... Luis, Giusi??

interesting indeed, but the lectures I'd be interested in already took place

No good, 80% of these people are top quoted in my research area... I really should have been there. Evidently the organisers need to work on advertising.