Power consumption and mean time between failure are the current major challenges. What good would a exa-scale supercomputer be, if you couldn not run a full-system job on it for even one hour?
Roadrunner consists of 6480 QS22 Blades. Using Cellminer each will yield approx. 56 MHash/s, or 363 GHash/s in total. Using the Bitcoin profitability calculator we can then estimate that one will gain ~27 BTC/day (ATM $2667) while paying $3360/day for power (assuming cheap $0.07/kWh). So yes: mining on Roadrunner would not be cost-effective.
Whenever there is a story on supercomputers on/., there will be a comment stating that there was no barrier whatsoever. But that's not quite true.
The truth is that the performance of supercomputers grows that fast because engineers continuously solve problems, which were deemed intractable before (e.g. power consumption, reliability, network performance). The research may not be groundbreaking in the sense of earth-shattering, but definitely in the sense of "wow, I didn't think one could do that!"
Let me quote TFA: "...performing even longer flights, one including a full circle. (Flying a 360 circle was the accepted standard for proving an aircraft was controlable in early aviation.)"
Good spotting! Looks like the umlauts got converted to the HTML encoding of Unicode automatically. I had totally forgotten to use the abominations of ß (for ß) and friends.
If maneuverability and high-G turns are still an advantage for fighter jets, then the imposed on those drones by having the pilots sitting on the other side of the globe would result in a tactical disadvantage. We've learned from game streaming (Gaikai, OnLive) that this lag can be significant boon (even with good network equipment) as the speed of light cannot be cheated.
At least Germany military experts doubt its efficiency in bad weather conditions (link is to a Google translated site as the original article is available in German only, sorry). I'd say their expertise weights more than the argument they're from Norway so they'll be alright.
Also, at 144k€ per unit this seems ridiculously expensive.
I just got my development kit from Leapmotion. Take a look at their 1m video. Pretty impressive, I would say. Also, the resolution of their sensor (1mm) seems to be much higher than that of what cameras can do today.
Did anyone else read the title and thought "Megaupload's downfall wasn't that bad, wasn't it?" Or did Kim's extensive physique just lose structural integrity?
Both, China and India are being hyped as the prime locations for outsorcing software engineering. But if you listen closely to the companies then you'll see that the first are already coming back to the US and Europe. And even if not: the people there that you'd have to communicate with all already speak English well. So congratulations, as an English native speaker you already have the best tool at hands to get around the world. But you might want to consider learning Spanish so that you can talk to the fastest growing minority in your own country.:-)
T-Mobile has the iPhone since ages in other countries. For instance here in Germany you could get any iPhone from T-Mobile IIRC. Don't know why that was different in the US. Probably because Apple tried to pay them through the nose. But the smartphone market in Germany is very different: very fey iPhones, many many Androids.
KDE got a lot of flak for the early 4.x versions, because they felt terrible. But what they did (replacing many internals, reworking the architecture) did yield us now a very flexible UI. Plasma (KDE's UI) is fully based on SVG and looks good on pretty much any screen, be it a notebook, workstation, or even tablets. And its not such a CPU/memory hog as the people generally claim.
Chaotic storage works because the barcode of each shelved item is scanned together with the barcode of its shelve, so that the computer can later on tell the humans where to find the stuff for a certain order.
Apparently there is no reason why this wouldn't work with robots. Apparently robots are still to expensive or not smart (in terms of physical skills) enough.I wonder when we'll see Amazon experimenting with robots.
Yeah, it still sounds like b/s to me. Here is why:
You could try reading the fine article. If you did so you would learn that he is talking about overclocking enthusiasts, who are now having fun overclocking ARM chips. He linked an article about Android running at 3.0 GHz (on an OMAP chip rated for about 1 GHz).
And why do those guys enjoy overclocking? Because they either try to achieve higher performance levels, or because they try to get insane frequencies, or both. An ARM core will never achieve any of those goals. Overclockers seldom boast themselves as haven o/c'ed their CPU by XY%, but rather to X.Y GHz.
Also, the quoted section says that "any headroom will have been screened out at the fab" implying that there won't be any overclocking potential in the new chips. I don't quite understand exactly how that will work, however. The reason overclocking works is that CPUs at high clock rates cost more, which means fewer are needed at the highest clock rates; yet the lower-cost slower CPUs are basically the same silicon, just not guaranteed at the higher speed. In other words there is a supply of chips capable of higher speeds, being sold more cheaply as a lower-speed part. How does improved screening change this situation? Unless Intel has some way of making the chip yields come out to exactly the number of fast chips they need, it seems like there will always be a chance to find a chip that could have been in a more expensive bin. It hardly seems likely that Intel will just shred chips that overperform their bins!
