A couple of them have 1gbps NICs but the bottle neck for getting an actual sustained flow of data into the CPU is still the speed of the ISA bus (since none of the promised next generation of super fast buses ever actually appeared in any hardware) Umm yes they did, PCI beat out ISA years ago and now PCIe is gradually pushing out PCI.
Plus on modern boards the nic is often integrated in the southbridge anyway.
I doubt sealing to the board would be very practical, it would be very hard to get a good enough seal there and anyway most of the heat comes out of a CPU through direct conduction to the heatsink anyway so I don't see a whole lot of point in immersing the CPU itself.
What you can get easily are "waterblocks" which attatch in place of a regular heatsink and take the heat from the CPU in the regular manner but are designed to transfer it to piped water rather than to the air.
It seems given that they run off the same line it would actually be MORE trouble to make a poorer product that looks perfect than to just make exact duplicates in every way. But actually making the product isn't the only part of the process that leads to a high quality product. After making the product it needs to be tested. Preferably under conditions more intense than it will see in service.
Consider for example that a factory could make products, sells the one that pass the test legitimately, scraps the ones that don't work at all and sells the products that appear to work but fail some tests out the back door (telling the customer that they were destroyed).
Or consider that a factory could run a third shift but sells the results without bothering to test them.
And even if you limit your considerations to the actual production there are plenty of ways to reduce the cost and reliability without significantly changing the process. E.G. using worse tolerance passives (assuming you are making sufficiant parts that you are changing reels frequently anyway).
I'd interpret "same quality" as meaning the following
1: the components are the same quality 2: the testing procedure is the same 3: the reject critera are the same
Unless you are actually monitoring the complete process both for legitimate products and illegal ones or you have access to LARGE samples of both legit and pirate material and a comprehensive testsuite you have no way of verifying this.
See, for a lot of people (ie. non gamers and people not doing CPU intensive stuff) being CPU bound is rarely something they'll encounter. True but I doubt those people were in the market for intels extreme edition chips in the first place.
Multiple cores have the benefit of making the operating system more responsive since a busy app doesn't make the whole system crawl. This is true up to a point (it's very true going from single core to dual-core and somewhat true going from dual-core to quad-core) but you quickly get into diminishing returns. Typically an app will either hog just one or maybe two cores or it will be built to use as many cores as it can.
Assuming a comparable aggregate performance I'd definately take four faster cores over six slower ones.
and if we are dealing with highly parallel applications then you get a decent performance bump by throwing cores at the problem Only if those cores are actually fast enough. AMD hasn't even reached the top end of intels quad-core stuff let alone come close to touching the intel 6-core i7.
Given the smaller die area and the way intel is known for getting very high yealids I'd expect the intel eqivilients to these chips have a lower marginal cost. That means if intel desires they can push the prices to a level where AMD takes a loss on every chip without doing so themselves.
Unfortunately for AMD there is little money in being the underdog of PC processors. Intels better process technology and income from high end chips like the 6-core i7* means they can set prices on the low-midrange stuff at a level that is comfortable enough for them while being extremely painful for AMD.
*Which unlike most extreme edition chips (which tend to cost a shitload of extra money for a marginal improvement over thier regular counterparts) doesn't seem that bad a deal to me. Afaict it will get you the performance of a lower end dual-quad (and more in processes where only some stages are multithreaded) at a similar CPU cost (2.4 GHz quad core 5500 series chips seem to be about $500 each, the 6-core i7 is about $1000) without the expense of a workstation board and chassis. The dual-quad soloution supports more ram though and has more total cache (though of course in the dual-quad that cache is split between two chips) which will be an advantage in some applications.
Asimov's "3 laws of robotics" (which are what I presume you are referring to) are FAR too wishy-washy, if we ever have sentiant robots with brilliant machine vision etc they may be appropriate but that is a long way off if indeed it ever comes.
In the meantime we have to deal with simpler issues of machines (including but not limited to robots) accidently cutting/stabbing/crushing stuff without realising it is there (or realising too late)
Also when I say robots need such safety rules I mean both the robots AND the humans they interact with need those rules.
Currently the response to the dangers of machines including robots is largely to keep the humans and machines seperate. Where humans and machines have to interact guards are placed and the operators trained to keep the risk of a dangerous interaction to a minimum. This works OK for fixed industrial stuff but isn't much good for a robotic helper arround the house.
Potentially more force, more speed (which both translates to force through inertia and less time to react and stop things) but IMO most crucially different control systems.
