which indicates that 'for reasons of national security' one RTG-worth of plutonium-238 had been reclaimed from NASA about five years ago.
There are various national-security applications for plutonium-238 - it's perfect stuff for powering, for example, bits of equipment to sit in a cave in Afghanistan or next to an undersea cable off Taiwan quietly recording all that passes to be collected later; it lets you build satellites without shiny solar panels. Lincoln Experimental Satellites 8 and 9 used RTGs; http://www.osti.gov/energycitations/product.biblio.jsp?osti_id=5562667 is an abstract claiming there are ten American satellites in Earth orbit with RTGs on board, though I rather doubt it will list names and purposes.
Except for Blue Gene (and that's a big exception, since the 2^17-CPU Blue Gene at LLNL pumps out about a billion processor hours a year on its own, and has had another 80,000 CPUs added recently) these CPUs are either 2.4GHz Opterons or 3GHz Core2s; those were and are the sweet spots for building big HPCs.
They seem mostly to be about fire, in power stations, in supernovae or in fission reactors.
Some nice examples: 27Mhours on lattice quantum chromodynamics, 21Mhours to simulating thermonuclear burning in type-1B supernovas, 18Mhours to figuring out how biofuels burn, 17.5Mhours to determine from first quantum principles how the nickel-56 nucleus holds together, 16Mhours to simulating thermonuclear burning in type-2 supernovas, 12Mhours to attempting to design a carbamate hydrolase enzyme de-novo, 10Mhours to simulating lead-telluride / silver-antinomy-telluride thermoelectric materials, 4.5Mhours to optimise the design of the next-generation linear collider, 5M hours to figuring out why enormous temperature gradients persist in liquid-sodium-cooled fast-breeder reactors and a further 14Mhours to liquid-sodium reactor design in general, 4M hours to figuring out exactly how multiple burners in large power-station combustion chambers light one another, 3.5Mhours to trying to understand why it's so hard to hydrolyse cellulose, 3.5Mhours to understanding how flame fronts move in the complicated gas mixtures obtained from coal gasification, half a million hours for oceanic circulation, three quarters of a million hours for flow of dense suspensions, ten million hours on catalyst design.
And, for some reason, a million hours on porting Plan-9 to the Blue Gene system. I presume this allows you to crash and reboot the entire 200kcpu system enough to identify ten bugs. Also eight million CPU-hours to developing better HPC libraries.
I would be interested to know the amount of idle time there is on these supercomputers; a friend of mine from mersenneforum.org got a week on several hundred Opterons in France over Christmas, which was enough to do most of the work required to factorise a few numbers of fairly unreasonable size - sadly, there's a second step in the factorisation which requires an SMP machine, and the biggest SMP machine I have is an Intel Q6600, so completing the factorisation is taking three weeks on a single desktop in my back bedroom.
Nuclear fusion is an absurdly poor source of He
on
Helium Crisis Approaching
·
· Score: 5, Informative
Yes, nuclear fusion produces helium.
The fusion of 1kg of deuterium produces near enough 1kg of helium, and, umm, 2.7MeV per fusion * 6*10^23 atoms per mole * 500 moles of D atoms per kilogram / 2 deuteriums per fusion * 1.6e-19 joules per eV = 64.8 terajoules of energy.
So, a one-gigawatt fusion power plant would produce a kilo of helium every eighteen days; if the current electricity use of France were provided entirely by fusion plants, you'd get thirty tons a year. The large hadron collider uses 120 tons of helium, but efficiently; present planetary helium use is about seventy-five tons a day.
For comparison, the US produces from natural gas about 76 million cubic metres of helium a year; a cubic metre of helium weighs 1000/22.4*4 grams, so 76 million cubic metres weigh about fifteen thousand tons.
You can buy helium from the US government at $2.037 per cubic metre, whilst the commercial price is nearer $3 per cubic metre; adjusting that would seem to make some kind of sense, since the US has 600 million cubic metres of the stuff in Amarillo.
There are plants at Skikda in Algeria and somewhere in Qatar which aim to extract 25 million cubic metres from natural gas a year, but there have been some issues in getting them to work; both Algeria and Qatar's natural gas reserves contain about as much helium as the US total reserves do.
It is impossible to substitute for helium for cryogenics; nothing else stays liquid at that low a temperature, and the ultra-refrigerators that get to liquid helium temperatures use helium as working fluid.
I did my PhD at Nottingham University, which uses a fair amount of liquid helium; the arrangement there is that it's delivered to the MRI building at the top of the hill, and the boil-off passes through a liquifier and is used by the theoretical physicists at the bottom of the hill. I don't know what the theoretical physicists do with their boil-off; there are obvious practical problems with running piping from lots of separate labs to a central liquifier, and liquifiers are bulky and vibrating enough that you don't want to have them in the same lab as your delicate semiconductor-physics experiment.
I got the impression that this was a matter of an author googling around a topic and not realising that the convincing page early up in the hits was written by a lunatic, then a number of editors deeply fed up of dealing with the consequence of this lunatic jumping on him. I could easily see myself doing the same.
