One thing you're missing out are isssues to do with memory sizes and pin limitiations (ie costs) at the low end.... bascily SDRAM works well when it's wide (64-bits or 128-bits) if you want to use the current state-of-the-art DRAM technologies that puts a low end to the size of your memory (and as a result a low end to the possible cost of the memory subsystem). Using so many pins also limits your package size (and potentially even die sized because the pad ring needs the equivalent of maybe 30-60 extra pads).
This is the main reason that RDRAM first appeared in graphics subsystems (with maximum frame buffer sizes - rambus appeared at about the same time that VRAM sizes grew so that a 64-bit wide framebuffer needed more VRAM that the size of the framebuffer) - it also showed up in game systems about the same time for the same reasons.
I also think there are real high-end places that RDRAM could be better (assuming the price comes down) - this is not so much due to the high bandwidth but due to the high potential concurrency (basicly the ability to sense many DRAM rows in parallel) - for today's CPUs with slow serializing buses between the CPU and the memory controller very little concurrency gets exposed to the memory subsystem - latency becomes everything - but on the other side of that bus is a superscalar CPU which is highly concurrent - potentially we could see 5-10 councurrent memory accesses (maybe more if you had speculative row senses) - applied to the memory if it were 'close by'.
Anyway - my point - there are real places where rambus makes sense - where they are changes from time to time (the higher cost IS a big deal).
(disclaimer - I'm a chip designer - I don't work for rambus, don't own any stock - but I've done 3 rambus based designs over the years for the above reasons [and 2 VRAM based ones and 1 SDRAM based one - you pick your design's sweet-spot whereever you can find it])
It's a relatively common CPU originally out of the UK - now made by a number of manufacturers (Compaq, Intel, Cirrus...) it's a RISCier RISC than most - but it has just enough goodies to happily boot Linux
he sais "it's an intelectual property destroyer"... of course Linux doesn't destroy M$'s IP - they still own what they've always owned.... it just may not be worth as much.... but that's always been the case for a supplier who suddenly finds a low-cost competitor in the market.... it ought to be an incentive for M$ to make better IP more cheaply.... (ie no more M$ cash cow for certain senior M$ executives....)
What I think if unfair is the fact that only select few are allowed to do this.
Do you mean launch bigger rockets.... check out Tripoli find a local group and bowl on down. The Arliss boosters are Ms so you need to work your way up to a level 3 confirmation if you want to do an exact cansat type flight but you could do close enough off of a minimum diameter K. Arliss itself maay start happening in other places - there's been some mention of expanding it across the US so that school kids all over could get a chance to build a payload
the point is to get the experience of launching cheaply - it costs ~$400 (the AP cost) to put 3 cansats up and for that 3 university or school teams get to run their airframe thru the G forces of putting it up there and test their telemetry in a real-world environment - weather baloons are MORE expensive, less fun, harder to recover (they drift for miles on their way up) and don't provide the G shock on launch and deployment
The idea is that the chutes and altitude chosen keep it up for about the same time as a low orbit pass - the builder gets to run their design thru all the G force of launch and deployment - then have the time for one 'pass' to get their telemetry up and running and downlink while the airframe is still in the air.
We know we're not putting them in orbit (we launch to 100k ft from the same site - and even 20 miles up aint LEO either:-) the idea is to get some really cheap experience at doing something real.
BTW they have an autonomous rover competition - same payload size as the usual 3 coke can deployment system - we drop it from 15k - you have to make it come back to a defined spot (it's on an almost flat playa so it's easier than you think - you do have to dodge the cars parked around the launch site:-)
The coke cans are dropped out at only 12k ft (the launch vehicle puts up 3 at a time) they are dropped on a parachute - the hang-time is about the same as for the sky-time in a single micro-sat pass so it's a great way to test if your payload can handle the stresses of launch and test your downlick hardware and software in real-world conditions....
Arliss is growing.... there are more and more payloads going up every year (dates for this year are here) - and now they have a rover contest - launch your rover to 10k ft have it return and find it's way back autonomously to a designated target. I
also hear plans are being made to extend the launch sites across the country
KDE has done the same for a long time (pre 2.0) to provide this sort of network transparency... I read between the lines here to see the Gnome mob checking a bullet on their "KDE feature list" in the ongoing KDE/Gnome mini-conflict (unlike many I don't think this competition is a bad thing - having KDE continually raising the bar for Gnome and vice-versa means we all win)
well sure - I was assuming that people actually want to buy what you're making:-) when you reach market tends to effect how much profit you can skim - not the basic economics of what the chip costs (and what percentage of the cost is NRE and what is fab cost - which was what I was addressing in my post). Basicly if someone beats you to market with an equivalent product you're price is set by their price and demand - they get to skim the early profit and work on their volume prices and manufacturability - you have a much harder road - you have to spend a lot longer writing off your NRE.
