informit.com— As Moore's Law continues to hold, IC designers are finding that they have more and more silicon real estate to play with. David Chisnall hazards some guesses as to what they might do with it.
Oct 29, 2006View in Crawl 4
Whats next? For me, the only future for CPU's is the DIY set.. If, in 5 years, I can't just print myself a multi-core system on my desktop with integrated components on-chip, I'm leaving the computing industry ...Honest, its time we broke this hegemony that chip-makers have over us computer-consumers. Give us the tools to build these machines ourselves, please, won't someone ..
"More coprocessors will be added to both CPUs and GPUs, until the two eventually merge"Why would you want this? CPUs are designed to be generalized hardware, so that they can run any code quickly. Adding a GPU, hardware specifically designed to run graphics code quickly, defeats the purpose of the GPU. Even as the GPU is becoming more generalized and retaskable itself, it's orchestrating its own demise.Eventually, people will realize it's not worth paying hundreds of dollars for an additional floating point core, and the CPU manufacturers will just throw smarter FPUs in their CPUs, ridding us of the scar that is the GPU for another development cycle (yes, of course, someone will respawn it again, and we'll probably go through this whole cycle of integration vs. seperation again). The real future of computers won't come until retaskable logic becomes more accessible. Instead of manufacturer's bickering over who should add what hardware to which piece of the computer, we just stick a blank cartridge into our machines and load the exact firmware image we need. Augment them with fast logic (like Xilinx's PPC-enhanced models) and you've got an architecture that supports being a GPU, or a CPU, or a PPU, or anywhere between. And with gate densities pushing higher and higher, why not? Most of the disadvantages of retaskable logic goes away as you have more densely constructed processors and can drive up the clock rate. Another future almost completely ignored by the general public is the microcontroller revolution; computing is now finally cheap enough that literally every product we buy can have a computer in it. ARMs have went from being the soldiers of the cable box to being in almost every piece of consumer electronics you can think of, from iPods to the Nintendo DS to your kid's toys like the Pixter. These computers are getting smaller, faster, and more feature rich monthly. Computers more powerful than the original Apple II can be purchased in bulk for less than $4/processor. This is something huge to think about, especially as the more traditional chip manufacturer's don't compete in this segment of the market at all (neither AMD nor Intel have very deep embeddable designs; AMD has Geode, which is more embeddable than most of Intel's offerings based on the Pentium M, but neither company has anything that can compete with ARM9's and -7's.) We're even starting to see dual-core MCUs like the iPod's former SoC from PortalPlayer.Very clearly, the future is change. Even as AMD and Intel compete for the fastest CPU, more and more people are realizing that what they've got is clearly fast enough, and more and more engineers are realizing they can do much, much more with less. We're starting to see companies trying to build computers for the low-end; instead of seeing how cheaply we can stick a Pentium 4 in something and sell it, we're designing computers like the OLPC CM1, which, while being a radical departure from "faster is better", still is adeqate to do most of the tasks we actually need our computers to do. [As an aside, imagine how cheaply you could make, manufacture, and embed an original Pentium Pro processor, redesigned for a much smaller manufacturing process like 90nm].
@geminitojanusGreat comment.@toomuchgreenteaWhat are you on about? CMOS stands for Complementary Metal Oxide Semiconductor, as far as I know. If you're aware of materials that can be used to make transistors that aren't metal oxides, let me know!
"Well i buried as lame because anyone that can be bothered reading past the first paragraph must be a somewhat hardcore geek"How the hell does that make it lame? Some people who read digg.com do actually give a f*** about the future of CPU technology you know.
While quantum computing is a good pointer, any logic done on the spin (its not actually spin as generally conceived in the macroscopic world, just a name for a property of matter) of an electron (lepton) or a quark isn't exactly electronics -- it's spintronics (<a class="user" href="http://en.wikipedia.org/wiki/Spintronics).">http://en.wikipedia.org/wiki/Spintronics).</a> It has nothing to do with the charge on the electron.
