Furthermore, Cameron's chip will not have the raw clockrate that John Henry's servers have.
Are you kidding? Her chip is from 20 years in the
future. Look how far we've come in 20 years; a freakin' PS3 is over 1000 times more powerful than a Cray-2 super computer from 1989. And that wasn't even built by a super advanced AI that has no problem tackling things like liquid metal machines and time travel.
For me the biggest problem with the scene was that there's no way to get the chip from Cam to John Henry without both of them having to be shut down at the same time.
The PS3's Cell architecture both a highly parallel architecture and a highly specialized architecture. It includes a "PPE", which is essentially a hyperthreading single core general purpose processor and eight "SPEs", RISC floating point vector processors, only six of which are enabled at any one time, two are redundant failsafes. In terms of clockrate, it is 3.2 GHz, which is pretty much the hard limit on reliable commercial processors these days, though one can overclock much high using extreme cooling methods, these are unreliable.
The SPEs are great for certain types of operations, and crappy at others, which is why you have to very specifically write your programs to spawn the correct threads on the correct cores at the correct times, and it's best to do so in assembly, which is one of the reasons writing for the PS3 is fairly difficult.
Currently, parallel processing is the prefered method of increasing processing power. New chips have more and more cores, and they're more efficient, but they don't have any high clockrates than the previous two generations of chips.
The issue is Ohms Law, and certain related facts. Current = Voltage/Resistence.
Resistence = (Length*resistivitity)/cross sextional area
Power = Current*Voltage
Waste Heat = Power in - Power out
As manufacturing processes become smaller, more transistors can be fit onto a single die, and archetectures can be made more efficient. However, smaller transistors means thinner transistors, and thinner transistors have greater resistence.
In order to perform at higher clockrates, a chip must be supplied with more current, which means more power, which means more waste heat. As the transistors become thinner and thinner, the amount of current they can withstand without burning becomes less and less. But, more imporantly, with all of those transistors packed tightly together, the more power that is fed into them the more waste heat will be produced, and if this heat is not removed in a timely manner the whole chip will melt. Which is why very fast chips have giant heatsinks with giant fans, and overclocked chips tend to to use absurdly complicated and expensive cooling methods.
Terminators don't have giant fans in their heads, that I'm aware of. And their chips are very small but contain massive amount of storage as well as processors. For this reason, I must assume that they use a very small fabrication process and rely on architecture efficiency and simultaneous multi-threading instead of raw clockspeed. This would allow them to ourperform any modern hardware in terms of operations per second, while still having a realatively low clockspeed. And this is great for any native program running on the hardware. The problem lies in emulation. A quick and dirty hardware emulator would not be able to take advantage of greater archetecture efficiency to simulate greater clockspeed. A very well written one might be, but I'm not sure about that. I've never actually seen someone try to emulate a procesor with a very high clockrate a processor that can complete more Ops but which has a lower clockrate. There isn't much call for it these days.