To the best of my understanding the improvement here is in lowering the thermal resistance right at the point of contact between the IC and the cooling system, which doesn't change the fact that the waste heat still has to go somewhere; but does allow heat to be removed from that very small contact area faster, which increases the safe maximum wattage for a given die size.
It requires good engineering to do robustly, quietly, in a relatively compact unit, etc; but dumping large amounts of waste heat once you've gotten it into the cooling system is the relatively easy part. We'll always wish we could get away with something lighter or smaller or quieter; but we can build you a heat exchanger large enough to throw at pretty much any thermal load(chips are very much on the low end: engine cooling can easily bring you into the multiple kilowatts and power plant heat exchangers into the multiple megawatts); but what we cannot so easily do is move heat faster from a small, fixed, dissipation area. If your IC is only 1 square cm, you have to be able to extract all the waste heat generated by the power level you are operating it at through that single square cm, with a low enough delta T to keep the IC from cooking itself.
The relatively simple method is just a solid heat spreader; but if you want to hit power densities higher than the conductivity of copper will allow, you have limited options(diamond would be nice, if you can get in the size you need). Direct liquid cooling of the die is tricky, even if you have access to cryogens, because the Leidenfrost effect tends to produce an insulating layer of vapor right where you don't want it, surrounding the item to be cooled, which ruins the efficiency of heat transfer. My understanding is that this microfluidic apparatus is aimed at solving this problem; by forcing the liquid coolant onto the chip in a way that disrupts the vapor layer in a way that naive 'just pour it on/pump it across' approaches don't.
It might prove to be the case, especially if somebody comes up with a clever way of producing the stuff in bulk, that these microfluidic interfaces will end up being used in larger cooling applications as well; but at present they are aimed at the problem of cooling small, relatively fixed size(since die area is expensive; and sometimes design constraints related to clocking and signal propagation time don't even allow you to spread the same circuit over a larger die, even if you are willing to pay for the extra area), heat sources that are bumping up against the limits of how fast we can get heat away from them. Whatever cooler ends up dumping the heat into the atmosphere or chilled water supply or whatever will be purely conventional; it's the point of contact between the IC and the conventional cooling system that is being improved.
Surely DARPA has enough nerds on hand to know that adding a 'small thermal exhaust port' to expensive military hardware is going to end in disaster, no?
If you could actually get people to use it both honestly and correctly(which is a big if, I know) complex numbers would actually make for a great version numbering system: You set the imaginary term to reflect the scope of the project, then bump the real term and (hopefully) reduce the imaginary term as subsequent releases implement the various features intended. When the version number has only a real component, you know you are finished. To account for real-world scenarios, version number updates can increase the imaginary term(either because the customer changed the specs again or because you made a horrifying discovery about how difficult something would actually turn out to be) and it is not required that the value of the real term increase with each release.
It does make it a little tricky to know which version is actually the most current; but if you plot each version number on a graph you get a very informative picture of the trajectory of the project. Nice, neat, line heading toward the real axis? Success! Bowel-churning random walk back and forth on the real axis as the imaginary component grows without apparent bound? Not so good...
It might well be something for the uplink/downlink between the ground station and the satellites, if that happens to be a bandwidth bottleneck; but it would presumably be of minimal use for the actual phone-to-satellite link: The earth's surface is loaded with optically opaque clutter, and users probably wouldn't stand for a cellphone that only receives calls when it has a clear view of the sky. For some of the other users of satellite phone/data(ships, say) the requirement that the modem-pod be mounted on top of the ship, and then local RF or wire links used to distribute the connection from there, wouldn't be nearly so difficult. Either way, it wouldn't have much effect on the latency that makes satellite voice and data so annoying; the speed of light gets you either way, and any orbit you can maintain without massive fuel burn to compensate for still being in the upper atmosphere is a long way away compared to a cell tower.
Using an optical link would also, presumably, require knowing where the recipient is: they must be using some elegant tricks(perhaps borrowed from Team Astronomy, which has been battling atmospheric distortion for years?) to keep the laser on target even through thermal distortions in the atmosphere; but they still need to know where 'on target' is. Again, much bigger challenge for a human with a handset, much smaller challenge for a satellite, aircraft, or largish ship.
