Anyone else feeling reminded of Bullfrog's classic PC game Syndicate? Can't wait to see corporations (e.g. Academi/Blackwater) buying these to supervise their missions.Oh, and please let them send out cyborgs with miniguns and mind control devices.
I do completely agree with the parent. We've recently been to Audi to discuss a research project and they gave us a nice tour, sharing some insights into their safety systems. Current Audis (and thus also Volkswagen, Seat, Scoda etc. which all belong to the VW group) have two classes of safety systems: pre and post crash. Pre crash systems analyze the situation and might act preemtively. The more interesting ones will break before you hit an obstacle or even avoid it by steering to the side. Technologically these systems are mostly mature. The sensors aboard can track >30 objects with a latency of just a few ms form object movement to servo action. The reason why they are not found in the current fleet is simple: insurance. What do we do in the 0.01% of the cases in which the system errs? These legal problems had to be resolved first.
Contrary to their name, post crash systems may kick in before the crash, but only if it's unavoidable according to the laws of physics. These systems will try to mitigate adverse effects, e.g. your bones being crumbled. For this the computers have a physical model of your body (and the car) and can detect through sensors in the seat if a heavy or light adult, or a kid in a booster seat is sitting in the car. It can then e.g. dose the strength of the belt pretensioner. One peculiar aspect of this is that the system knows whether you will break your neck during the crash. Scary!
This is the state of the art, no SciFi. Your car already knows much more about its surroundings that you might think it does. So relax people, and welcome your new robotic overlords... ehm, drivers. They'll be better drivers than you. Or me.
The MegaFLOpD metric you propose is interesting, but also limited. A $700 dual-GPU gaming rig will always be more cost efficient than an HPC system with similar performance, and cost efficiency gets worse the larger the machine gets (the K computer was $1.2 Billion IIRC). And yet these machines are sometimes the only ones that can run such simulations.
In essence, I don't think there is a single sufficient metric. PetaFLOPS are one thing, FLOPS/$ are the other. Both have to be kept in mind today already. For instance ORNL (who are currently upgrading Jaguar to Titan) are especially proud of achieving a PUE of 1.26 (meaning that they lose only a fifth of the total facility power on cooling, AC/DC converters etc. and 4/5 go straight to the CPUs/GPUs. Efficient cooling means lower cost of ownership means more FLOPS per $.
BTW: are you a materials scientist or physicist? Just curious because its so rare to talk to someone who knows about solidification processes.:-)
Here is one example our chair is working on: simulation of dendritic growth. Ever heard of that? Doesn't sound particularly relevant to your everyday life? Well, it is. Material scientist are interested in understanding how crystalline structures form in cooling metal alloys, The crytal structure is ke to building stronger, lighter metal work pieces. Ultimately a solid understanding of this will lead to e.g. higher fuel efficiency in jet planes and lighter cars.
Currently material scientists are building computational models of these processes. To check whether a new models works out we need to simulate it. This takes Terabytes of RAM and Petabytes of disk space. That's what such a machine is good for.
Here is a list of other flagship applications. Most of them are simulation codes that replace experiments which would either be too costly, happen under too extreme physical stress/forces, or simply could not be carried out at all in practice because of scale.
nuclear weapons stewardship (if the nuke sitting in that cupboard for 20 years is ignited tomorrow, will it work, despite the nuclear decay?)
fuel combustion (which compounds will burn in which mixture, pressure, conditions with the highest efficiency?)
weather/climate simulation (will it rain on sunday? and if not, will the Netherlands be flooded by 2100?)
It's not really about having the fastest machine (because then it would be stupid to build multiple 20 PFLOPS machines on US soil), but about having enought compute power to maintain leadership in a number of key sciences (e.g. simulation of fuel compustion, nuclear weapons, drugs development). For many disciplines these supercomputers are the only way to further the state of the art.
While that is basically true, I must still add that those offices are occupied by computer scientists. And those are not really known for partying hard, aren't they? And yes, I'm a CS PhD student, too.:-/
Yes, people have tested passive cooling ("wind tunnel") in server rooms. Integration density nowadays is way too high. In our (modest) server room CPUs and GPUs will quickly jump from die temperatures of 70C to 90C, which essentially means that the HW will be fried.:-/
The problem with simply "dumping the waste heat" somewhere is that you need to find a place where to dump it. As the story indicated, this is not so much a problem in winter, but in summer, when no one wants your heat (no offices b/c sweaty people won't be smart, no industry b/c the water is still too cool). The innovative aspect of SuperMUC is that they achieve free air cooling, even in the German summer when delta t is very very low. Another cooling fluid is not an option as refrigerators are not as power efficient. Its all about power efficiency. And TCO: LRZ saves 500k €/a by this hot water cooling compared to classic AC with "refrigerators".
