I'm skeptical. I don't know the exact breakpoint, but pushing enough power through the small (and very dense) circuits of a processor can't go beyond a certain level based on our current CPU cooling technology. The heat at that density just becomes too high and too localized.
Liquid nitrogen or helium. But if someone went to the trouble to use these substances, they'd likely show it in action instead of some computer screen.
Late reply, I know :( - Actually, it's not even the cooling outside the CPU that's the real problem, it's moving heat away at a microscopic level near each transistor. With enough power going through enough circuits nearby, the material the CPU is made out of becomes the issue.
Consider a block of metal submerged in a (hypothetical) liquid at near 0 degrees K. Despite how much heat this liquid can draw out of the block, if heat is being generated too quickly at the very center of the block, it's possible that the heat conductivity of the metal block itself limits the efficacy of outside cooling, resulting in an "overheating" center.
I'm a bit saddened that nearly ten years after we reached 3GHz, desktop clock speed increase has come to a halt.
I know that FSB speeds are irrelevant now more than ever with multithreading and multi-core architectures, but the performance afforded by high-clock chips in high-demand areas, along with multi-core technology, it seems could be far greater than it is now.
Especially with several 8GHz nodes, it seems like cheaper supercomputing could be more viable.
Am I wrong, right, misguided, inaccurate, incorrect?
Clock speed hasn't increased, but instructions per second per core has more than doubled over the last decade, and instructions per clock cycle per core continues to increase (http://en.wikipedia.org/wiki/Instructions_per_second). Slower than Moore's law, but it's something.
> I know that FSB speeds are irrelevant now more than ever with multithreading and multi-core architectures
Clock speeds have been irrelevant ever since there has been more than one architecture for the same instruction set. You cannot compare a 3GHz i7 with a 3GHz Pentium 4 of equal core count and say that, just because the clock is ticking at the same speed, Intel isn't doing enough to improve performance. There would be a vast effective difference in just about every kind of workload, just due to architecture improvements.
This cap also has to do with physical limitations of signal propagation times and the limitation of memory as we know it today. Even if external memory wasn't the bottleneck annihilating the purpose of a faster clock, internal registers need a certain time to retain electricity. Silicon is only so speedy! More compact architecture (1nm dies would be nice, right?) could allow for quicker propagation times and thus faster clock speeds.
There's a reason for that clock speed stall and it certainly isn't for lack of trying. Billions have been thrown at the problem. It's just a fundamental limit of transistor design and heat dissipation. If it were possible to manufacture 8ghz cpus that could run reliably for years they'd be on the market.