We could do that, or we could let advances in killbot technology reduce the costs of suppressing the squalid underclass. I'm not confident that we'll choose wisely.
Isn't "Becoming old without a pension" less of a 'fear' and more of a 'guarantee'? With the exception of a few labor unions that have really dug in and not quite been extirpated yet, we are basically all playing the tables with our 401ks(if that). 'Pensions' are what the old people who accuse you of being an entitled, lazy, little shit have.
In other pedantry, isn't 'seeing your skills and tools become irrelevant' an apt description of what would happen if an AI started doing your job?
I imagine that the tricky bit is that, legally, the penalties increase drastically when you shift from 'stealing nothing' to 'stealing something', so there's an incentive to either not touch so much as a few quarters for the parking meter; or make off with enough loot to be worth the risk and effort required to circumvent whatever security measures are in place.
The only exception would be if security and audits were pitifully lax; and you knew that discrepancies of under 1% by weight are classified as 'eh, close enough, the coins probably dried out a bit in storage...' in which case the risk would be so low that the reward wouldn't have to amount to much. In the presence of greater risk, the fact that you don't have to steal all that much for it to qualify as a felony(and that the Federal Reserve, and Brinks, probably like to make examples in order to discourage copycat offenders) would make stealing modest amounts a pretty harrowing business.
As recent college students know, laundromats almost always accept quarters as a mechanism for buying machine time, so laundering the proceeds must have been particularly easy and convenient. This guy is obviously a cerebral master of crime.
How dare the dead hand of state interference meddle with an industry that has gone to all the trouble of developing a ceremonial 'self regulation' procedure?
Those are valid points as well(though the peripheral drivers issue is arguably a sub-issue of the 'only supported kernel is pretty creaky' problem). They do offer a case; but not much variety, so that may or may not be an issue; EOL-ing is definitely an unknown. The kernel issue struck me as the most important, in that if a device is mainlined, at least the hardware you already own tends to take ages to be EOLed(the kernel maintainers do eventually drop platforms and devices that appear to have nobody who cares; but it takes a while; while vendor-specific BSP-forks are often slightly elderly at birth; and either dangerous or wildly obnoxious, or both, well before the hardware is showing any signs of age.
USB is certainly capable; but arguably overqualified for IR work: The classic LIRC DIY can be driven off a serial port; and I assume that a few GPIO pins that aren't too crippled would work as well. Bit-banging the ~36KHz carrier frequencies common in IR applications would probably be a pain in the ass; but the sheer ubiquity of consumer IR remotes and receivers means that handy little 'most of the low-level details and the emitter/receiver all in an IR bandpass plastic package' modules are cheap, and the actual data rates on IR remote protocols tend to be trivial.
Soft power is a bit harder to just tack on; but would be really nice, especially for embedded applications. Yes, it isn't rocket surgery to have a microcontroller set up as a watchdog and capable of switching the +5v line on the rPi; but one would think that the same feature could be integrated into the board more cheaply and elegantly than doing it after the fact.
Do you know if the SATA on the Series 4 (looks like a Marvell Kirkwood?) supports port multipliers?
My need for absolute performance in NASes is fairly limited, so the fact that a multiplier bottlenecks all the drives on it to the speed of a single SATA link isn't that important; but my experience has been that support is...deeply uneven; and when you don't have some PCIe slots to shove more HBAs into, your options are limited if a multiplier turns out not to work.
All the UDOO models have LVDS exposed. I'm afraid that I can't tell you anything useful about how versatile/well-documented that aspect of the board is. I have one that has played reasonably nicely with some headless stuff; but I haven't tried driving a panel directly.
(Incidentally, where did you get your hands on a 1600x480 LCD? I didn't even know that those existed.)
I'm not in the position to have had need of one; but doesn't team rPi provide the 'compute module' version(DDR-2 SODIMM form factor, snap it in to the carrier board of your choice) for that purpose?
They don't seem to have an updated version with the guts of the newer model as yet, though I assume that that is coming.
In that vein, the current pain-point for the ODroid-2 is the fact that the AMLogic S905 SoC it is based on has no mainline kernel support; and the current vendor fork is of a version heading toward EOL uncomfortably quickly. There is supposed to be a mainlining effort that will fix this before the current option actually goes EOL; but that remains to be seen.