There's a lot of dire technical-management-substitute involved, at least in part because there are no very useful side-channels: it's not clear that there's a way of saying 'umm, you realise that you're linking to a site written by a lunatic, maybe reconsider?'. Since editors have to spend most of their time dealing with people motivated by malice, it's forgiveable that they assume of malicious motivation...
I don't see it as a critical failing of an encyclopaedia that its governance structures are such as to bias its pool of contributors towards the stoic.
This is just another part of the slow-motion meltdown of overstock.com, and the 'naked short-selling' financial conspiracy theory. There seem to be a lot of financial conspiracy theories around at the moment, presumably since there is some degree of actual financial chaos in the background, and the things financiers have demonstrably got away with are crazy enough that it's difficult to reject conspiracy theories on the mere grounds of strangeness in appearance.
Disclaimer: yes, I write stuff on wikipedia, my handle is fivemack. Mostly I write about chemistry; it's pretty clear that wikipedia is the most comprehensive and reliable site for chemistry on the Web, since chemistry is advanced stamp-collecting and wikipedia is a superb medium for presenting stamps in multiple series. The science side of wikipedia is a wonderful resource, and doesn't seem too prone to the kind of lunacy that afflicts other parts of the encyclopedia; people have less heated feelings about the melting-point of tellurium or the carcinogenicity of tetramethylhydrazine than they do about whether Mount Ararat is a Turkish or an Armenian mountain.
This sounds like a nice engineering exercise - very small motors with adequate torque, very small angle encoders, elaborate chassis to contain it all with the right number of degrees of freedom. The interesting part is always the control software, and I can't work out from the article whether the robot is running software that can play the violin ab-initio, or playing back a sequence of moves figured out by infinite grad-student labour.
I'd probably guess the latter - 'can it play any other piece' and 'can it play _this_ violin' (offering one of a slightly different size - with the popularity of the Suzuki method, children's violins are made in a variety of sizes) would be the killer questions.
I've heard of robots (WABOT-2 back in 1985) which can play piano from a score - OCR on musical notation turns out not to be especially difficult - but the extra mechanical linkages in a piano mean that you can press the keys in almost any way and get something note-like.
I'm rather more impressed that Toyota have managed (http://www.toyota.co.jp/en/special/robot/) to get robotic lips to play the trumpet - that's a nasty problem in engineering active soft materials.
Yes, the Skynet project includes a huge amount of ground infrastructure - terminals in jeeps, terminals on ships, terminals in shipping containers, terminals anywhere that an MoD Procurement Executive employee could imagine a member of the armed forces needing a terminal. So that was an unfair comparison to use; sorry.
I suspect there isn't very much deep-black material on Skynet, it was constructed via a complicated scheme of industrial contracts as a showpiece of private-finance-initiative procurement, and I can't see the kind of people who have weird packages to fly wanting the exposure risk of having extra clauses in the contract detailing the packages. I've worked a small amount on projects near Skynet, but on the documentation and assurance side rather than anywhere near either money or bent metal - yes, in a past life I was a civil servant.
The DVB satellites are off-the-rack ones (in as much as satellites ever are) which are supported by infrastructure that's already there, so I'm a bit more confident about that $400 million figure.
Actually, http://www.solarstorms.org/Sinsurance.html has some interesting numbers - Intelsat 'have declined to purchase insurance for satellites costing less than $150M', suggesting that some of their satellites cost more than that.
'Comparatively speaking, the price of a satellite is usually something like 95% launch costs, 5% satellite design and build costs' is not, in fact, anything close to true.
Launch costs are about five thousand dollars per kilogram to low-Earth orbit, about ten thousand dollars per kilogram to geostationary; maybe a bit cheaper if you're launching something fairly small on a Soyuz, maybe a bit more expensive if you're launching something enormous on a Delta 4 Heavy.
The Skynet 5 program is building and launching three large comsats for the British ministry of defence for $7.2 billion, of which about a thirtieth is the cost of the three half-Ariane-5 launches. For straightforward digital-broadcasting satellites, you're talking $400 million of which a quarter is launch cost.
In case you haven't noticed, Russia doesn't get very near the equator; they built the original facility in Baikonur because that was as far south as you could get in the Soviet Union and have a reasonable region of Soviet Union over which to drop discarded rocket stages.
The southernmost points of Russia are in the Caucasus, but that's a decidedly unstable area of the world, and rocket stages dropped off by things heading east would drop on Kazakhstan, which the Kazakhs obviously don't want. If you rule out the Caucasus, the next-southernmost points are at the North Korean border in the far east; there is a constant Russian worry that the Chinese might want to expand into Siberia if it's left empty, and so they'd like to build facilities there, especially the sort of facilities which set up clusters of skilled people who'd bring non-resource-dependent income to Amur.
The proposed site is Svobodny, which is just over the Amur from China, and not too far from the Komsomolsk-na-Amur rocket factory.