Afaict most control systems are designed both electrically (though PID etc) and mechanically (through worm drives etc) to control position as tightly as possible regardless of external applied force. That is what makes "machining" possible. It is what makes it possible for a machine to put components on PCBs at breakneck speed.
Humans don't work like that we control force. If we hit an unexpected resistance we have to consciously apply more force. We will also generally stop applying force if either we feel pain or the person we are working with feels pain and screams. On the flip-side if a resistance we are pressing against disappears we slip all over the place.
What this means is unless the tools are extremely sharp unpowered held tools only do serious damage under very particular situations e.g. when they slip out of a cut or when someone deliberately swings them with lots of force and misses. We have safety rules to deal with this.
Robots either need very different safety rules or they need systems developed to make them respond more like humans (the people in the article seem to be working on such a system).
The only thing that matters is what Intel builds into its chipsets. Absoloutely, built into intel chipsets means cheap and easy to support which brings success.
This is why esata is fairly common these days. If the design has a sata port to spare routing it out is relatively cheap. The downside of ESATa is that you generally only get one port and the desigeners didn't include power (yes I know powered ESATA soloutions exist now but there are many unpowered ports still out there).
This is where things get interesting with USB 3, intel claim to be behind it but afaict they haven't actually put it in any of their chipsets yet. Intel are also behind light peak afaict.
Though personally I think lightpeak will end up as a niche product whatever happens. Even if the core functionality is in the chipset those optical transcievers will add hugely to the cost.
If you put mains down there you have huge product safety issues since the mains would have to be safely insulated all the way from the PSU to the display connector or you would need huge warnings to stop system builders electricuting themselves.
If you put 12V down there then you would need to seriously uprate the PSU (depending on how big a monitors you want to support you may be talking hundreds of watts). You may also have volt drop issues.
If you put a higher but still relatively safe DC voltage down there (e.g. the 25V that apple used) the PSU would have to be completely redesigned to add another high current rail.
I agree that it would have been a good idea to put USB down there though.
The trouble is that the LGA1156 platforms are somewhat lacking in terms of PCIe expandability. There are 16 2.0 lanes off the CPU which can be used for either one x16 device or two x8 devices. Much the same applies to lower end LGA775 stuff (with LGA775 the chipset determines PCIe configuration)
There are some (6-8 depending on chipset) lanes off the southbridge but they only run at 1.0 speeds.
PCIe bridges can provide a protential way out of this predicament (e.g. by taking 1.0 x4 from the southbridge and providing 2.0 x1 to a couple of devices) but add to the cost. Bridges that can work with the x16 port are even more expensive and tend to run hot as well so they are generally only seen on gamer boards.
It still stikes me as strange to split the graphics port though, especially as to connect two devices they would still need a bridge (afaict the lanes from the CPU cannot be split into anything smaller than x8)
By contrast the LGA1366 with X58 and ICH10 (afaict the only common LGA1366 configuration) gives you 20 lanes of fast PCIe off the northbridge and 6 lanes of slower PCIe off the southbridge.
One big problem i've run into is that while PCIe configuration is an important issue when selecting a motherboard (particulally a LGA1156 one with lots of extra IO devices onboard) it's often difficult to find that information. Workstation and server boards usually have a block diagram in the manual but consumer and gamer boards don't so you have to either guess or try to find a review that gives you the information.
The thing with putting a SSD in an ordinary external enclosure is that ordinary external enclosures are USB 2 at the moment and while probablly fast enough for a laptop drive (at least a laptop drive doing random access) they are going to seriously bottleneck a decent SSD.
What is the point in paying the extra for superfast storage only to bottleneck it with a shitty bus all the time?
wtf where they thinking back then? Before the advent of classless routing the only options were a class A, a class B or a class C. Therefore companies/institutions who needed more than a class B got a class A.
L1 and sometimes L2 caches are small not because of die area but because there is a tradeoff between cache size and cache speed. Only the lowest level cache (L3 on the i series) takes significant chip area (and it already takes a pretty large proportion on both the quad and hex core chips).
I'd say games qualify for "little uses". Believe it or not, most people don't use their computers for high end gaming. Afaict most people don't use their computers for anything that strains the CPU at all. Most people would be perfectly happy with a bottom of the range C2D, i3, late P4 or maybe even less as long as the system was kept free of crapware and had enough ram for their OS and applications of choice.
However of those apps that DO strain the CPU (e.g. games, video encoding, scientific software, software build systems) a lot do now have the ability to use multiple cores.
wifi is still more expensive than Ethernet Depends how you do your costing, getting a wired network proffessionally installed will almost certainly cost more than using wifi, especially if you don't want ugly cables on show.