You got this exactly right and the blog post (can we please stop calling it an "article"?) was pretty wrong: there will always be a demand for these CPUs, and there will be supply, too. The only thing which is likely to change is that you won't buy CPU and MB separated from each other, but together. My $0.02.
...doesn't understand the first thing about supercomputers, or even HPC. Currently virtually every HPC application uses MPI. And MPI doesn't take well to failing nodes. The supercomputer as a whole might still work, but the job will inevitably crash and needs to be restarted. HPC apps are usually tightly coupled. That sets them apart from loosely coupled codes such as a giant website (e.g. Google and friends)
Fault tolerance is a huge problem in the community and we don't have the answers yet. Some say that fault tolerance within the MPI layer (e.g. here) will be sufficient. I personally very much doubt that. My bet is on higher-level frameworks, e.g. HPX, which can "abstract away" the location of a task from the node where its actually being executed.
Checkpoint/restart is actually a rather poor workaround as the ration of IO bandwidth to compute performance is shrinking with every new generation of supercomputers. Soon enough we'll spend more time writing checkpoints than doing actual computations.
I personally believe that we'll see some sort of redundancy on the node level in the mid-term future (i.e. the road to exascale), which will sadly require source code-level adaptation.
The Top500 reports actual performance as measured with LINPACK, hardware vendors report the theoretical performance of their chips, which in the case of GPUs is often quite a bit more than you'd be able to squeeze out with LINPACK.
For comparison: Tsubame 2.0 consists of 1400 nodes with approx. 4200 NVIDIA Tesla C2075, which should yield -- according to your estimate -- 2.1 PFLOPS (4200 * 0.5 TFLOPS), yet it is listed at 1.2 PFLOPS. So just add two years to your estimate and you should be fine...
You're laughing, but here in Germany the GEMA (fills in for the RIAA in Germany) is charging the educators at Kindergartens for the songs they sing with the kids.
Slashdot Mirror
Power consumption and mean time between failure are the current major challenges. What good would a exa-scale supercomputer be, if you couldn not run a full-system job on it for even one hour?
Phase transition from gaseous to liquid dissipates thermal energy. News at 18:00.
Roadrunner consists of 6480 QS22 Blades. Using Cellminer each will yield approx. 56 MHash/s, or 363 GHash/s in total. Using the Bitcoin profitability calculator we can then estimate that one will gain ~27 BTC/day (ATM $2667) while paying $3360/day for power (assuming cheap $0.07/kWh). So yes: mining on Roadrunner would not be cost-effective.
Whenever there is a story on supercomputers on /., there will be a comment stating that there was no barrier whatsoever. But that's not quite true.
The truth is that the performance of supercomputers grows that fast because engineers continuously solve problems, which were deemed intractable before (e.g. power consumption, reliability, network performance). The research may not be groundbreaking in the sense of earth-shattering, but definitely in the sense of "wow, I didn't think one could do that!"
Let me quote TFA: "...performing even longer flights, one including a full circle. (Flying a 360 circle was the accepted standard for proving an aircraft was controlable in early aviation.)"
Good spotting! Looks like the umlauts got converted to the HTML encoding of Unicode automatically. I had totally forgotten to use the abominations of ß (for ß) and friends.
If maneuverability and high-G turns are still an advantage for fighter jets, then the imposed on those drones by having the pilots sitting on the other side of the globe would result in a tactical disadvantage. We've learned from game streaming (Gaikai, OnLive) that this lag can be significant boon (even with good network equipment) as the speed of light cannot be cheated.
At least Germany military experts doubt its efficiency in bad weather conditions (link is to a Google translated site as the original article is available in German only, sorry). I'd say their expertise weights more than the argument they're from Norway so they'll be alright.
Also, at 144k€ per unit this seems ridiculously expensive.
I just got my development kit from Leapmotion. Take a look at their 1m video. Pretty impressive, I would say. Also, the resolution of their sensor (1mm) seems to be much higher than that of what cameras can do today.
Amen. I just came here to say that. Thank you!
Did anyone else read the title and thought "Megaupload's downfall wasn't that bad, wasn't it?" Or did Kim's extensive physique just lose structural integrity?
Perhaps Peter Northon is more for you.
Nice blend of geek culture and auto-biographical blog: blog.beetlebum.de. German only, though.