Obviously, an equatorial site would be better, and indeed there's a Soyuz pad being built at Sinnamary, in French Guyana, five degrees from the equator and about twenty miles from the Arianespace facility at Kourou. First launch from there will be late 2008, but it's only Soyuz so not particularly heavy lift, and I suspect the Russians might be less keen than EU nations at having their military satellites launched from French soil.
It means that you will finish the maximum-likelihood refinement of a protein model against X-ray data in forty minutes rather than ninety, so can check the mode in which ligands bind to your protein kinase about twice as fast as you could on the slower computer. Which, literally, helps some of our customers find treatments for cancer more quickly.
(yes, of course the software is multi-threaded, the core parts are Fourier transforms and exponential maps on large 3D arrays, and those parallelise nicely).
I expect Intel to ignore it entirely. The SSE5 proposal rewrites quite a chunk of the instruction set and would require a complete redesign of the processor pipeline, and Intel's competitive position at the moment is such that there's no need for it to do the work; they will argue that, if you want that kind of vector processing, you should get one of their Larrabee vector processors.
There is no particular guarantee that AMD will survive long enough actually to produce any processors using this planned SSE5 instruction set; it's for the chip that comes after Barcelona, and Barcelona is a year late and much slower than anyone had planned.
16GB of memory is possible on any dual-Xeon motherboard using eight 2GB FBDIMMs, and on most dual-Opteron motherboards using eight 2GB DDR modules, though some of the older, cheaper dual-Opteron boards only attached memory to one CPU. Some systems (the Sun X4150 server and X6250 blade, Supermicro's X7DWN+ board) have sixteen memory slots and support 4GB FBDIMMs, though those are horribly expensive - it's much cheaper to go from 16GB to 32GB by replacing the motherboard and processor and buying eight more 2GB FBDIMMs than it is to replace eight 2GB FBDIMMs with 4GB FBDIMMs.
is something like the work being reported on; 'A 1 Gsymbol/s, 64 QAM coherent signal was successfully transmitted over 150 km using heterodyne detection with a frequency-stabilized fiber laser and an optical phase-locked-loop technique. The spectral efficiency reached as high as 3 bit/s/Hz.'
looks very plausible in the context of this work; 'coherent optical transmission' is I think the relevant buzz-word. Going from 1Gsymbol/s to 10Tsymbol/s is clearly a lot more work, but being able to do optical QAM at all is pretty spectacular.
It'll take only about a year or so; Samsung 8Gbit Flash chips are $9 on the spot market at the moment.
Though the interface for connecting eighty of them onto a single SATA channel wouldn't be completely straightforward just from the point of view of I/O pins; each chip presents eight data pins and nine control pins, so you need a total of about 1500 pins on the disc-controller ICs.
For the early market you'd probably use FPGAs, you'd need six XC3S400A in one of the larger packages to get enough I/Os, which would be another couple of hundred dollars; the master one can speak SATA directly and do a bit of buffering, and using this baroque configuration means you get the full data rate from each chip so reads are very fast. Say $1000 for 80G SATA-attached - but that's making mass-manufacturing assumptions, which are clearly false if what you're manufacturing are 80G disc drives for $1000.
Five hundred dollars for 80G SATA-attached sounds a reasonable sort of pricing once Samsung have got one more generation of Flash factory online, and once the devices are more plausibly consumer items.
But most people don't need a video card nowadays, it's integrated in the north-bridge chip on the motherboard, and the difference between a board with video on the motherboard and without video is about twelve pounds, for video a lot better than the G400MAX can offer. The way to get things really cheap is to lose the PCB and interconnects entirely, and put them on a chip that you'd have had to manufacture anyway.
Given that most proteins contain tryptophan, and tryptophan fluoresces under UV, and UV lasers are not that hard to come by, wouldn't it be easier to shine a UV laser at the crystallisation plate and detect by subtraction where the glowy bit is?
Or, as a lot of molbio automation companies are offering, actually shine an X-ray beam through the putative crystal onto a detector and see if it diffracts.
Fully automated high-throughput crystal growing strikes me as a bit of a boondoggle; the sophisticated robots required for the last steps of automation are an order of magnitude more expensive than having three shifts of trained Indian or Chinese workers moving plates around and looking through microscopes.
Generally these absolute-top-of-the-range chips don't go into supercomputers; for supercomputing, you do the price-performance calculation and find that more, slower chips suffice.
Though there are some applications where you need very large memories on the nodes, and if you're buying 32GB of RAM the difference between two 3.3GHz quad-core processors and two 2.4GHz quad-core processors is lost in the noise.
Tom's Hardware just did a series of power-consumption tests on various overclocks of a Q9650, the first available 45nm processor.
At 3GHz, it uses 8.79W when doing nothing, and 73W when running all four cores flat-out
At 4GHz, it uses 16.83W to do nothing, and 135W with all four cores flat-out; on the other hand this required a voltage increase to 1.44V from the 1.25V that sufficed up to 3.33GHz.
Fitting curves suggests that you would be using something like 350W for four cores at 5GHz, which is quite impressive.
