They could continue adding power to the chips, but unfortunately there's a thermal limit to how much heat can be dissipated by fans blowing atmospheric air. Not everyone has liquid nitrogen tanks to cool their computers :(
fizzy
There are predictions that the Transistor Density, which has been following Moore's Law, will flatten out sometime soon because of physical limitations of how small a transistor can get correct? Then parallelism becomes more important unless some other method is found to increase performance in a chip.
pagerank
This page argues for the necessity of learning parallel computing as a developer. The performance of a single core depends on the frequency and the instruction level parallelism (ILP). The frequency has a limit because of the power. ILP is limited because a lot of dependency is unavoidable in a single thread program. So here comes the multi-core processors. To make the program running on multiple cores behave efficiently and correctly, developers should consider more design choices beyond the workflow of the original signal source program.
Drizzle
Is there any reason that Intel has not added better coolers to their chips? For example, it's not particularly safe to overclock on intel's default cooler, but you can increase performance by up to 20/30% by overclocking with a reasonable fan / liquid cooler. If single thread speed is so important, it seems like spending a bit more on the cooler would be a priority, given the increase in speedup.
Proof of concept: I have a 75 dollar liquid cooler on my i7-4790k and I have it clocked up to 4.9GHZ from a base of 4.0GHZ
rjvani
As an extension to @fizzy, it seems like increasing the number of transistors for microprocessors was really important during initial development (allowing for the transition from 8 to 32-bit processors). As a result, these physical limitations of transistor density has enabled engineering multi-level caches on a chip. These decisions further imply consideration of what actually goes on and off a chip to ensure performance actually increases as time goes on.
bebe
Since there probably is programmed that cannot take advantage of parallel execution, what other ways can we hope to improve the performance of this program. Is there anyway program-wised to improve the performance of this program? Is there no solution other than improving the hardware?
They could continue adding power to the chips, but unfortunately there's a thermal limit to how much heat can be dissipated by fans blowing atmospheric air. Not everyone has liquid nitrogen tanks to cool their computers :(
There are predictions that the Transistor Density, which has been following Moore's Law, will flatten out sometime soon because of physical limitations of how small a transistor can get correct? Then parallelism becomes more important unless some other method is found to increase performance in a chip.
This page argues for the necessity of learning parallel computing as a developer. The performance of a single core depends on the frequency and the instruction level parallelism (ILP). The frequency has a limit because of the power. ILP is limited because a lot of dependency is unavoidable in a single thread program. So here comes the multi-core processors. To make the program running on multiple cores behave efficiently and correctly, developers should consider more design choices beyond the workflow of the original signal source program.
Is there any reason that Intel has not added better coolers to their chips? For example, it's not particularly safe to overclock on intel's default cooler, but you can increase performance by up to 20/30% by overclocking with a reasonable fan / liquid cooler. If single thread speed is so important, it seems like spending a bit more on the cooler would be a priority, given the increase in speedup.
Proof of concept: I have a 75 dollar liquid cooler on my i7-4790k and I have it clocked up to 4.9GHZ from a base of 4.0GHZ
As an extension to @fizzy, it seems like increasing the number of transistors for microprocessors was really important during initial development (allowing for the transition from 8 to 32-bit processors). As a result, these physical limitations of transistor density has enabled engineering multi-level caches on a chip. These decisions further imply consideration of what actually goes on and off a chip to ensure performance actually increases as time goes on.
Since there probably is programmed that cannot take advantage of parallel execution, what other ways can we hope to improve the performance of this program. Is there anyway program-wised to improve the performance of this program? Is there no solution other than improving the hardware?