x86 is an architecture, so it basically says which words the processor should understand and what should it do after it understood them. So, how can different processors be different if they have the same number of cores and same speed? I.e. if I have an i5, quad core, no strange Technologies like HT, and an equally quad core from AMD with the same clock speed, what makes the difference in performance? Smaller transistors may make it consume less power, but it should be equal, right? or is it like with logic gates, that you may have the same result with two different approaches?
x86 is an instruction set, not an architecture.
Architectures can be x86 but x86 isn't an architecture by itself
>>55388514
X86 is an instruction set.
AMD and intel both make hardware which implement these instructions. A "core" is not one thing, they are all vastly, vastly different.
>>55388514
Because one instruction does not necessarily take one cycle to execute.
A single SIMD instruction can take anywhere from 2 to 8 cycles to fully execute.
Each subsequent generation of processors generally push down the number of cycles required to execute fat instructions like SIMD.
>>55388555
>>55388552
>>55388538
Oh, ok, I see. But then how can Intel stop you from making an x86 compatible processor?
>>55388587
Licensing I imagine. I think it is all publicly documented though
>>55388598
couldn't a company just make a processor that has the same instructions as an x86 just renamed some way? instruction X on x86 becomes Y in this processor, but still does the same thing. It's easy then to set up a converter
>>55388587
Patents.
>>55388587
Licensing of the instruction set.
It's legally murky whether they can actually stop competitors from implementing compatible CPU based on publicly available documentation, but that hasn't stopped Intel from throwing billions of dollars worth of IP lawyers to bully competitors into leaving the x86 market.
>>55388514
The way these instructions are executed has nothing to do with the instructions themselves. Look up pipelining for a naive example of a strategy that may be different across different processors.
>>55388651
yes, I've noticed that. still, >>55388627?
>>55388514
x86 is an ISA - Instruction Set Architecture
How an ISA is implemented is defined by the micro-architecture\
>>55388587
It's a grey area.
IIRC nvidias project denver pretty much transpiles x86 instructions on the fly into ARM instructions and executes them this way. But I think they somehow failed somehow anyways.
On another level AMD does something similar with CUDA. They basically have a compiler which translates CUDA into native C++ which in turn gets transpiled into opencl. This way they can circumvent any nvidia patent infringements.
>>55388859
ARM and x86 are different instruction sets, aren't they? there's no X instruction on ARM for Y instruction on x86, right?
>>55388916
ARM instructions are quite different from that of any other architecture in that jump and branch can happen from any instruction.
>>55388916
yes. I was sort of talking out of my ass but I just looked it up.
Denver apparently transpiles ARM and x86 instructions into one common instruction set which the gets executed. But they still needed an x86 license and Intel said nope. That's probably why denver is kill aside from HPC and automobile sectors.
>>55389039
I think someone should just do >>55388627, it would be legal imo
