Slashdot Mirror
ORNL's Newest Petaflop Climate Computer To Come Online For NOAA
bricko writes with a description of NOAA's Gaea supercomputer, being assembled at the Oak Ridge National Laboratory. It's some big iron: 1.1 petaflops, based on 16-core Interlagos chips from AMD, and built by Cray. "The system, which is used for climate modeling and resource, also includes two separate Lustre parallel file systems 'that handle data sets that rank among the world's largest,' ORNL said. 'NOAA research partners access the system remotely through speedy wide area connections. Two 10-gigabit (billion bit) lambdas, or optical waves, pass data to NOAA's national research network through peering points at Atlanta and Chicago.'"
How much a Petaflop Climate Super Computers contribute to carbon footprint...
A feeling of having made the same mistake before: Deja Foobar
As an out of work Biologist, I have a great interest in the population size of the Guatemalan Diarrhea Toad.
5-day forecasts are quite accurate. Check for probabilities, and if they're low, check back in a few days. I've been rarely surprised by weather in the last few years - as in, was expecting a warm, sunny day and got instead cold rain all day.
Those who can, do. Those who can't, sue.
Spend this money on putting people back to work and improving the infrastructure of this country.
Yea, 'cause it's not like any people are involved with the design, engineering, manufacture, transport, operation, and maintenance of a supercomputer!
What a maroon...
An enigma, wrapped in a riddle, shrouded in bacon and cheese
5-day forecasts are quite accurate. Check for probabilities, and if they're low, check back in a few days. I've been rarely surprised by weather in the last few years - as in, was expecting a warm, sunny day and got instead cold rain all day.
Depends where you live. Right off the Pacific Ocean a prediction is only good for about a day, sometimes not even that. All depends how directly a cell is moving relative to you position. Finding temperature and direction of ocean currents has a bit to say on the matters and doesn't necessarily move in a 2 Dimensional plane.
Living in the midwest, ah, you could see it coming days away. Very predictable over large land area.
A feeling of having made the same mistake before: Deja Foobar
It's disheartening that most of the first posters are all trolls wondering why we would ever need this or are just trying to get cheap jabs in on a site for nerds. If you don't like the science behind it (climate sciences), or you don't like the technology behind it (computing systems), then why come here to comment?
Personally, I don't put much stock in the climate modeling capabilities of it just because that is not my area of study or interest. But having another large supercomputer with interconnects running at this speed is pretty cool.
I've worked at a company that had 8 of these 10Gb waves worth of bandwidth between Chicago and NY (and at an extremely low latency), now THAT was fun! On the other hand, the prisms and optics you need in order to separate out the lightwaves were hideously expensive :)
- Toast
On a related note, today I worked 28.8 million milliseconds.
none
Living in the midwest, ah, you could see it coming days away. Very predictable over large land area.
Not during the summer, at least not where I live. Winter, sure, 5 day prediction no problem. Summer? Storms can hit hours after they give a prediction for sunny skies. And visa-versa: I've seen 95% chance of rain with not a single drop falling (hell, sometimes barely even any clouds). Always the possibility of tornadoes too: they can barely predict a likelihood for those during the storm.
"None can love freedom heartily, but good men; the rest love not freedom, but license." --John Milton
Ever since the 1960's and big Control Data Corporation iron the latest/greatest supercomputer always seems headed towards weather forecasting/climate modeling. And we still can't accurately say where the next hurricane will make landfall in 3 days, or if I'm going to get rained on tomorrow.
"It's the height of ridiculousness to say for those 9 lines you get hundreds of millions."
[looking at picture linked] Where's the dining table? The napkins? A couple ASIMO waiters?
CS majors know the time/space tradeoff, but they never get taught the 3rd, crucial, tradeoff of the set: comprehension!
The main reason forecasts are poorer near the Pacific coast is that we have few observations available over the ocean compared to what we have over land. Since there is predominately westerly flow in the mid-latitudes, storm systems are poorly sampled when they arrive at the west coast, but we can observe them thoroughly before they reach the Midwest. A lot of research is currently being done to develop ways to use current and future satellite observations in forecast models, partly for the purpose of making up for the shortage of obs over the oceans.
Thunderstorms can be quite tricky to forecast since their development can be very sensitive to small details in things like the vertical temperature profile of the atmosphere. Often times during the summer in the Midwest, the difference between sunny skies (or at least just a bunch of puffy clouds) and severe thunderstorms can be only a degree or two in the temperature, either at the surface or in the low- to mid-levels of the atmosphere. Thus, if our observations of temperature have a small error, or if a small amount of error exists in a model forecast, it may be quite difficult to predict whether storms will occur or not. Similar sensitivities exist when it comes to moisture in the low-levels of the atmosphere.
Tornadoes are even trickier, since in their case we are still trying to figure out exactly what all to look for in a storm and the storm's environment that determines if it will produce a tornado or not. We know a lot more about that than we used to, but we are still a way from being able to say with much certainty whether any specific storm will produce a tornado, especially with more than a short lead-time (at least in the case of storms occurring in environments favorable for tornadoes -- there are plenty storms which we can pretty certainly say will *not* produce a tornado). Even if we figure out all the details of what causes tornadoes to occur and thus what to look for, predicting them will likely still, in many cases, be subject to the sort of sensitivity to small scale details mentioned above.