By themselves R&D times just push up the basic NRE - in the real world of course things like these are never that simple - I've done chips where we were so far ahead of the competition (1-2 complete cycles) where we just wallowed in the $$$ - and others where management dithered changing the project back and forward from month to month to the point that the R&D was eventually abandoned because the market window had been missed (guess which project took the longest - and I worked the hardest on.... and threw the most usefull work away on...:-)
Don't forget Arliss which the cubesat project has grown out of.... this is a project where students build coke-can sized payloads (the launch vehicle puts up 3 at a time) that are launched to 12k ft and dropped on a parachute - the hang-time is about the same as for the sky-time in a single micro-sat pass so it's a great way to test if your payload can handle the stresses of launch and test your downlick hardware and software in real-world conditions....
Arliss is growing.... there are more and more payloads going up every year - and now they have a rover contest - launch your rover to 10k ft have it return and find it's way back autonomously to a designated target
First, an idea must be developed into a design, and then the design must be verified and tested. Then, after that, a great deal of resources will be put into setting up a production line to produce the design. These two steps require anywhere from 75% to 95% of the total resources involved in developing a new chip. Once the design is in production, the incremental cost of making another chip (from raw materials and running the production line itself) is extremely small.
This is certainly true for small production runs (ie under say 100k parts) design costs and tapeout NRE dominate - and it's true - those 2 things do cost the bulk of developing a new chip (because those 2 things basicly are 'developing a new chip') - but for really large production runs you care about yield, wafer cost, testing and packaging - to a first estimate:
cost per die = cost per wafer/(dies per wafer * yield)
cost per chip = cost per die + package cost
testing cost is a little harder to figure in since some of it is done for diesort and some after packaging (where you have to include all the bad packages you toss away). Yield is also highly proportional to dies/wafer - if you have 10 random defects on a wafer then if you are pulling 1000 small dies off of the wafer you're likely to get a 99.9% yield - but with 10 dies per wafer you might get a 10% yield if you are lucky.
How you do actually accounting of the NRE depends a lot on your business model - but for a lot of companies who are using an outside fab it's it's more likely to be tossed in on top of the packaged chip cost along with things like profit and marketting costs.
Back to the original point - if you spent $5M of engineer costs and NRE to develop a chip that costs $20 packaged and you are selling 10 Million of them you only need to add 50c on top of the packaged chip cost - this is chicken feed next to the package itself which might cost you anything from $1-5. On the other hand if you are selling 10k chips you have to amortize your R&D at $500/chip....... and unless you have one of those military contracts for hammers your boss will probably be having his head handed to him on a plate by senior management....
why every year they organize the production of a flu vaccine expected to match the viruses live in the general population for that year.... with this long term example of how to handle infectious inofrmation diseases (be they bits or RNA) it would appear to me to be obvious to anyone working in the field...
Imagine a screen of 1000x768 pixels, at 50fps. Thats 36 million pixels per second. Therefore, logically, when we reach 50 million polygons/second in calculation for a graphics chip, it is effectively impossible to make the graphics quality any better without improving the quality of the screen.
as many others have replied you're missing a lot about what pixel rates are about (hidden pixels, alpha blending, antialiasing etc etc)
However there is a grain of truth in what you're getting at that in the future may eventually result in a whole new generation of hardware. basicly it's this - the number of visible pixels on the screen isn't really changing much, certainly not at the same rate that the ability of silicon to manipulate them is.... which means that rendering techniques that are proportional to the number of pixels (rather than screen complexity) may become more interesting - for example - ray tracing - the number of rays is (to a 1st approximation) is proportional to the number of pixels rather than the scene complexity (however the cost of processing each ray also goes up with scene complexity - but not necessarily always at the same rate if you are carefull)....
Just about 100 years ago the cops couldn't tap phones to solve crimes, there weren't any - before that messages were carried by hand, or memorized... the window during which they've had the chance to do this is relatively small in the grand scheme of things - just a couple of generations (just long enough to forget how things used to work:-). These days the terrorists could equally be using coded short wave radio transmissions - the net's just aconvenient whipping boy because the feds think they have chance of regulating it
What we're really talking about here is a balance between our personal privacy and the public safety - we should be be carefull not to race off and give away our privacy when in practice all it will mean is that the black hats will use different technology - if that happens we've all lost and the feds have gained nothing...