AMD needs to get on the ARM bandwagon. I want an APU with an integrated ARM core that works as a service processor and low power auxiliary CPU when the big CPU is powered off. Good enough for email and browsing and if the GPU has good power management the battery should last forever.
So you would have a mix of instruction sets (ARM vs. X86). It's very unlikely that we'll see such a scenario, not just because of the hardware, but because of the software: porting an OS to run on two different architectures simultaneously is something (AFAIK) never done before. That would be coding hell. Too bad, because I like the idea, too.
Global interdependencies in our economies tie us all together. The only technology required for large scale global trading are containerships and telephones. Internet, global spot trading, airmail, Bitcoin, they're all just sugar icing on top.
"Africa has some of the poorest soils anywhere on the earth". Such a generic statement about a whole continent which contains huge portions of tropical rainforest and grassland is just wrong.
ARM is a good point, yes. But they only sell IP cores (i.e. they license their designs), they don't sell physical chips, not to mention complete systems. Infinion, Bosch and Siemens all have IT hardware related branches, but they're more into automotive and automation. None of them sells CPUs that are applicable for HPC.
I assume a "+1 funny" as otherwise I'd have to assume that you're oblivious to the numerous scientific contributions for which Europeans have received recognitions like the Nobel Prize or the Fields Medal. You've got a point though: research around the globe is tightly coupled and so the funded projects resemble each other. You could add Japan to the mix. Their K computer isn't just a copy of some IBM BlueGene or such. And it's currently the fastest machine, at least until BlueGene/Q results are in.
Europe on the other hand doesn't have a serious computer hardware industry. The only chip manufacturers left (e.g. IBM, AMD, Nvidia, Fujitsu etc.) are all non-european. For a layman, this may make it kind of hard to imagine what Europe would spend its funding on, if they can't build the hardware themselves. Well, it turns out that software is a major part of exascale computing because at that scale effects (e.g. reliability of the hardware, scalability of IO) play a major role, but didn't hurt as much on the Petaflop machines. Now, when you turn your face to the software aspect, then you will see that a sizeable part of the papers published at the relevant conferences (e.g. http://sc11.supercomputing.org/ ) are European, and in many aspects they set the benchmark in terms of scalability and performance.
That said, it's hard to find a purely European or US project nowadays as many research institutions collaborate
BigDog may look like a dog, but LS3 looks like a horse. Imagine how great this could be: every soldier gets his own LS3 to ride on. With these they could go effortlessly anywhere, even when no roads are available. Just like in medieval times...
Slashdot Mirror
...her "golden river" turns red every four weeks!
Touché!
Not until there is an Emacs version available from the App Store.
...and the coverage in the news. Here is a nice story on how people react on the verdict.
Anyone else feeling reminded of Bullfrog's classic PC game Syndicate? Can't wait to see corporations (e.g. Academi/Blackwater) buying these to supervise their missions.Oh, and please let them send out cyborgs with miniguns and mind control devices.
I do completely agree with the parent. We've recently been to Audi to discuss a research project and they gave us a nice tour, sharing some insights into their safety systems. Current Audis (and thus also Volkswagen, Seat, Scoda etc. which all belong to the VW group) have two classes of safety systems: pre and post crash. Pre crash systems analyze the situation and might act preemtively. The more interesting ones will break before you hit an obstacle or even avoid it by steering to the side. Technologically these systems are mostly mature. The sensors aboard can track >30 objects with a latency of just a few ms form object movement to servo action. The reason why they are not found in the current fleet is simple: insurance. What do we do in the 0.01% of the cases in which the system errs? These legal problems had to be resolved first.
Contrary to their name, post crash systems may kick in before the crash, but only if it's unavoidable according to the laws of physics. These systems will try to mitigate adverse effects, e.g. your bones being crumbled. For this the computers have a physical model of your body (and the car) and can detect through sensors in the seat if a heavy or light adult, or a kid in a booster seat is sitting in the car. It can then e.g. dose the strength of the belt pretensioner. One peculiar aspect of this is that the system knows whether you will break your neck during the crash. Scary!
This is the state of the art, no SciFi. Your car already knows much more about its surroundings that you might think it does. So relax people, and welcome your new robotic overlords... ehm, drivers. They'll be better drivers than you. Or me.
The MegaFLOpD metric you propose is interesting, but also limited. A $700 dual-GPU gaming rig will always be more cost efficient than an HPC system with similar performance, and cost efficiency gets worse the larger the machine gets (the K computer was $1.2 Billion IIRC). And yet these machines are sometimes the only ones that can run such simulations.
In essence, I don't think there is a single sufficient metric. PetaFLOPS are one thing, FLOPS/$ are the other. Both have to be kept in mind today already. For instance ORNL (who are currently upgrading Jaguar to Titan) are especially proud of achieving a PUE of 1.26 (meaning that they lose only a fifth of the total facility power on cooling, AC/DC converters etc. and 4/5 go straight to the CPUs/GPUs. Efficient cooling means lower cost of ownership means more FLOPS per $.
