Type of computer architecture
One approach to order PC designs is by the quantity of directions executed per clock.
Many registering machines read each guidance in turn and execute it (or put a ton of exertion into going about as though they do that, regardless of whether they do extravagant and faulty superscalar things inside). I call these machines "von Neumann" machines, since they all have a von Neumann bottleneck.
Such machines incorporate CISC, RISC, MISC, TTA and DSP models. These machines incorporate collecting machines, sales enlists and stacking machines. Different machines peruse and execute various directions without a moment's delay (VLIW, super-scalar), which break the constraint of one guidance for each clock, yet keep on hitting the von Neumann bottleneck on marginally more guidelines per clock. In any case, different machines are not constrained by the von Neumann bottleneck, as they load the entirety of their tasks once at power-up and afterward process the information moving along without any more guidelines.
Such non-Von-Neumann machines incorporate information stream designs.
Another approach to order PC structures is through the association (s) between the CPU and memory. A few machines have bound together memory, with the goal that a solitary location compares to a solitary area in memory, and when that memory is RAM, you can utilize that address to peruse and compose information, or burden that address into the program counter to run the code. I call these machines princeton machines. Different machines have a few separate memory spaces, so the program counter consistently alludes to "program memory" paying little heed to the location that is stacked, and typical peruses and composes consistently go to "information memory", which is a different area that for the most part contains diverse data in any event, when the bits of the information address end up being indistinguishable from the bits of the memory address of the program. Those machines are "unadulterated Harvard"
A few people utilize a tight meaning of "von Neumann machine" that does exclude Harvard machines. In the event that you are one of those individuals, what term would you use for the more broad idea of "a machine that has a von Neumann bottleneck", which incorporates machines from Harvard and Princeton, and avoids NON-VON?
Most installed frameworks utilize the Harvard design. A few CPUs are "unadulterated Harvard," which is maybe the least complex course of action for building equipment: the location transport to the read-just program memory is associated only to the program counter, in the same way as other Microchip PICmicros.
What's more, some changed Harvard machines additionally place constants in program memory, which can be perused with an uncommon "read consistent information from program memory" guidance (unique in relation to the "read information memory" guidance) . The product that sudden spikes in demand for the above kinds of Harvard machines can't change the memory of the program, which is viably ROM for that product.
Some installed frameworks are "self-programmable", commonly with streak memory program memory and an uncommon "streak memory eradicate square" guidance and a unique "streak memory compose square" guidance (not quite the same as the "guidance"). write to "typical" information memory, notwithstanding the "read information from program memory" guidance. A few more up to date Microchip PICmicros and Atmel AVR are self-programmable adjusted Harvard machines.
Another approach to classify CPUs is by their clock. Most PCs are synchronous, they have a solitary worldwide clock. A few CPUs are nonconcurrent (they don't have a clock), including ILLIAC I and ILLIAC II, which were at one time the world's quickest supercomputers.
One approach to order PC designs is by the quantity of directions executed per clock.
Many registering machines read each guidance in turn and execute it (or put a ton of exertion into going about as though they do that, regardless of whether they do extravagant and faulty superscalar things inside). I call these machines "von Neumann" machines, since they all have a von Neumann bottleneck.
Such machines incorporate CISC, RISC, MISC, TTA and DSP models. These machines incorporate collecting machines, sales enlists and stacking machines. Different machines peruse and execute various directions without a moment's delay (VLIW, super-scalar), which break the constraint of one guidance for each clock, yet keep on hitting the von Neumann bottleneck on marginally more guidelines per clock. In any case, different machines are not constrained by the von Neumann bottleneck, as they load the entirety of their tasks once at power-up and afterward process the information moving along without any more guidelines.
Such non-Von-Neumann machines incorporate information stream designs.
Another approach to order PC structures is through the association (s) between the CPU and memory. A few machines have bound together memory, with the goal that a solitary location compares to a solitary area in memory, and when that memory is RAM, you can utilize that address to peruse and compose information, or burden that address into the program counter to run the code. I call these machines princeton machines. Different machines have a few separate memory spaces, so the program counter consistently alludes to "program memory" paying little heed to the location that is stacked, and typical peruses and composes consistently go to "information memory", which is a different area that for the most part contains diverse data in any event, when the bits of the information address end up being indistinguishable from the bits of the memory address of the program. Those machines are "unadulterated Harvard"
A few people utilize a tight meaning of "von Neumann machine" that does exclude Harvard machines. In the event that you are one of those individuals, what term would you use for the more broad idea of "a machine that has a von Neumann bottleneck", which incorporates machines from Harvard and Princeton, and avoids NON-VON?
Most installed frameworks utilize the Harvard design. A few CPUs are "unadulterated Harvard," which is maybe the least complex course of action for building equipment: the location transport to the read-just program memory is associated only to the program counter, in the same way as other Microchip PICmicros.
What's more, some changed Harvard machines additionally place constants in program memory, which can be perused with an uncommon "read consistent information from program memory" guidance (unique in relation to the "read information memory" guidance) . The product that sudden spikes in demand for the above kinds of Harvard machines can't change the memory of the program, which is viably ROM for that product.
Some installed frameworks are "self-programmable", commonly with streak memory program memory and an uncommon "streak memory eradicate square" guidance and a unique "streak memory compose square" guidance (not quite the same as the "guidance"). write to "typical" information memory, notwithstanding the "read information from program memory" guidance. A few more up to date Microchip PICmicros and Atmel AVR are self-programmable adjusted Harvard machines.
Another approach to classify CPUs is by their clock. Most PCs are synchronous, they have a solitary worldwide clock. A few CPUs are nonconcurrent (they don't have a clock), including ILLIAC I and ILLIAC II, which were at one time the world's quickest supercomputers.
No comments:
Post a Comment