One application is to reduce the cost of fiber to the home. If you can eliminate the $150 transceiver that converts fiber to ethernet, you've lowered the cost of entry into that market. Getting that cost down will help new companies and neighborhood coops make ftth more common. I very much doubt a system designed for short distance hookup between desktop devices (and therefore probably based on relatively cheap but lossy optical components) will be compatible with a system designed to send data kilometers.
Nor would I think you would want to connect a local desktop interconnect system directly to a relatively untrusted line.
Yes you DO take some loss of quality in the intitial digitisation and in the case of video also in the compression that is needed to get it to a manageable size and you also take some loss in the reconstruction.
However no matter how many copies of that digital file you produce you do not take any further loss and at least for audio and video the loss of the digitisation, compression and reconstruction compares favourably to even one generation of storage in analogue formats that are available to consumers.
There is no question at this point that the virtually free and relatively easy IPV4 IPs we enjoy today will stop in the not too distant future and there is NOTHING that can be done about that. Recovering legacy allocations will as you say make no significant difference to when this happens.
The question is what happens after that. It seems certain that IPs will be reassigned from less important/lucrative uses to more important/lucrative ones. The only question is will that only be allowed to happen inside companies/ISPs, will there be a grey market in IPs or will there be some kind of official resale structure.
nah companies will keep using them because their content suppliers will sue them into the ground if they don't. Their content suppliers will do this because the next level up of content suppliers would sue them into the ground if they didn't.
Broadcasters are generally limited to a particular geographical region and want to be able to advertise an exclusive in that region so content owners sell them an exclusive license for that region.
As long as internet broadcasters are taking reasonable steps to perform geographic restriction they will probablly be ok just as some spill over boarders is tolerated for traditional radio based transmissions.
What it means is that assuming you are on a normal home connection you have a very high chance of ending up behind ISP level IPV4 nat (which practically speaking means no incoming connections) as your ISP reclaims your IP for more lucrative customers and/or resale.
If you are lucky you may be offered public IPV6 addressing as well.
Slashdot Mirror
A couple of them have 1gbps NICs but the bottle neck for getting an actual sustained flow of data into the CPU is still the speed of the ISA bus (since none of the promised next generation of super fast buses ever actually appeared in any hardware)
Umm yes they did, PCI beat out ISA years ago and now PCIe is gradually pushing out PCI.
Plus on modern boards the nic is often integrated in the southbridge anyway.
I doubt sealing to the board would be very practical, it would be very hard to get a good enough seal there and anyway most of the heat comes out of a CPU through direct conduction to the heatsink anyway so I don't see a whole lot of point in immersing the CPU itself.
What you can get easily are "waterblocks" which attatch in place of a regular heatsink and take the heat from the CPU in the regular manner but are designed to transfer it to piped water rather than to the air.
It seems given that they run off the same line it would actually be MORE trouble to make a poorer product that looks perfect than to just make exact duplicates in every way.
But actually making the product isn't the only part of the process that leads to a high quality product. After making the product it needs to be tested. Preferably under conditions more intense than it will see in service.
Consider for example that a factory could make products, sells the one that pass the test legitimately, scraps the ones that don't work at all and sells the products that appear to work but fail some tests out the back door (telling the customer that they were destroyed).
Or consider that a factory could run a third shift but sells the results without bothering to test them.
And even if you limit your considerations to the actual production there are plenty of ways to reduce the cost and reliability without significantly changing the process. E.G. using worse tolerance passives (assuming you are making sufficiant parts that you are changing reels frequently anyway).
I'd interpret "same quality" as meaning the following
1: the components are the same quality
2: the testing procedure is the same
3: the reject critera are the same
Unless you are actually monitoring the complete process both for legitimate products and illegal ones or you have access to LARGE samples of both legit and pirate material and a comprehensive testsuite you have no way of verifying this.
See, for a lot of people (ie. non gamers and people not doing CPU intensive stuff) being CPU bound is rarely something they'll encounter.
True but I doubt those people were in the market for intels extreme edition chips in the first place.
Multiple cores have the benefit of making the operating system more responsive since a busy app doesn't make the whole system crawl.
This is true up to a point (it's very true going from single core to dual-core and somewhat true going from dual-core to quad-core) but you quickly get into diminishing returns. Typically an app will either hog just one or maybe two cores or it will be built to use as many cores as it can.
Assuming a comparable aggregate performance I'd definately take four faster cores over six slower ones.
and if we are dealing with highly parallel applications then you get a decent performance bump by throwing cores at the problem
Only if those cores are actually fast enough. AMD hasn't even reached the top end of intels quad-core stuff let alone come close to touching the intel 6-core i7.