I must admit that (having come into linux back in the delightful days when Broadcom wireless meant screwing around with NDISwrapper) it's a bit of a shock; but the rPi actually has an atypically high plays-well-with-others factor. You can get them cheaper; and you can get them better; but until the 'every ARM SoC is its own dysfunctional port' issue gets ironed out, some very promising hardware can end up hobbled by neurotic and antique software.
It's always entertaining to see the latest episode of 'Bitcoin: rediscovering the reasons behind the various messy hacks that we dislike about more typical currencies...'
I suspect that these starry-eyed optimists wouldn't be entirely pleased with Trump's cost reduction strategies during his years in real estate, which have included trying to go cheap on the pesky human resources; but they are correct that he is basically the only option on the republican side who is even interested in pretending to care about the filthy peons who aren't good enough to realize their income in capital gains rather than 'wages'.
It's almost as though people can't be made to vote against their economic interests by promising to keep the scary gays away from school prayer forever. Crazy stuff.
What's really sneaky is that every lab demo of a biological supercomputer is actually a production-ready biological supercomputer showing off some crude little toy that will be 5 years out for the next few decades.
The trick is just to ignore whatever is on the benchtop and grab the guy talking about it.
The whole point of adblockers is to 'disrupt the relationship between advertisers and consumers'; because that 'relationship' is inherently somewhere between 'adversarial' and 'cold war'. We don't go to varying levels of hassle just for fun; we do so because we fucking hate you and your 'product'.
Pretty much all of the brute-force prevention strategies involve a compromise between vulnerability to denial of service attacks and vulnerability to brute force; it's really just a matter of picking your poison based on what works best for your environment.
The 'if you screw up, the data gets wiped' approach is a good fit when the device it protects is assumed to be a mere local cache(often one with a dangerously high risk of loss/theft) of something that is backed up elsewhere. Protecting the data isn't a serious concern; but keeping it from leaking is.
For on-site situations(and anywhere where any idiot on the network can bounce authentication attempts in your name off the server as often as they like) the progressive timeout/temporary lockout ones are usually a better bet.
Not that his opinion matters nearly as much as the others(he's still loaded; but he's more busy playing the Hunter S. Thompson of tech than being a tech leader these days); but I thought that McAfee's position wasn't so much 'pro unlock' as "Me and my hacker posse will hack the shit out of it!"; which is a vote in favor of getting the contents of the phone(not that anyone is really against that, if there were some non-problematic way to do it); but not obviously a vote in favor of the feds having the right to force Apple to make it so.
Before we start worrying about cash; why not actually crack down on all those 'respectable' financial institutions that are more than happy to launder money?
Sure, maybe filthy proles use stacks of grubby bills to do their crimes; but so long as HSBC and the like can launder billions in drug money and handle transactions for a variety of theoretically-embargoed states(and face zero criminal penalties either as a company or on the part of any of the people involved); only little people smuggle cash.
Without a flamethrower-cleanup of the financial sector, making noises about high-value notes is basically just bigger criminals whining about the fact that they have to deal with lesser criminals who might find using money mules cheaper than buying bespoke financial services for the multijurisdictional client who values discretion.
Yeah, that was my intention: only cancers that are clonally transmissible, not other types of pathogens that are transmissible in the usual manner of viruses(do any bacteria or eukaryotic parasites do that sort of thing?) and often provoke the development of cancer in their host.
So far, the one that horrifies me the most is the case of the poor bastard who was parasitized by one or more tapeworms; but ultimately died when the tapeworm developed cancer and its cancer spread beyond the tapeworm and aggressively into his tissues. That's multiple levels of "people who thought that nature was red in tooth and claw were optimists". For the sake of my sound sleep, I'm pretending that that could only have happened because he was immunocompromised; and certainly represents no risk whatsoever to anyone else; because otherwise I'm going to have to start sleeping with a flamethrower under my pillow.
Slashdot Mirror
To the best of my understanding the improvement here is in lowering the thermal resistance right at the point of contact between the IC and the cooling system, which doesn't change the fact that the waste heat still has to go somewhere; but does allow heat to be removed from that very small contact area faster, which increases the safe maximum wattage for a given die size.