Both, China and India are being hyped as the prime locations for outsorcing software engineering. But if you listen closely to the companies then you'll see that the first are already coming back to the US and Europe. And even if not: the people there that you'd have to communicate with all already speak English well. So congratulations, as an English native speaker you already have the best tool at hands to get around the world. But you might want to consider learning Spanish so that you can talk to the fastest growing minority in your own country. :-)
But do those trolls eat toejam, too?
T-Mobile has the iPhone since ages in other countries. For instance here in Germany you could get any iPhone from T-Mobile IIRC. Don't know why that was different in the US. Probably because Apple tried to pay them through the nose. But the smartphone market in Germany is very different: very fey iPhones, many many Androids.
KDE got a lot of flak for the early 4.x versions, because they felt terrible. But what they did (replacing many internals, reworking the architecture) did yield us now a very flexible UI. Plasma (KDE's UI) is fully based on SVG and looks good on pretty much any screen, be it a notebook, workstation, or even tablets. And its not such a CPU/memory hog as the people generally claim.
Chaotic storage works because the barcode of each shelved item is scanned together with the barcode of its shelve, so that the computer can later on tell the humans where to find the stuff for a certain order.
Apparently there is no reason why this wouldn't work with robots. Apparently robots are still to expensive or not smart (in terms of physical skills) enough.I wonder when we'll see Amazon experimenting with robots.
You could try reading the fine article. If you did so you would learn that he is talking about overclocking enthusiasts, who are now having fun overclocking ARM chips. He linked an article about Android running at 3.0 GHz (on an OMAP chip rated for about 1 GHz).
And why do those guys enjoy overclocking? Because they either try to achieve higher performance levels, or because they try to get insane frequencies, or both. An ARM core will never achieve any of those goals. Overclockers seldom boast themselves as haven o/c'ed their CPU by XY%, but rather to X.Y GHz.
Also, the quoted section says that "any headroom will have been screened out at the fab" implying that there won't be any overclocking potential in the new chips. I don't quite understand exactly how that will work, however. The reason overclocking works is that CPUs at high clock rates cost more, which means fewer are needed at the highest clock rates; yet the lower-cost slower CPUs are basically the same silicon, just not guaranteed at the higher speed. In other words there is a supply of chips capable of higher speeds, being sold more cheaply as a lower-speed part. How does improved screening change this situation? Unless Intel has some way of making the chip yields come out to exactly the number of fast chips they need, it seems like there will always be a chance to find a chip that could have been in a more expensive bin. It hardly seems likely that Intel will just shred chips that overperform their bins!
You got this exactly right and the blog post (can we please stop calling it an "article"?) was pretty wrong: there will always be a demand for these CPUs, and there will be supply, too. The only thing which is likely to change is that you won't buy CPU and MB separated from each other, but together. My $0.02.
why would any "enthusiast" go for an ARM CPU with about one tenth of the power a current Intel CPU has? I call this story b/s.
...doesn't understand the first thing about supercomputers, or even HPC. Currently virtually every HPC application uses MPI. And MPI doesn't take well to failing nodes. The supercomputer as a whole might still work, but the job will inevitably crash and needs to be restarted. HPC apps are usually tightly coupled. That sets them apart from loosely coupled codes such as a giant website (e.g. Google and friends)
Fault tolerance is a huge problem in the community and we don't have the answers yet. Some say that fault tolerance within the MPI layer (e.g. here) will be sufficient. I personally very much doubt that. My bet is on higher-level frameworks, e.g. HPX, which can "abstract away" the location of a task from the node where its actually being executed.
Checkpoint/restart is actually a rather poor workaround as the ration of IO bandwidth to compute performance is shrinking with every new generation of supercomputers. Soon enough we'll spend more time writing checkpoints than doing actual computations.
I personally believe that we'll see some sort of redundancy on the node level in the mid-term future (i.e. the road to exascale), which will sadly require source code-level adaptation.
The Top500 reports actual performance as measured with LINPACK, hardware vendors report the theoretical performance of their chips, which in the case of GPUs is often quite a bit more than you'd be able to squeeze out with LINPACK.
For comparison: Tsubame 2.0 consists of 1400 nodes with approx. 4200 NVIDIA Tesla C2075, which should yield -- according to your estimate -- 2.1 PFLOPS (4200 * 0.5 TFLOPS), yet it is listed at 1.2 PFLOPS. So just add two years to your estimate and you should be fine...
You're laughing, but here in Germany the GEMA (fills in for the RIAA in Germany) is charging the educators at Kindergartens for the songs they sing with the kids.
Citation needed.