I've seen a number of sources that talk as though computing power will just keep increasing. This seems at odds with the hardware.
Back in 1975, a processor like the 6502 had an 8 bit data bus. This doubled until it reached 64 bits in 2003. It hasn't gone up since then. 64 bits seems to be an optimal size. So, no further increase here.
A 6502 ran at 1 MHz. This doubled every 18 months until it reached 1.5 GHz in 2001 and then the pace slowed. Over the next two years it increased 60% to 2.4 GHz in 2003. Over the next three years, this increased 50% to 3.6 GHz. However from 2006 to 2011, the increase was only 600 MHz to 4.2 GHz and there has been no appreciable increase since then. So, I'm not seeing an increase here either.
A 6502 had some limited pipelining. The depth has steadily increased and we've now seemed to have topped out.
You used to only be able to see things like longer pipelines, vector processing, super scalar and multi-processing on supercomputers like the Cray XMP. These have all now been incorporated into desktop processors. And, we've already discarded things like RISC and VLIW. There isn't a lot left here. The only thing left seems to be incorporating GPGPU processors.
And then you have process size. If memory serves, a 6502 used a 3,500 nm process. We are at 22n m now. The first processor at 22 nm came out in 2011. Intel is still trying to move down to 14 nm. It looks like everyone will eventually move down to 16 or 14 nm, but it still isn't clear if there is any advantage to the 11 or 10 nm size. So, this seems about used up too.
Also, it looks like the primary method of increasing memory speed since EDO has been by interleaving the banks. This method obviously cannot continue since it has the same latency disadvantage as pipelining.
As far as I can tell, all of these things are true yet tech publications and journalists don't seem to ever mention it. I suppose this must seem like the kiss of death to a magazine like Wired which was built on steadily claiming that the next new, hot thing was just around the corner.
Back in 1975, a processor like the 6502 had an 8 bit data bus. This doubled until it reached 64 bits in 2003. It hasn't gone up since then. 64 bits seems to be an optimal size. So, no further increase here.
A 6502 ran at 1 MHz. This doubled every 18 months until it reached 1.5 GHz in 2001 and then the pace slowed. Over the next two years it increased 60% to 2.4 GHz in 2003. Over the next three years, this increased 50% to 3.6 GHz. However from 2006 to 2011, the increase was only 600 MHz to 4.2 GHz and there has been no appreciable increase since then. So, I'm not seeing an increase here either.
A 6502 had some limited pipelining. The depth has steadily increased and we've now seemed to have topped out.
You used to only be able to see things like longer pipelines, vector processing, super scalar and multi-processing on supercomputers like the Cray XMP. These have all now been incorporated into desktop processors. And, we've already discarded things like RISC and VLIW. There isn't a lot left here. The only thing left seems to be incorporating GPGPU processors.
And then you have process size. If memory serves, a 6502 used a 3,500 nm process. We are at 22n m now. The first processor at 22 nm came out in 2011. Intel is still trying to move down to 14 nm. It looks like everyone will eventually move down to 16 or 14 nm, but it still isn't clear if there is any advantage to the 11 or 10 nm size. So, this seems about used up too.
Also, it looks like the primary method of increasing memory speed since EDO has been by interleaving the banks. This method obviously cannot continue since it has the same latency disadvantage as pipelining.
As far as I can tell, all of these things are true yet tech publications and journalists don't seem to ever mention it. I suppose this must seem like the kiss of death to a magazine like Wired which was built on steadily claiming that the next new, hot thing was just around the corner.
via International Skeptics Forum http://ift.tt/2jmtfLl
Aucun commentaire:
Enregistrer un commentaire