It requires good engineering to do robustly, quietly, in a relatively compact unit, etc; but dumping large amounts of waste heat once you've gotten it into the cooling system is the relatively easy part. We'll always wish we could get away with something lighter or smaller or quieter; but we can build you a heat exchanger large enough to throw at pretty much any thermal load(chips are very much on the low end: engine cooling can easily bring you into the multiple kilowatts and power plant heat exchangers into the multiple megawatts); but what we cannot so easily do is move heat faster from a small, fixed, dissipation area. If your IC is only 1 square cm, you have to be able to extract all the waste heat generated by the power level you are operating it at through that single square cm, with a low enough delta T to keep the IC from cooking itself.
The relatively simple method is just a solid heat spreader; but if you want to hit power densities higher than the conductivity of copper will allow, you have limited options(diamond would be nice, if you can get in the size you need). Direct liquid cooling of the die is tricky, even if you have access to cryogens, because the Leidenfrost effect tends to produce an insulating layer of vapor right where you don't want it, surrounding the item to be cooled, which ruins the efficiency of heat transfer. My understanding is that this microfluidic apparatus is aimed at solving this problem; by forcing the liquid coolant onto the chip in a way that disrupts the vapor layer in a way that naive 'just pour it on/pump it across' approaches don't.
It might prove to be the case, especially if somebody comes up with a clever way of producing the stuff in bulk, that these microfluidic interfaces will end up being used in larger cooling applications as well; but at present they are aimed at the problem of cooling small, relatively fixed size(since die area is expensive; and sometimes design constraints related to clocking and signal propagation time don't even allow you to spread the same circuit over a larger die, even if you are willing to pay for the extra area), heat sources that are bumping up against the limits of how fast we can get heat away from them. Whatever cooler ends up dumping the heat into the atmosphere or chilled water supply or whatever will be purely conventional; it's the point of contact between the IC and the conventional cooling system that is being improved.
Surely DARPA has enough nerds on hand to know that adding a 'small thermal exhaust port' to expensive military hardware is going to end in disaster, no?
If you could actually get people to use it both honestly and correctly(which is a big if, I know) complex numbers would actually make for a great version numbering system: You set the imaginary term to reflect the scope of the project, then bump the real term and (hopefully) reduce the imaginary term as subsequent releases implement the various features intended. When the version number has only a real component, you know you are finished. To account for real-world scenarios, version number updates can increase the imaginary term(either because the customer changed the specs again or because you made a horrifying discovery about how difficult something would actually turn out to be) and it is not required that the value of the real term increase with each release.
It does make it a little tricky to know which version is actually the most current; but if you plot each version number on a graph you get a very informative picture of the trajectory of the project. Nice, neat, line heading toward the real axis? Success! Bowel-churning random walk back and forth on the real axis as the imaginary component grows without apparent bound? Not so good...
It might well be something for the uplink/downlink between the ground station and the satellites, if that happens to be a bandwidth bottleneck; but it would presumably be of minimal use for the actual phone-to-satellite link: The earth's surface is loaded with optically opaque clutter, and users probably wouldn't stand for a cellphone that only receives calls when it has a clear view of the sky. For some of the other users of satellite phone/data(ships, say) the requirement that the modem-pod be mounted on top of the ship, and then local RF or wire links used to distribute the connection from there, wouldn't be nearly so difficult. Either way, it wouldn't have much effect on the latency that makes satellite voice and data so annoying; the speed of light gets you either way, and any orbit you can maintain without massive fuel burn to compensate for still being in the upper atmosphere is a long way away compared to a cell tower.
Using an optical link would also, presumably, require knowing where the recipient is: they must be using some elegant tricks(perhaps borrowed from Team Astronomy, which has been battling atmospheric distortion for years?) to keep the laser on target even through thermal distortions in the atmosphere; but they still need to know where 'on target' is. Again, much bigger challenge for a human with a handset, much smaller challenge for a satellite, aircraft, or largish ship.
We could do that, or we could let advances in killbot technology reduce the costs of suppressing the squalid underclass. I'm not confident that we'll choose wisely.