The mechanisms that drive large-scale precipitation, like often occurs during Midwest winters, are less sensitive to small details, which allows it to be predicted with more certainty much further in advance.
Having many different hurricane models that don't always agree with each other is actually quite useful -- the degree of agreement across models can often be used as a measure of uncertainty in the forecast. If all of the models show a storm plowing into New Orleans as a Cat 5 in three days, it is probably time to start evacuating. If one of the models says that, but the rest of them show it fizzling out into a tropical depression, chances are it won't warrant evacuations, and calling for them will most likely have a very large and unnecessary negative impact on the local economy.
If you think we should cut funding for hurricane forecast R&D, take a look at the right side of the millions section in XKCD's great money chart: http://xkcd.com/980/huge/
For those who don't want to hunt for that, we've spent $440 million on improving hurricane forecasts since 1989, or about $20 million per year on average. The estimated economic savings from limiting evacuations due to Hurricane Irene this year (one storm, during one year) is $700 million. Seems pretty cost-effective to me.
UH... CPUs used in a supercomputer have fuckall to do with the CPU that I would want to use in any standard PC, mobile device, or even a regular server for that matter. Cray made a deal with AMD in 2005 (back when they WERE the faster chip) and is basically stuck with AMD for the foreseeable future due to being tied to AMD's hypertransport. The interconnect is what matters, not the CPU.
You want the fastest supercomputer in the world? Guess what: it uses SPARC chips. Now I'm sure you hate Intel enough to convince yourself that you should go out and only use SPARC chips since they are 'teh fastest' , but believe me, you wouldn't like the results.
Considering benchmarks of freshly release Interlagos servers with 32 cores only beat Intel models from 2010 with 12 cores by 20% at best... while losing at other benchmarks (and cost over twice as much so STFU before you go on about how "cheap" the AMD solution is) you get a good idea as to why AMD's share of the server market is 5% when it was more than 25% just a few years ago.
AntiFA: An abbreviation for Anti First Amendment.
Oh how I weep for humanity that factual and useful statements are avoided and shunned by morons.
I count supercomputers as infrastructure. How many people worked on building the thing? Now they can start working on the next supercomputer.
So if 1/3 show a decline, and 2/3 show an increase, isn't there just a smidge of a possibility that the overall average is going up? (It's not a gurantee, obviously; if the increases are small and the declines are large the overall number would be down.)
Climate models aren't intended to explain the local variations; they're there for the big picture.
The cagey way you phrase this makes this a likely troll. The climate folks have looked at all the numbers and concluded that, overall, temperatures are increasing. Yes, some local temperatures are declining. You yourself admit it's even a minority of local temperatures. That you try to turn it into some sort of accusation is trolling.
-DwS
Of course it's going up, NASA has confirmed that with satellite information as well as several other sources all showing quite clearly that the temperature is rising.
Basing models on data that is at least 1/3 bogus is fucking stupid; NOAA puts a LOT of weight on the land-based temperature data in their models.
What I said is that maybe the should find out why a large (huge in terms of science) amount of their data is invalid before they go about throwing huge amounts of power at more models.
You misread intense anger at the stupidity and political motivation as trolling. I would like nothing more than accurate climate models but we'll never get them until people admit that the data we have is shit.
Of course it's going up, NASA has confirmed that with satellite information as well as several other sources all showing quite clearly that the temperature is rising.
News flash: the satellite and surface station temperature records closely agree.
Basing models on data that is at least 1/3 bogus is fucking stupid
News flash: "data that shows cooling" != "bogus data". Parts of the Earth do cool from time to time, you know (and are expected to, even with the enhanced greenhouse effect). The satellite data shows this as well.
NOAA puts a LOT of weight on the land-based temperature data in their models.
Climate models usually don't use temperature data at all (i.e., it's not an input to the model). They're run freely using only the forcing data (greenhouse gases, solar varations, aerosol loadings, etc.) and allowed to predict their own temperatures, without reference to any temperature observations.
When they are initialized with temperature data (in "data assimilation" mode), land gets exactly the weight it should: about 30% of the Earth's surface.
Temperature observations are used for testing the predictions of climate models, but the above remains true: land data gets exactly as much weight as its area average. And models are compared to a variety of temperature records (surface and satellite), not that it matters much, since (as noted above), they all agree pretty closely.
I would like nothing more than accurate climate models but we'll never get them until people admit that the data we have is shit.
As amply demonstrated above, you have no idea what you're talking about.
You're a moron, go and actually LOOK at the NOAA data. A cluster of stations within a 50 mile radius at about 2/3 warming and 1/3 cooling.
Stop being a retard, do some fucking research and piss off you gnat.
A cluster of stations within a 50 mile radius at about 2/3 warming and 1/3 cooling.
Again, so what? This does not imply there is anything wrong with the data. You are deeply confused about what should happen, meteorologically speaking, on microclimatic scales.
You should go and hit yourself with a hammer, I swear to god you're so dumb it'd make you smarter.