But a 5GHz Skulltrail would be a chip hand-picked by Intel to run at lower voltages in general, and would be running cryo-cooled which I think also allows a lower voltage to be used; probably 200W would be a better estimate for the chip power consumption, though the cooler will be using a comparable amount of power. This is, indeed, moderately crazy.
There are four PDFs there; the brochure is a four-colour glossy, but there is some real information. Sadly, the interesting-looking white papers are for the SX6, two generations earlier.
SX9 summary: 65nm technology, 3.2GHz clock speed, eight vector elements handled per cycle with two multiply and two add units, which is where the 102.4Gflop/CPU figure comes from. 16 CPUs in a box about the size of a standard 42U rack.
Totally absurdly fast (ten 64-bit words per cycle per CPU) access to a large (options are 512GB or 1TB) shared main memory; absurdly fast (128GB/second) inter-node bandwidth.
I hadn't realized ESA was thinking of supplying the mirror for SPICA; I did an arxiv search for SPICA, and aside from the occasional confounding reference to Alpha Virginis, there was a paper on surface quality of carbon fibre / silicon carbide mirrors which made me assume that JAXA was planning to build one such.
[also a very nifty paper on coronographs using binary masks very close to the focal plane, which in principle masked out the star enough that you could almost detect Jupiters in reasonably distant orbits around reasonably close stars]
The mirror paper surprised me slightly, since they were testing the mirrors at 95 kelvin and assuming that the results could be extrapolated to liquid helium temperatures, which I'd have thought wasn't guaranteed.
You are, I suspect, the friend I was thinking of - Dave of Cardiff with whom I talked about infra-red astronomy at Recombination.
The Shuttle stack - the solid rocket boosters and the main engines - gets just under thirty tons of payload *and 68 tons of shuttle* into orbit, which is pretty close to the weight of the whole Apollo stack. In comparison, a big comsat might be seven tons, Envisat is eight tons, Hubble is eleven tons.
This is why the Shuttle-C project, which would have used the rockets from the Shuttle to launch something much more like a standard payload, was so appealing; you could launch most of a Space Station, or enough fuel for a manned mission to Mars, in a single shot.
Space missions aren't particularly resource-intensive; after all, the probes rarely weigh more than a few tonnes. They're expensive because they use a lot of staff time and rocket scientists aren't cheap, but about the only resource that they use which is in any sense rare is xenon for ion drives. Bits of space probe are quite often machined from solid ingots of relatively costly materials like inconel, but generally you can recycle the chips.
An Ariane 5 burns 25 tons of liquid hydrogen in 130 tons of liquid oxygen, and is assisted by about 480 tons of ammonium perchlorate and aluminium powder in the solid rocket boosters; the materials costs are trivial in comparison to the cost of the engineers who designed and assembled the machine. Liquid oxygen is significantly cheaper than milk (say 10 cents per kilogram), and 130 tons is much less than the daily consumption by even a modest steelworks; liquid hydrogen is cheaper than beer at about $4 per kilo.
It's interesting to see ESA proposing Galileo- or Cassini-type missions, given that NASA has rather missed the boat on those - they focussed on JIMO, which would have been a spectacular truly next-generation mission, but didn't have the political backing. Either mission would be a twenty-year commitment and cost several billion Euros, but the ESA is confident at the moment and the various 'Europe, centre of innovation' pushes by the EU will likely be able to find several billion Euros.
I wouldn't be surprised if JAXA flew Marco Polo whether ESA is interested or not - Hayabusa was a beautiful mission plagued by bugs, and Marco Polo would be a re-run with the bugs fixed.
DUNE/SPACE are close to duplicating a NASA effort called the Joint Dark Energy Mission; PLATO is very specialised and comes close to duplicating NASA's Kepler. I would bet on Xeus since XMM/Newton has been a great success, and ESA's reputation in X-ray astronomy is particularly good. Spica, again, I could see the Japanese flying without ESA input, though I suppose I should push for that since a friend of mine's working on the transition-edge detectors planned for it.
JDEM is a wide-field optical telescope in space; Hubble's not going to last forever, and the JWST's inability to see wavelengths shorter than yellow means it will have some trouble producing Hubble-class spectacular images, so it may well be that the classic NASA images from the 2010s will have to come from JDEM.
Slashdot Mirror
The RTG references for this are, I think, mostly traceable back to
http://www.space.com/news/nasa_plutonium_020724.html
which indicates that 'for reasons of national security' one RTG-worth of plutonium-238 had been reclaimed from NASA about five years ago.
There are various national-security applications for plutonium-238 - it's perfect stuff for powering, for example, bits of equipment to sit in a cave in Afghanistan or next to an undersea cable off Taiwan quietly recording all that passes to be collected later; it lets you build satellites without shiny solar panels. Lincoln Experimental Satellites 8 and 9 used RTGs; http://www.osti.gov/energycitations/product.biblio.jsp?osti_id=5562667 is an abstract claiming there are ten American satellites in Earth orbit with RTGs on board, though I rather doubt it will list names and purposes.