Isn't "Becoming old without a pension" less of a 'fear' and more of a 'guarantee'? With the exception of a few labor unions that have really dug in and not quite been extirpated yet, we are basically all playing the tables with our 401ks(if that). 'Pensions' are what the old people who accuse you of being an entitled, lazy, little shit have.
In other pedantry, isn't 'seeing your skills and tools become irrelevant' an apt description of what would happen if an AI started doing your job?
I imagine that the tricky bit is that, legally, the penalties increase drastically when you shift from 'stealing nothing' to 'stealing something', so there's an incentive to either not touch so much as a few quarters for the parking meter; or make off with enough loot to be worth the risk and effort required to circumvent whatever security measures are in place.
The only exception would be if security and audits were pitifully lax; and you knew that discrepancies of under 1% by weight are classified as 'eh, close enough, the coins probably dried out a bit in storage...' in which case the risk would be so low that the reward wouldn't have to amount to much. In the presence of greater risk, the fact that you don't have to steal all that much for it to qualify as a felony(and that the Federal Reserve, and Brinks, probably like to make examples in order to discourage copycat offenders) would make stealing modest amounts a pretty harrowing business.
As recent college students know, laundromats almost always accept quarters as a mechanism for buying machine time, so laundering the proceeds must have been particularly easy and convenient. This guy is obviously a cerebral master of crime.
So a secret court stamped 'legal' on a classified set of rules governing what access the FBI has to the data that the NSA officially does not collect?
I think, comrade, that we might have made an error somewhere in the process of implementing this 'representative government' concept...
How dare the dead hand of state interference meddle with an industry that has gone to all the trouble of developing a ceremonial 'self regulation' procedure?
Those are valid points as well(though the peripheral drivers issue is arguably a sub-issue of the 'only supported kernel is pretty creaky' problem). They do offer a case; but not much variety, so that may or may not be an issue; EOL-ing is definitely an unknown. The kernel issue struck me as the most important, in that if a device is mainlined, at least the hardware you already own tends to take ages to be EOLed(the kernel maintainers do eventually drop platforms and devices that appear to have nobody who cares; but it takes a while; while vendor-specific BSP-forks are often slightly elderly at birth; and either dangerous or wildly obnoxious, or both, well before the hardware is showing any signs of age.
USB is certainly capable; but arguably overqualified for IR work: The classic LIRC DIY can be driven off a serial port; and I assume that a few GPIO pins that aren't too crippled would work as well. Bit-banging the ~36KHz carrier frequencies common in IR applications would probably be a pain in the ass; but the sheer ubiquity of consumer IR remotes and receivers means that handy little 'most of the low-level details and the emitter/receiver all in an IR bandpass plastic package' modules are cheap, and the actual data rates on IR remote protocols tend to be trivial.
Soft power is a bit harder to just tack on; but would be really nice, especially for embedded applications. Yes, it isn't rocket surgery to have a microcontroller set up as a watchdog and capable of switching the +5v line on the rPi; but one would think that the same feature could be integrated into the board more cheaply and elegantly than doing it after the fact.
Do you know if the SATA on the Series 4 (looks like a Marvell Kirkwood?) supports port multipliers?
My need for absolute performance in NASes is fairly limited, so the fact that a multiplier bottlenecks all the drives on it to the speed of a single SATA link isn't that important; but my experience has been that support is...deeply uneven; and when you don't have some PCIe slots to shove more HBAs into, your options are limited if a multiplier turns out not to work.
All the UDOO models have LVDS exposed. I'm afraid that I can't tell you anything useful about how versatile/well-documented that aspect of the board is. I have one that has played reasonably nicely with some headless stuff; but I haven't tried driving a panel directly.
(Incidentally, where did you get your hands on a 1600x480 LCD? I didn't even know that those existed.)
I'm not in the position to have had need of one; but doesn't team rPi provide the 'compute module' version(DDR-2 SODIMM form factor, snap it in to the carrier board of your choice) for that purpose?
They don't seem to have an updated version with the guts of the newer model as yet, though I assume that that is coming.