Given the smaller die area and the way intel is known for getting very high yealids I'd expect the intel eqivilients to these chips have a lower marginal cost. That means if intel desires they can push the prices to a level where AMD takes a loss on every chip without doing so themselves.
Unfortunately for AMD there is little money in being the underdog of PC processors. Intels better process technology and income from high end chips like the 6-core i7* means they can set prices on the low-midrange stuff at a level that is comfortable enough for them while being extremely painful for AMD.
*Which unlike most extreme edition chips (which tend to cost a shitload of extra money for a marginal improvement over thier regular counterparts) doesn't seem that bad a deal to me. Afaict it will get you the performance of a lower end dual-quad (and more in processes where only some stages are multithreaded) at a similar CPU cost (2.4 GHz quad core 5500 series chips seem to be about $500 each, the 6-core i7 is about $1000) without the expense of a workstation board and chassis. The dual-quad soloution supports more ram though and has more total cache (though of course in the dual-quad that cache is split between two chips) which will be an advantage in some applications.
Asimov's "3 laws of robotics" (which are what I presume you are referring to) are FAR too wishy-washy, if we ever have sentiant robots with brilliant machine vision etc they may be appropriate but that is a long way off if indeed it ever comes.
In the meantime we have to deal with simpler issues of machines (including but not limited to robots) accidently cutting/stabbing/crushing stuff without realising it is there (or realising too late)
Also when I say robots need such safety rules I mean both the robots AND the humans they interact with need those rules.
Currently the response to the dangers of machines including robots is largely to keep the humans and machines seperate. Where humans and machines have to interact guards are placed and the operators trained to keep the risk of a dangerous interaction to a minimum. This works OK for fixed industrial stuff but isn't much good for a robotic helper arround the house.
Potentially more force, more speed (which both translates to force through inertia and less time to react and stop things) but IMO most crucially different control systems.
Afaict most control systems are designed both electrically (though PID etc) and mechanically (through worm drives etc) to control position as tightly as possible regardless of external applied force. That is what makes "machining" possible. It is what makes it possible for a machine to put components on PCBs at breakneck speed.
Humans don't work like that we control force. If we hit an unexpected resistance we have to consciously apply more force. We will also generally stop applying force if either we feel pain or the person we are working with feels pain and screams. On the flip-side if a resistance we are pressing against disappears we slip all over the place.
What this means is unless the tools are extremely sharp unpowered held tools only do serious damage under very particular situations e.g. when they slip out of a cut or when someone deliberately swings them with lots of force and misses. We have safety rules to deal with this.
Robots either need very different safety rules or they need systems developed to make them respond more like humans (the people in the article seem to be working on such a system).
According to the article some of these patches were only marked as important not critical.
The only thing that matters is what Intel builds into its chipsets.
Absoloutely, built into intel chipsets means cheap and easy to support which brings success.
This is why esata is fairly common these days. If the design has a sata port to spare routing it out is relatively cheap. The downside of ESATa is that you generally only get one port and the desigeners didn't include power (yes I know powered ESATA soloutions exist now but there are many unpowered ports still out there).
This is where things get interesting with USB 3, intel claim to be behind it but afaict they haven't actually put it in any of their chipsets yet. Intel are also behind light peak afaict.
Though personally I think lightpeak will end up as a niche product whatever happens. Even if the core functionality is in the chipset those optical transcievers will add hugely to the cost.
Power down a monitor cable is IMO a bad idea.
If you put mains down there you have huge product safety issues since the mains would have to be safely insulated all the way from the PSU to the display connector or you would need huge warnings to stop system builders electricuting themselves.
If you put 12V down there then you would need to seriously uprate the PSU (depending on how big a monitors you want to support you may be talking hundreds of watts). You may also have volt drop issues.
If you put a higher but still relatively safe DC voltage down there (e.g. the 25V that apple used) the PSU would have to be completely redesigned to add another high current rail.
I agree that it would have been a good idea to put USB down there though.
The trouble is that the LGA1156 platforms are somewhat lacking in terms of PCIe expandability. There are 16 2.0 lanes off the CPU which can be used for either one x16 device or two x8 devices. Much the same applies to lower end LGA775 stuff (with LGA775 the chipset determines PCIe configuration)
There are some (6-8 depending on chipset) lanes off the southbridge but they only run at 1.0 speeds.