Except for Blue Gene (and that's a big exception, since the 2^17-CPU Blue Gene at LLNL pumps out about a billion processor hours a year on its own, and has had another 80,000 CPUs added recently) these CPUs are either 2.4GHz Opterons or 3GHz Core2s; those were and are the sweet spots for building big HPCs.
The list of projects is at
http://www.science.doe.gov/ascr/incite/2008INCITEFactsheets.pdf
They seem mostly to be about fire, in power stations, in supernovae or in fission reactors.
Some nice examples: 27Mhours on lattice quantum chromodynamics, 21Mhours to simulating thermonuclear burning in type-1B supernovas, 18Mhours to figuring out how biofuels burn, 17.5Mhours to determine from first quantum principles how the nickel-56 nucleus holds together, 16Mhours to simulating thermonuclear burning in type-2 supernovas, 12Mhours to attempting to design a carbamate hydrolase enzyme de-novo, 10Mhours to simulating lead-telluride / silver-antinomy-telluride thermoelectric materials, 4.5Mhours to optimise the design of the next-generation linear collider, 5M hours to figuring out why enormous temperature gradients persist in liquid-sodium-cooled fast-breeder reactors and a further 14Mhours to liquid-sodium reactor design in general, 4M hours to figuring out exactly how multiple burners in large power-station combustion chambers light one another, 3.5Mhours to trying to understand why it's so hard to hydrolyse cellulose, 3.5Mhours to understanding how flame fronts move in the complicated gas mixtures obtained from coal gasification, half a million hours for oceanic circulation, three quarters of a million hours for flow of dense suspensions, ten million hours on catalyst design.
And, for some reason, a million hours on porting Plan-9 to the Blue Gene system. I presume this allows you to crash and reboot the entire 200kcpu system enough to identify ten bugs. Also eight million CPU-hours to developing better HPC libraries.
I would be interested to know the amount of idle time there is on these supercomputers; a friend of mine from mersenneforum.org got a week on several hundred Opterons in France over Christmas, which was enough to do most of the work required to factorise a few numbers of fairly unreasonable size - sadly, there's a second step in the factorisation which requires an SMP machine, and the biggest SMP machine I have is an Intel Q6600, so completing the factorisation is taking three weeks on a single desktop in my back bedroom.
Yes, nuclear fusion produces helium.
The fusion of 1kg of deuterium produces near enough 1kg of helium, and, umm, 2.7MeV per fusion * 6*10^23 atoms per mole * 500 moles of D atoms per kilogram / 2 deuteriums per fusion * 1.6e-19 joules per eV = 64.8 terajoules of energy.
So, a one-gigawatt fusion power plant would produce a kilo of helium every eighteen days; if the current electricity use of France were provided entirely by fusion plants, you'd get thirty tons a year. The large hadron collider uses 120 tons of helium, but efficiently; present planetary helium use is about seventy-five tons a day.
For comparison, the US produces from natural gas about 76 million cubic metres of helium a year; a cubic metre of helium weighs 1000/22.4*4 grams, so 76 million cubic metres weigh about fifteen thousand tons.
The USGS compiles a large quantity of useful information about mineral production and consumption, including helium:
http://minerals.usgs.gov/minerals/pubs/commodity/helium/
You can buy helium from the US government at $2.037 per cubic metre, whilst the commercial price is nearer $3 per cubic metre; adjusting that would seem to make some kind of sense, since the US has 600 million cubic metres of the stuff in Amarillo.
There are plants at Skikda in Algeria and somewhere in Qatar which aim to extract 25 million cubic metres from natural gas a year, but there have been some issues in getting them to work; both Algeria and Qatar's natural gas reserves contain about as much helium as the US total reserves do.
It is impossible to substitute for helium for cryogenics; nothing else stays liquid at that low a temperature, and the ultra-refrigerators that get to liquid helium temperatures use helium as working fluid.
I did my PhD at Nottingham University, which uses a fair amount of liquid helium; the arrangement there is that it's delivered to the MRI building at the top of the hill, and the boil-off passes through a liquifier and is used by the theoretical physicists at the bottom of the hill. I don't know what the theoretical physicists do with their boil-off; there are obvious practical problems with running piping from lots of separate labs to a central liquifier, and liquifiers are bulky and vibrating enough that you don't want to have them in the same lab as your delicate semiconductor-physics experiment.
Yes, I had a look at http://en.wikipedia.org/wiki/Wikipedia:Requests_for_comment/Cla68 and there is clearly a lot of weird toxicity going on there.
...
I got the impression that this was a matter of an author googling around a topic and not realising that the convincing page early up in the hits was written by a lunatic, then a number of editors deeply fed up of dealing with the consequence of this lunatic jumping on him. I could easily see myself doing the same.
There's a lot of dire technical-management-substitute involved, at least in part because there are no very useful side-channels: it's not clear that there's a way of saying 'umm, you realise that you're linking to a site written by a lunatic, maybe reconsider?'. Since editors have to spend most of their time dealing with people motivated by malice, it's forgiveable that they assume of malicious motivation
I don't see it as a critical failing of an encyclopaedia that its governance structures are such as to bias its pool of contributors towards the stoic.