In that vein, the current pain-point for the ODroid-2 is the fact that the AMLogic S905 SoC it is based on has no mainline kernel support; and the current vendor fork is of a version heading toward EOL uncomfortably quickly. There is supposed to be a mainlining effort that will fix this before the current option actually goes EOL; but that remains to be seen.
I must admit that (having come into linux back in the delightful days when Broadcom wireless meant screwing around with NDISwrapper) it's a bit of a shock; but the rPi actually has an atypically high plays-well-with-others factor. You can get them cheaper; and you can get them better; but until the 'every ARM SoC is its own dysfunctional port' issue gets ironed out, some very promising hardware can end up hobbled by neurotic and antique software.
It's always entertaining to see the latest episode of 'Bitcoin: rediscovering the reasons behind the various messy hacks that we dislike about more typical currencies...'
I suspect that these starry-eyed optimists wouldn't be entirely pleased with Trump's cost reduction strategies during his years in real estate, which have included trying to go cheap on the pesky human resources; but they are correct that he is basically the only option on the republican side who is even interested in pretending to care about the filthy peons who aren't good enough to realize their income in capital gains rather than 'wages'.
It's almost as though people can't be made to vote against their economic interests by promising to keep the scary gays away from school prayer forever. Crazy stuff.
What's really sneaky is that every lab demo of a biological supercomputer is actually a production-ready biological supercomputer showing off some crude little toy that will be 5 years out for the next few decades.
The trick is just to ignore whatever is on the benchtop and grab the guy talking about it.
Plenty of room in the tape safe, thanks to the innovations in backup efficiency.
The whole point of adblockers is to 'disrupt the relationship between advertisers and consumers'; because that 'relationship' is inherently somewhere between 'adversarial' and 'cold war'. We don't go to varying levels of hassle just for fun; we do so because we fucking hate you and your 'product'.
Pretty much all of the brute-force prevention strategies involve a compromise between vulnerability to denial of service attacks and vulnerability to brute force; it's really just a matter of picking your poison based on what works best for your environment.
The 'if you screw up, the data gets wiped' approach is a good fit when the device it protects is assumed to be a mere local cache(often one with a dangerously high risk of loss/theft) of something that is backed up elsewhere. Protecting the data isn't a serious concern; but keeping it from leaking is.
For on-site situations(and anywhere where any idiot on the network can bounce authentication attempts in your name off the server as often as they like) the progressive timeout/temporary lockout ones are usually a better bet.
Not that his opinion matters nearly as much as the others(he's still loaded; but he's more busy playing the Hunter S. Thompson of tech than being a tech leader these days); but I thought that McAfee's position wasn't so much 'pro unlock' as "Me and my hacker posse will hack the shit out of it!"; which is a vote in favor of getting the contents of the phone(not that anyone is really against that, if there were some non-problematic way to do it); but not obviously a vote in favor of the feds having the right to force Apple to make it so.
Before we start worrying about cash; why not actually crack down on all those 'respectable' financial institutions that are more than happy to launder money?
Sure, maybe filthy proles use stacks of grubby bills to do their crimes; but so long as HSBC and the like can launder billions in drug money and handle transactions for a variety of theoretically-embargoed states(and face zero criminal penalties either as a company or on the part of any of the people involved); only little people smuggle cash.
Without a flamethrower-cleanup of the financial sector, making noises about high-value notes is basically just bigger criminals whining about the fact that they have to deal with lesser criminals who might find using money mules cheaper than buying bespoke financial services for the multijurisdictional client who values discretion.
Yeah, that was my intention: only cancers that are clonally transmissible, not other types of pathogens that are transmissible in the usual manner of viruses(do any bacteria or eukaryotic parasites do that sort of thing?) and often provoke the development of cancer in their host.
So far, the one that horrifies me the most is the case of the poor bastard who was parasitized by one or more tapeworms; but ultimately died when the tapeworm developed cancer and its cancer spread beyond the tapeworm and aggressively into his tissues. That's multiple levels of "people who thought that nature was red in tooth and claw were optimists". For the sake of my sound sleep, I'm pretending that that could only have happened because he was immunocompromised; and certainly represents no risk whatsoever to anyone else; because otherwise I'm going to have to start sleeping with a flamethrower under my pillow.