Moore's law will be replaced with a new law that is much faster once silicon approaches its limits. I heared that one cubic centimeter of nanotube based semiconductors could theoretically exceed the computational power of the human brain (in terms of how much information the brain can process, not in terms of how well we do math) by a factor of one million. So there is certainly a lot of room to grow. Moore's law is only one example of the overall trend of technology to improve exponentially.
roprotOct 30, 2006
Whats next? For me, the only future for CPU's is the DIY set.. If, in 5 years, I can't just print myself a multi-core system on my desktop with integrated components on-chip, I'm leaving the computing industry ...Honest, its time we broke this hegemony that chip-makers have over us computer-consumers. Give us the tools to build these machines ourselves, please, won't someone ..
geminitojanusOct 30, 2006
"More coprocessors will be added to both CPUs and GPUs, until the two eventually merge"Why would you want this? CPUs are designed to be generalized hardware, so that they can run any code quickly. Adding a GPU, hardware specifically designed to run graphics code quickly, defeats the purpose of the GPU. Even as the GPU is becoming more generalized and retaskable itself, it's orchestrating its own demise.Eventually, people will realize it's not worth paying hundreds of dollars for an additional floating point core, and the CPU manufacturers will just throw smarter FPUs in their CPUs, ridding us of the scar that is the GPU for another development cycle (yes, of course, someone will respawn it again, and we'll probably go through this whole cycle of integration vs. seperation again). The real future of computers won't come until retaskable logic becomes more accessible. Instead of manufacturer's bickering over who should add what hardware to which piece of the computer, we just stick a blank cartridge into our machines and load the exact firmware image we need. Augment them with fast logic (like Xilinx's PPC-enhanced models) and you've got an architecture that supports being a GPU, or a CPU, or a PPU, or anywhere between. And with gate densities pushing higher and higher, why not? Most of the disadvantages of retaskable logic goes away as you have more densely constructed processors and can drive up the clock rate. Another future almost completely ignored by the general public is the microcontroller revolution; computing is now finally cheap enough that literally every product we buy can have a computer in it. ARMs have went from being the soldiers of the cable box to being in almost every piece of consumer electronics you can think of, from iPods to the Nintendo DS to your kid's toys like the Pixter. These computers are getting smaller, faster, and more feature rich monthly. Computers more powerful than the original Apple II can be purchased in bulk for less than $4/processor. This is something huge to think about, especially as the more traditional chip manufacturer's don't compete in this segment of the market at all (neither AMD nor Intel have very deep embeddable designs; AMD has Geode, which is more embeddable than most of Intel's offerings based on the Pentium M, but neither company has anything that can compete with ARM9's and -7's.) We're even starting to see dual-core MCUs like the iPod's former SoC from PortalPlayer.Very clearly, the future is change. Even as AMD and Intel compete for the fastest CPU, more and more people are realizing that what they've got is clearly fast enough, and more and more engineers are realizing they can do much, much more with less. We're starting to see companies trying to build computers for the low-end; instead of seeing how cheaply we can stick a Pentium 4 in something and sell it, we're designing computers like the OLPC CM1, which, while being a radical departure from "faster is better", still is adeqate to do most of the tasks we actually need our computers to do. [As an aside, imagine how cheaply you could make, manufacture, and embed an original Pentium Pro processor, redesigned for a much smaller manufacturing process like 90nm].
jamboxOct 30, 2006
@geminitojanusGreat comment.@toomuchgreenteaWhat are you on about? CMOS stands for Complementary Metal Oxide Semiconductor, as far as I know. If you're aware of materials that can be used to make transistors that aren't metal oxides, let me know!
svpirateOct 30, 2006
"Well i buried as lame because anyone that can be bothered reading past the first paragraph must be a somewhat hardcore geek"How the hell does that make it lame? Some people who read digg.com do actually give a f*** about the future of CPU technology you know.
haeliosOct 30, 2006
While quantum computing is a good pointer, any logic done on the spin (its not actually spin as generally conceived in the macroscopic world, just a name for a property of matter) of an electron (lepton) or a quark isn't exactly electronics -- it's spintronics (<a class="user" href="http://en.wikipedia.org/wiki/Spintronics).">http://en.wikipedia.org/wiki/Spintronics).</a> It has nothing to do with the charge on the electron.
almadielOct 30, 2006
Moore's law will be replaced with a new law that is much faster once silicon approaches its limits. I heared that one cubic centimeter of nanotube based semiconductors could theoretically exceed the computational power of the human brain (in terms of how much information the brain can process, not in terms of how well we do math) by a factor of one million. So there is certainly a lot of room to grow. Moore's law is only one example of the overall trend of technology to improve exponentially.
mastertopOct 30, 2006
I think somewhen, binary system won't exist anymore, it will be like hexadecimal and could be much more efficient...
tmcdiggNov 2, 2006
LASERS