PCIe bridges can provide a protential way out of this predicament (e.g. by taking 1.0 x4 from the southbridge and providing 2.0 x1 to a couple of devices) but add to the cost. Bridges that can work with the x16 port are even more expensive and tend to run hot as well so they are generally only seen on gamer boards.
It still stikes me as strange to split the graphics port though, especially as to connect two devices they would still need a bridge (afaict the lanes from the CPU cannot be split into anything smaller than x8)
By contrast the LGA1366 with X58 and ICH10 (afaict the only common LGA1366 configuration) gives you 20 lanes of fast PCIe off the northbridge and 6 lanes of slower PCIe off the southbridge.
One big problem i've run into is that while PCIe configuration is an important issue when selecting a motherboard (particulally a LGA1156 one with lots of extra IO devices onboard) it's often difficult to find that information. Workstation and server boards usually have a block diagram in the manual but consumer and gamer boards don't so you have to either guess or try to find a review that gives you the information.
The thing with putting a SSD in an ordinary external enclosure is that ordinary external enclosures are USB 2 at the moment and while probablly fast enough for a laptop drive (at least a laptop drive doing random access) they are going to seriously bottleneck a decent SSD.
What is the point in paying the extra for superfast storage only to bottleneck it with a shitty bus all the time?
According to that article there was a "dead mans switch" system that should have closed the valve. They also attempted to close the switch with a ROV.
This makes me wonder if yet another method for closing the valve would really have made a difference.
wtf where they thinking back then?
Before the advent of classless routing the only options were a class A, a class B or a class C. Therefore companies/institutions who needed more than a class B got a class A.
L1 and sometimes L2 caches are small not because of die area but because there is a tradeoff between cache size and cache speed. Only the lowest level cache (L3 on the i series) takes significant chip area (and it already takes a pretty large proportion on both the quad and hex core chips).
I'd say games qualify for "little uses". Believe it or not, most people don't use their computers for high end gaming.
Afaict most people don't use their computers for anything that strains the CPU at all. Most people would be perfectly happy with a bottom of the range C2D, i3, late P4 or maybe even less as long as the system was kept free of crapware and had enough ram for their OS and applications of choice.
However of those apps that DO strain the CPU (e.g. games, video encoding, scientific software, software build systems) a lot do now have the ability to use multiple cores.
wifi is still more expensive than Ethernet
Depends how you do your costing, getting a wired network proffessionally installed will almost certainly cost more than using wifi, especially if you don't want ugly cables on show.
Yeah, that's like what I was thinking too: mobs of villagers looking for werewolves or somesuch, carring torches and pitchforks. ;) http://www.wickedlasers.com/lasers/The_Torch-74-41.html
Torches like this?
One application is to reduce the cost of fiber to the home. If you can eliminate the $150 transceiver that converts fiber to ethernet, you've lowered the cost of entry into that market. Getting that cost down will help new companies and neighborhood coops make ftth more common.
I very much doubt a system designed for short distance hookup between desktop devices (and therefore probably based on relatively cheap but lossy optical components) will be compatible with a system designed to send data kilometers.
Nor would I think you would want to connect a local desktop interconnect system directly to a relatively untrusted line.
Yes you DO take some loss of quality in the intitial digitisation and in the case of video also in the compression that is needed to get it to a manageable size and you also take some loss in the reconstruction.
However no matter how many copies of that digital file you produce you do not take any further loss and at least for audio and video the loss of the digitisation, compression and reconstruction compares favourably to even one generation of storage in analogue formats that are available to consumers.
There is no question at this point that the virtually free and relatively easy IPV4 IPs we enjoy today will stop in the not too distant future and there is NOTHING that can be done about that. Recovering legacy allocations will as you say make no significant difference to when this happens.
The question is what happens after that. It seems certain that IPs will be reassigned from less important/lucrative uses to more important/lucrative ones. The only question is will that only be allowed to happen inside companies/ISPs, will there be a grey market in IPs or will there be some kind of official resale structure.
nah companies will keep using them because their content suppliers will sue them into the ground if they don't. Their content suppliers will do this because the next level up of content suppliers would sue them into the ground if they didn't.
Broadcasters are generally limited to a particular geographical region and want to be able to advertise an exclusive in that region so content owners sell them an exclusive license for that region.
As long as internet broadcasters are taking reasonable steps to perform geographic restriction they will probablly be ok just as some spill over boarders is tolerated for traditional radio based transmissions.
What it means is that assuming you are on a normal home connection you have a very high chance of ending up behind ISP level IPV4 nat (which practically speaking means no incoming connections) as your ISP reclaims your IP for more lucrative customers and/or resale.
If you are lucky you may be offered public IPV6 addressing as well.