This is just another part of the slow-motion meltdown of overstock.com, and the 'naked short-selling' financial conspiracy theory. There seem to be a lot of financial conspiracy theories around at the moment, presumably since there is some degree of actual financial chaos in the background, and the things financiers have demonstrably got away with are crazy enough that it's difficult to reject conspiracy theories on the mere grounds of strangeness in appearance.
Disclaimer: yes, I write stuff on wikipedia, my handle is fivemack. Mostly I write about chemistry; it's pretty clear that wikipedia is the most comprehensive and reliable site for chemistry on the Web, since chemistry is advanced stamp-collecting and wikipedia is a superb medium for presenting stamps in multiple series. The science side of wikipedia is a wonderful resource, and doesn't seem too prone to the kind of lunacy that afflicts other parts of the encyclopedia; people have less heated feelings about the melting-point of tellurium or the carcinogenicity of tetramethylhydrazine than they do about whether Mount Ararat is a Turkish or an Armenian mountain.
This sounds like a nice engineering exercise - very small motors with adequate torque, very small angle encoders, elaborate chassis to contain it all with the right number of degrees of freedom. The interesting part is always the control software, and I can't work out from the article whether the robot is running software that can play the violin ab-initio, or playing back a sequence of moves figured out by infinite grad-student labour.
I'd probably guess the latter - 'can it play any other piece' and 'can it play _this_ violin' (offering one of a slightly different size - with the popularity of the Suzuki method, children's violins are made in a variety of sizes) would be the killer questions.
I've heard of robots (WABOT-2 back in 1985) which can play piano from a score - OCR on musical notation turns out not to be especially difficult - but the extra mechanical linkages in a piano mean that you can press the keys in almost any way and get something note-like.
I'm rather more impressed that Toyota have managed (http://www.toyota.co.jp/en/special/robot/) to get robotic lips to play the trumpet - that's a nasty problem in engineering active soft materials.
Yes, the Skynet project includes a huge amount of ground infrastructure - terminals in jeeps, terminals on ships, terminals in shipping containers, terminals anywhere that an MoD Procurement Executive employee could imagine a member of the armed forces needing a terminal. So that was an unfair comparison to use; sorry.
I suspect there isn't very much deep-black material on Skynet, it was constructed via a complicated scheme of industrial contracts as a showpiece of private-finance-initiative procurement, and I can't see the kind of people who have weird packages to fly wanting the exposure risk of having extra clauses in the contract detailing the packages. I've worked a small amount on projects near Skynet, but on the documentation and assurance side rather than anywhere near either money or bent metal - yes, in a past life I was a civil servant.
The DVB satellites are off-the-rack ones (in as much as satellites ever are) which are supported by infrastructure that's already there, so I'm a bit more confident about that $400 million figure.
Actually, http://www.solarstorms.org/Sinsurance.html has some interesting numbers - Intelsat 'have declined to purchase insurance for satellites costing less than $150M', suggesting that some of their satellites cost more than that.
Envisat-1 (according to http://www.hq.nasa.gov/webaccess/CommSpaceTrans/SpaceCommTransSec34/CommSpacTransSec34.html) has 70% of its (http://ec.europa.eu/research/press/2002/pr0103en.html) about 2.3 billion Euro cost for satellite and launch services, though that is a large satellite bristling with novel detectors.
'Comparatively speaking, the price of a satellite is usually something like 95% launch costs, 5% satellite design and build costs' is not, in fact, anything close to true.
Launch costs are about five thousand dollars per kilogram to low-Earth orbit, about ten thousand dollars per kilogram to geostationary; maybe a bit cheaper if you're launching something fairly small on a Soyuz, maybe a bit more expensive if you're launching something enormous on a Delta 4 Heavy.
The Skynet 5 program is building and launching three large comsats for the British ministry of defence for $7.2 billion, of which about a thirtieth is the cost of the three half-Ariane-5 launches. For straightforward digital-broadcasting satellites, you're talking $400 million of which a quarter is launch cost.
In case you haven't noticed, Russia doesn't get very near the equator; they built the original facility in Baikonur because that was as far south as you could get in the Soviet Union and have a reasonable region of Soviet Union over which to drop discarded rocket stages.
The southernmost points of Russia are in the Caucasus, but that's a decidedly unstable area of the world, and rocket stages dropped off by things heading east would drop on Kazakhstan, which the Kazakhs obviously don't want. If you rule out the Caucasus, the next-southernmost points are at the North Korean border in the far east; there is a constant Russian worry that the Chinese might want to expand into Siberia if it's left empty, and so they'd like to build facilities there, especially the sort of facilities which set up clusters of skilled people who'd bring non-resource-dependent income to Amur.