As other have pointed out the bulk of most large chips these days are designed in high-level languages (Verilog, and to a lesser extent VHDL) which look a lot like existing programming languages.
You can learn either language relatively quickly if you know some other programming langauages - but using them for logic design does require a grounding in hardware that's outside of normal programming experience (wires, low level concurrency etc etc) - and converting a working simulation into a physical device requires a lot more infrastructure than most people have on hand....
is that creating an OS core is that it's a lot harder and more expensive to make up a CPU design and run it thru a fab than it is to hack up a piece of software and post it on the net.
Software you can do at home - hardware takes infrastructure that most people don't have access to (fabs, NRE, synthesis and layout tools etc)
After all to build a real high-end CPU you need a the work of maybe 100 people for a couple of years...
For these reasons I think that things like OpenCores make sense for companies but not so much for individuals (however much I wish it were otherwise:-( )
just remember - no taxation without representation..... and if you're not a citizen the US will happily tax you and not let you vote (unlike some other western countries).... all that boston tea party patriotic rah-rah is just window dressing. The ONLY way you can make yourself heard here is with your money - spending it with a group like the EFF is an excellent way to do it.
Beside don't you think the RIAA has members that are based outside the US that are funding its operations?
I doubt this technology will work for high-density state-of-the-art sorts of things (like CPUs for example) - it probably makes sense for something where transistor size doesn't have to be submicron (like in a TFT display). However for a 1GHz CPU it's not going to cut it (they have to be small because of little things like the speed of light and RC effects in wires).
On the other hand there's a whole range of electronics out there where this sort of density is not an issue and this could make a lot of older fabs that are building this stuff redundant.
I could imagine a cool disk drive replacement with this technology - basicly a pile of mylar sheets - I bet you could get comparable densities at similar prices.... and you wouldn't have to spin them....
in their day I think the TOPs guys hated the VMS guys more - because the vax ended up killing the 10/20 line and both groups were kind of competing for DECs internal R&D attention - back then Unix was a bit of a side show - its 'rise' sort of happened just about the time the 10/20 line was axed so there wasn't a lot of overlap
it starts with - 'if (document.all)
location.href="http://aspalliance.com/dagon/";
else
if (document.referrer.toLowerCase().indexOf("slashdot.org")>-1)
(insert rant here)'.
That's right it special cases people who come from slashdot.... so I suspect people are seeing several different things on this page (Konqueror just ignores this stuff and continues on to the main page)
Other research showed that cells called melanocytes found in the uveal layer started growing and dividing more rapidly when exposed to microwave radiation.
Since uveal melanoma starts within such cells, there is a ready-made mechanism by which mobile phone radiation might help to initiate cancer, especially in people with a genetic predisposition to the condition.
This is a poor explanation. If microwaves cause unusual growth rates among the melanocytes, then microwave radiation could act as a promoter, not an initiator, of cancer. Ultraviolet light is both a promoter and an initiator, because it is high-enough energy to ionize DNA molecules and cause mutations. Microwaves are non-ionizing, so there is no known mechanism by which they could act as an initiator.
Not necessarily - for example UV causing initiaion of cancerous cells might be common in the light sensitive cells in the eye - but the body's normal defense mechanisms might be able to keep it normally under check - if microwave heating causes these cells to reproduce more quickly it might raise the chances of cancer taking hold - maybe even in a non-linear manner.
Personally I found the article to be appropriately non-committal reporting a result but reccomending other studies reproducing the results. I think that suggesting possible mechanisms for explaining the information that their staistical study has discovered is also appropriate - they may be wrong - but if you figure out somethings up the way to figure out how the correlation is to start hypothesising and testing those hypotheses - that's the scientific method.
This sort of way of doing science is normal - it's just that in this situation the results are potentially a PR firestorm for the cell phone industry so anything anyone sais will be explosive
Slashdot Mirror
This is the main reason that RDRAM first appeared in graphics subsystems (with maximum frame buffer sizes - rambus appeared at about the same time that VRAM sizes grew so that a 64-bit wide framebuffer needed more VRAM that the size of the framebuffer) - it also showed up in game systems about the same time for the same reasons.
I also think there are real high-end places that RDRAM could be better (assuming the price comes down) - this is not so much due to the high bandwidth but due to the high potential concurrency (basicly the ability to sense many DRAM rows in parallel) - for today's CPUs with slow serializing buses between the CPU and the memory controller very little concurrency gets exposed to the memory subsystem - latency becomes everything - but on the other side of that bus is a superscalar CPU which is highly concurrent - potentially we could see 5-10 councurrent memory accesses (maybe more if you had speculative row senses) - applied to the memory if it were 'close by'.