BTW: are you a materials scientist or physicist? Just curious because its so rare to talk to someone who knows about solidification processes. :-)
Currently material scientists are building computational models of these processes. To check whether a new models works out we need to simulate it. This takes Terabytes of RAM and Petabytes of disk space. That's what such a machine is good for.
Here is a list of other flagship applications. Most of them are simulation codes that replace experiments which would either be too costly, happen under too extreme physical stress/forces, or simply could not be carried out at all in practice because of scale.
It's not really about having the fastest machine (because then it would be stupid to build multiple 20 PFLOPS machines on US soil), but about having enought compute power to maintain leadership in a number of key sciences (e.g. simulation of fuel compustion, nuclear weapons, drugs development). For many disciplines these supercomputers are the only way to further the state of the art.
While that is basically true, I must still add that those offices are occupied by computer scientists. And those are not really known for partying hard, aren't they? And yes, I'm a CS PhD student, too. :-/
The problem with simply "dumping the waste heat" somewhere is that you need to find a place where to dump it. As the story indicated, this is not so much a problem in winter, but in summer, when no one wants your heat (no offices b/c sweaty people won't be smart, no industry b/c the water is still too cool). The innovative aspect of SuperMUC is that they achieve free air cooling, even in the German summer when delta t is very very low. Another cooling fluid is not an option as refrigerators are not as power efficient. Its all about power efficiency. And TCO: LRZ saves 500k €/a by this hot water cooling compared to classic AC with "refrigerators".
Interesting. Thanks!
Got that already: Smartphone + Notebook. I switch to whatever device better suits my needs. ^^ Battery on the ARM device is indeed not an issue.
Which chip in the Wii is an ARM chip? AFAIK it's just a PowerPC CPU + ATI GPU.
AMD needs to get on the ARM bandwagon. I want an APU with an integrated ARM core that works as a service processor and low power auxiliary CPU when the big CPU is powered off. Good enough for email and browsing and if the GPU has good power management the battery should last forever.
So you would have a mix of instruction sets (ARM vs. X86). It's very unlikely that we'll see such a scenario, not just because of the hardware, but because of the software: porting an OS to run on two different architectures simultaneously is something (AFAIK) never done before. That would be coding hell. Too bad, because I like the idea, too.
Global interdependencies in our economies tie us all together. The only technology required for large scale global trading are containerships and telephones. Internet, global spot trading, airmail, Bitcoin, they're all just sugar icing on top.
Buy Firefox bonds!
Dear Sir, If I had any mod points left, I'd spend one on this post. DNA as a ubiquitous fingerprint is already bad enough. Kind regards
Easy: just put a cup to the exhaust of your hydrogen car, add a bit of Earl Grey and you'll be good.
Any cost can be counted in Bitcoins.
"Africa has some of the poorest soils anywhere on the earth". Such a generic statement about a whole continent which contains huge portions of tropical rainforest and grassland is just wrong.
...which is why it's easy to scale up. Thus the speedup isn't that impressive. Scalability on tightly coupled apps would be much more interesting.
ARM is a good point, yes. But they only sell IP cores (i.e. they license their designs), they don't sell physical chips, not to mention complete systems. Infinion, Bosch and Siemens all have IT hardware related branches, but they're more into automotive and automation. None of them sells CPUs that are applicable for HPC.
I assume a "+1 funny" as otherwise I'd have to assume that you're oblivious to the numerous scientific contributions for which Europeans have received recognitions like the Nobel Prize or the Fields Medal. You've got a point though: research around the globe is tightly coupled and so the funded projects resemble each other. You could add Japan to the mix. Their K computer isn't just a copy of some IBM BlueGene or such. And it's currently the fastest machine, at least until BlueGene/Q results are in.
Europe on the other hand doesn't have a serious computer hardware industry. The only chip manufacturers left (e.g. IBM, AMD, Nvidia, Fujitsu etc.) are all non-european. For a layman, this may make it kind of hard to imagine what Europe would spend its funding on, if they can't build the hardware themselves. Well, it turns out that software is a major part of exascale computing because at that scale effects (e.g. reliability of the hardware, scalability of IO) play a major role, but didn't hurt as much on the Petaflop machines. Now, when you turn your face to the software aspect, then you will see that a sizeable part of the papers published at the relevant conferences (e.g. http://sc11.supercomputing.org/ ) are European, and in many aspects they set the benchmark in terms of scalability and performance.
That said, it's hard to find a purely European or US project nowadays as many research institutions collaborate
BigDog may look like a dog, but LS3 looks like a horse. Imagine how great this could be: every soldier gets his own LS3 to ride on. With these they could go effortlessly anywhere, even when no roads are available. Just like in medieval times...