The proposed site is Svobodny, which is just over the Amur from China, and not too far from the Komsomolsk-na-Amur rocket factory.
http://maps.google.co.uk/maps?f=q&hl=en&geocode=&time=&date=&ttype=&sll=54.162434,-3.647461&sspn=8.188315,20.566406&ie=UTF8&ll=51.410771,128.19191&spn=0.272388,0.6427&t=k&z=11&iwloc=addr&om=1
Obviously, an equatorial site would be better, and indeed there's a Soyuz pad being built at Sinnamary, in French Guyana, five degrees from the equator and about twenty miles from the Arianespace facility at Kourou. First launch from there will be late 2008, but it's only Soyuz so not particularly heavy lift, and I suspect the Russians might be less keen than EU nations at having their military satellites launched from French soil.
It means that you will finish the maximum-likelihood refinement of a protein model against X-ray data in forty minutes rather than ninety, so can check the mode in which ligands bind to your protein kinase about twice as fast as you could on the slower computer. Which, literally, helps some of our customers find treatments for cancer more quickly.
(yes, of course the software is multi-threaded, the core parts are Fourier transforms and exponential maps on large 3D arrays, and those parallelise nicely).
I expect Intel to ignore it entirely. The SSE5 proposal rewrites quite a chunk of the instruction set and would require a complete redesign of the processor pipeline, and Intel's competitive position at the moment is such that there's no need for it to do the work; they will argue that, if you want that kind of vector processing, you should get one of their Larrabee vector processors.
There is no particular guarantee that AMD will survive long enough actually to produce any processors using this planned SSE5 instruction set; it's for the chip that comes after Barcelona, and Barcelona is a year late and much slower than anyone had planned.
16GB of memory is possible on any dual-Xeon motherboard using eight 2GB FBDIMMs, and on most dual-Opteron motherboards using eight 2GB DDR modules, though some of the older, cheaper dual-Opteron boards only attached memory to one CPU. Some systems (the Sun X4150 server and X6250 blade, Supermicro's X7DWN+ board) have sixteen memory slots and support 4GB FBDIMMs, though those are horribly expensive - it's much cheaper to go from 16GB to 32GB by replacing the motherboard and processor and buying eight more 2GB FBDIMMs than it is to replace eight 2GB FBDIMMs with 4GB FBDIMMs.
It looks as if
http://ieeexplore.ieee.org/xpl/freeabs_all.jsp?tp=&arnumber=4348615&isnumber=4348298
is something like the work being reported on; 'A 1 Gsymbol/s, 64 QAM coherent signal was successfully transmitted over 150 km using heterodyne detection with a frequency-stabilized fiber laser and an optical phase-locked-loop technique. The spectral efficiency reached as high as 3 bit/s/Hz.'
Masato YOSHIDA's list of papers at
http://db.tohoku.ac.jp/whois/Tunv_Title_All.php?&user_num=LTU0OA==&sel1=1&sel2=1&sel3=1&sel4=2&page=1&lang=E
looks very plausible in the context of this work; 'coherent optical transmission' is I think the relevant buzz-word. Going from 1Gsymbol/s to 10Tsymbol/s is clearly a lot more work, but being able to do optical QAM at all is pretty spectacular.
It'll take only about a year or so; Samsung 8Gbit Flash chips are $9 on the spot market at the moment.
Though the interface for connecting eighty of them onto a single SATA channel wouldn't be completely straightforward just from the point of view of I/O pins; each chip presents eight data pins and nine control pins, so you need a total of about 1500 pins on the disc-controller ICs.
For the early market you'd probably use FPGAs, you'd need six XC3S400A in one of the larger packages to get enough I/Os, which would be another couple of hundred dollars; the master one can speak SATA directly and do a bit of buffering, and using this baroque configuration means you get the full data rate from each chip so reads are very fast. Say $1000 for 80G SATA-attached - but that's making mass-manufacturing assumptions, which are clearly false if what you're manufacturing are 80G disc drives for $1000.
Five hundred dollars for 80G SATA-attached sounds a reasonable sort of pricing once Samsung have got one more generation of Flash factory online, and once the devices are more plausibly consumer items.
But most people don't need a video card nowadays, it's integrated in the north-bridge chip on the motherboard, and the difference between a board with video on the motherboard and without video is about twelve pounds, for video a lot better than the G400MAX can offer. The way to get things really cheap is to lose the PCB and interconnects entirely, and put them on a chip that you'd have had to manufacture anyway.
Given that most proteins contain tryptophan, and tryptophan fluoresces under UV, and UV lasers are not that hard to come by, wouldn't it be easier to shine a UV laser at the crystallisation plate and detect by subtraction where the glowy bit is?
Or, as a lot of molbio automation companies are offering, actually shine an X-ray beam through the putative crystal onto a detector and see if it diffracts.
Fully automated high-throughput crystal growing strikes me as a bit of a boondoggle; the sophisticated robots required for the last steps of automation are an order of magnitude more expensive than having three shifts of trained Indian or Chinese workers moving plates around and looking through microscopes.
Generally these absolute-top-of-the-range chips don't go into supercomputers; for supercomputing, you do the price-performance calculation and find that more, slower chips suffice.