Anyway - my point - there are real places where rambus makes sense - where they are changes from time to time (the higher cost IS a big deal).
(disclaimer - I'm a chip designer - I don't work for rambus, don't own any stock - but I've done 3 rambus based designs over the years for the above reasons [and 2 VRAM based ones and 1 SDRAM based one - you pick your design's sweet-spot whereever you can find it])
It's a relatively common CPU originally out of the UK - now made by a number of manufacturers (Compaq, Intel, Cirrus ...) it's a RISCier RISC than most - but it has just enough goodies to happily boot Linux
Well only if rats are your greatest fear .... the rest of us will get Win3.1 on our computers
(and so long as you don't try to break the encryption and attempt to make freedom for yourself)
he sais "it's an intelectual property destroyer" ... of course Linux doesn't destroy M$'s IP - they still own what they've always owned .... it just may not be worth as much .... but that's always been the case for a supplier who suddenly finds a low-cost competitor in the market .... it ought to be an incentive for M$ to make better IP more cheaply .... (ie no more M$ cash cow for certain senior M$ executives ....)
Do you mean launch bigger rockets .... check out Tripoli find a local group and bowl on down. The Arliss boosters are Ms so you need to work your way up to a level 3 confirmation if you want to do an exact cansat type flight but you could do close enough off of a minimum diameter K. Arliss itself maay start happening in other places - there's been some mention of expanding it across the US so that school kids all over could get a chance to build a payload
the point is to get the experience of launching cheaply - it costs ~$400 (the AP cost) to put 3 cansats up and for that 3 university or school teams get to run their airframe thru the G forces of putting it up there and test their telemetry in a real-world environment - weather baloons are MORE expensive, less fun, harder to recover (they drift for miles on their way up) and don't provide the G shock on launch and deployment
We know we're not putting them in orbit (we launch to 100k ft from the same site - and even 20 miles up aint LEO either :-) the idea is to get some really cheap experience at doing something real.
BTW they have an autonomous rover competition - same payload size as the usual 3 coke can deployment system - we drop it from 15k - you have to make it come back to a defined spot (it's on an almost flat playa so it's easier than you think - you do have to dodge the cars parked around the launch site :-)
The coke cans are dropped out at only 12k ft (the launch vehicle puts up 3 at a time) they are dropped on a parachute - the hang-time is about the same as for the sky-time in a single micro-sat pass so it's a great way to test if your payload can handle the stresses of launch and test your downlick hardware and software in real-world conditions....
Arliss is growing .... there are more and more payloads going up every year (dates for this year are here) - and now they have a rover contest - launch your rover to 10k ft have it return and find it's way back autonomously to a designated target. I
also hear plans are being made to extend the launch sites across the country
KDE has done the same for a long time (pre 2.0) to provide this sort of network transparency ... I read between the lines here to see the Gnome mob checking a bullet on their "KDE feature list" in the ongoing KDE/Gnome mini-conflict (unlike many I don't think this competition is a bad thing - having KDE continually raising the bar for Gnome and vice-versa means we all win)
By themselves R&D times just push up the basic NRE - in the real world of course things like these are never that simple - I've done chips where we were so far ahead of the competition (1-2 complete cycles) where we just wallowed in the $$$ - and others where management dithered changing the project back and forward from month to month to the point that the R&D was eventually abandoned because the market window had been missed (guess which project took the longest - and I worked the hardest on .... and threw the most usefull work away on ... :-)
Arliss is growing .... there are more and more payloads going up every year - and now they have a rover contest - launch your rover to 10k ft have it return and find it's way back autonomously to a designated target
This is certainly true for small production runs (ie under say 100k parts) design costs and tapeout NRE dominate - and it's true - those 2 things do cost the bulk of developing a new chip (because those 2 things basicly are 'developing a new chip') - but for really large production runs you care about yield, wafer cost, testing and packaging - to a first estimate:
testing cost is a little harder to figure in since some of it is done for diesort and some after packaging (where you have to include all the bad packages you toss away). Yield is also highly proportional to dies/wafer - if you have 10 random defects on a wafer then if you are pulling 1000 small dies off of the wafer you're likely to get a 99.9% yield - but with 10 dies per wafer you might get a 10% yield if you are lucky.How you do actually accounting of the NRE depends a lot on your business model - but for a lot of companies who are using an outside fab it's it's more likely to be tossed in on top of the packaged chip cost along with things like profit and marketting costs.
Back to the original point - if you spent $5M of engineer costs and NRE to develop a chip that costs $20 packaged and you are selling 10 Million of them you only need to add 50c on top of the packaged chip cost - this is chicken feed next to the package itself which might cost you anything from $1-5. On the other hand if you are selling 10k chips you have to amortize your R&D at $500/chip ....... and unless you have one of those military contracts for hammers your boss will probably be having his head handed to him on a plate by senior management ....
why every year they organize the production of a flu vaccine expected to match the viruses live in the general population for that year .... with this long term example of how to handle infectious inofrmation diseases (be they bits or RNA) it would appear to me to be obvious to anyone working in the field ...
as many others have replied you're missing a lot about what pixel rates are about (hidden pixels, alpha blending, antialiasing etc etc)
However there is a grain of truth in what you're getting at that in the future may eventually result in a whole new generation of hardware. basicly it's this - the number of visible pixels on the screen isn't really changing much, certainly not at the same rate that the ability of silicon to manipulate them is .... which means that rendering techniques that are proportional to the number of pixels (rather than screen complexity) may become more interesting - for example - ray tracing - the number of rays is (to a 1st approximation) is proportional to the number of pixels rather than the scene complexity (however the cost of processing each ray also goes up with scene complexity - but not necessarily always at the same rate if you are carefull) ....
What we're really talking about here is a balance between our personal privacy and the public safety - we should be be carefull not to race off and give away our privacy when in practice all it will mean is that the black hats will use different technology - if that happens we've all lost and the feds have gained nothing ...
You can learn either language relatively quickly if you know some other programming langauages - but using them for logic design does require a grounding in hardware that's outside of normal programming experience (wires, low level concurrency etc etc) - and converting a working simulation into a physical device requires a lot more infrastructure than most people have on hand ....
Software you can do at home - hardware takes infrastructure that most people don't have access to (fabs, NRE, synthesis and layout tools etc)
After all to build a real high-end CPU you need a the work of maybe 100 people for a couple of years ...
For these reasons I think that things like OpenCores make sense for companies but not so much for individuals (however much I wish it were otherwise :-( )
my first thought was ..... "cool now I can avoid PacBell and put my office directly on the grid ..."
Beside don't you think the RIAA has members that are based outside the US that are funding its operations?
On the other hand there's a whole range of electronics out there where this sort of density is not an issue and this could make a lot of older fabs that are building this stuff redundant.
I could imagine a cool disk drive replacement with this technology - basicly a pile of mylar sheets - I bet you could get comparable densities at similar prices .... and you wouldn't have to spin them ....
My firewall is logging one or 2 attempts at port 111 each day, and slightly more attempts to access my non-existant FTP server ....
in their day I think the TOPs guys hated the VMS guys more - because the vax ended up killing the 10/20 line and both groups were kind of competing for DECs internal R&D attention - back then Unix was a bit of a side show - its 'rise' sort of happened just about the time the 10/20 line was axed so there wasn't a lot of overlap
That's right it special cases people who come from slashdot .... so I suspect people are seeing several different things on this page (Konqueror just ignores this stuff and continues on to the main page)
Since uveal melanoma starts within such cells, there is a ready-made mechanism by which mobile phone radiation might help to initiate cancer, especially in people with a genetic predisposition to the condition.
This is a poor explanation. If microwaves cause unusual growth rates among the melanocytes, then microwave radiation could act as a promoter, not an initiator, of cancer. Ultraviolet light is both a promoter and an initiator, because it is high-enough energy to ionize DNA molecules and cause mutations. Microwaves are non-ionizing, so there is no known mechanism by which they could act as an initiator.
Not necessarily - for example UV causing initiaion of cancerous cells might be common in the light sensitive cells in the eye - but the body's normal defense mechanisms might be able to keep it normally under check - if microwave heating causes these cells to reproduce more quickly it might raise the chances of cancer taking hold - maybe even in a non-linear manner.
Personally I found the article to be appropriately non-committal reporting a result but reccomending other studies reproducing the results. I think that suggesting possible mechanisms for explaining the information that their staistical study has discovered is also appropriate - they may be wrong - but if you figure out somethings up the way to figure out how the correlation is to start hypothesising and testing those hypotheses - that's the scientific method.
This sort of way of doing science is normal - it's just that in this situation the results are potentially a PR firestorm for the cell phone industry so anything anyone sais will be explosive