Though there are some applications where you need very large memories on the nodes, and if you're buying 32GB of RAM the difference between two 3.3GHz quad-core processors and two 2.4GHz quad-core processors is lost in the noise.
Tom's Hardware just did a series of power-consumption tests on various overclocks of a Q9650, the first available 45nm processor.
At 3GHz, it uses 8.79W when doing nothing, and 73W when running all four cores flat-out
At 4GHz, it uses 16.83W to do nothing, and 135W with all four cores flat-out; on the other hand this required a voltage increase to 1.44V from the 1.25V that sufficed up to 3.33GHz.
Fitting curves suggests that you would be using something like 350W for four cores at 5GHz, which is quite impressive.
But a 5GHz Skulltrail would be a chip hand-picked by Intel to run at lower voltages in general, and would be running cryo-cooled which I think also allows a lower voltage to be used; probably 200W would be a better estimate for the chip power consumption, though the cooler will be using a comparable amount of power. This is, indeed, moderately crazy.
http://www.nec.de/hpc/hardware/sx-series/index.html
There are four PDFs there; the brochure is a four-colour glossy, but there is some real information. Sadly, the interesting-looking white papers are for the SX6, two generations earlier.
SX9 summary: 65nm technology, 3.2GHz clock speed, eight vector elements handled per cycle with two multiply and two add units, which is where the 102.4Gflop/CPU figure comes from. 16 CPUs in a box about the size of a standard 42U rack.
Totally absurdly fast (ten 64-bit words per cycle per CPU) access to a large (options are 512GB or 1TB) shared main memory; absurdly fast (128GB/second) inter-node bandwidth.
I hadn't realized ESA was thinking of supplying the mirror for SPICA; I did an arxiv search for SPICA, and aside from the occasional confounding reference to Alpha Virginis, there was a paper on surface quality of carbon fibre / silicon carbide mirrors which made me assume that JAXA was planning to build one such.
[also a very nifty paper on coronographs using binary masks very close to the focal plane, which in principle masked out the star enough that you could almost detect Jupiters in reasonably distant orbits around reasonably close stars]
The mirror paper surprised me slightly, since they were testing the mirrors at 95 kelvin and assuming that the results could be extrapolated to liquid helium temperatures, which I'd have thought wasn't guaranteed.
You are, I suspect, the friend I was thinking of - Dave of Cardiff with whom I talked about infra-red astronomy at Recombination.
It depends what you mean by weight to orbit.
The Shuttle stack - the solid rocket boosters and the main engines - gets just under thirty tons of payload *and 68 tons of shuttle* into orbit, which is pretty close to the weight of the whole Apollo stack. In comparison, a big comsat might be seven tons, Envisat is eight tons, Hubble is eleven tons.
This is why the Shuttle-C project, which would have used the rockets from the Shuttle to launch something much more like a standard payload, was so appealing; you could launch most of a Space Station, or enough fuel for a manned mission to Mars, in a single shot.
Space missions aren't particularly resource-intensive; after all, the probes rarely weigh more than a few tonnes. They're expensive because they use a lot of staff time and rocket scientists aren't cheap, but about the only resource that they use which is in any sense rare is xenon for ion drives. Bits of space probe are quite often machined from solid ingots of relatively costly materials like inconel, but generally you can recycle the chips.
An Ariane 5 burns 25 tons of liquid hydrogen in 130 tons of liquid oxygen, and is assisted by about 480 tons of ammonium perchlorate and aluminium powder in the solid rocket boosters; the materials costs are trivial in comparison to the cost of the engineers who designed and assembled the machine. Liquid oxygen is significantly cheaper than milk (say 10 cents per kilogram), and 130 tons is much less than the daily consumption by even a modest steelworks; liquid hydrogen is cheaper than beer at about $4 per kilo.
It's interesting to see ESA proposing Galileo- or Cassini-type missions, given that NASA has rather missed the boat on those - they focussed on JIMO, which would have been a spectacular truly next-generation mission, but didn't have the political backing. Either mission would be a twenty-year commitment and cost several billion Euros, but the ESA is confident at the moment and the various 'Europe, centre of innovation' pushes by the EU will likely be able to find several billion Euros.
I wouldn't be surprised if JAXA flew Marco Polo whether ESA is interested or not - Hayabusa was a beautiful mission plagued by bugs, and Marco Polo would be a re-run with the bugs fixed.
DUNE/SPACE are close to duplicating a NASA effort called the Joint Dark Energy Mission; PLATO is very specialised and comes close to duplicating NASA's Kepler. I would bet on Xeus since XMM/Newton has been a great success, and ESA's reputation in X-ray astronomy is particularly good. Spica, again, I could see the Japanese flying without ESA input, though I suppose I should push for that since a friend of mine's working on the transition-edge detectors planned for it.
JDEM is a wide-field optical telescope in space; Hubble's not going to last forever, and the JWST's inability to see wavelengths shorter than yellow means it will have some trouble producing Hubble-class spectacular images, so it may well be that the classic NASA images from the 2010s will have to come from JDEM.