If I’m reading this correctly, the reason the distinction doesn’t matter any more is because computers have gotten so fast and chips/features so small that what used to be defining features of...
If I’m reading this correctly, the reason the distinction doesn’t matter any more is because computers have gotten so fast and chips/features so small that what used to be defining features of each architecture are now just tiny pieces.
The bigger deal seems to be how each individual chip designer (Apple, Qualcomm, Intel, AMD, etc.) design other parts of the chip and how they keep information flowing to and from the processor.
I recall awhile back an article about how modern x86 is so far removed from the original that the x86 compatability is actually holding chip capabilities back. Open source everything and end this...
I recall awhile back an article about how modern x86 is so far removed from the original that the x86 compatability is actually holding chip capabilities back.
Open source everything and end this architecture dependent madness.
instruction sets on modern CPUs are kind of weird; it almost feels like assembly programming is not only outdated, but counterproductive; there's hardware put into CPUs to take apart your bad...
instruction sets on modern CPUs are kind of weird; it almost feels like assembly programming is not only outdated, but counterproductive; there's hardware put into CPUs to take apart your bad instructions and tear them into micro-instructions so that it can run more of them at the same time. Modern high performance CPUs have transcended the RISC/CISC paradigm and are essentially VLIW machines in everything but name. They're practically DSPs at this point, and that's before you get into multicore design.
I did a bit of high-performance programming recently, and it got me thinking: the predictions I had to make about how the compiler would transform the code are kind of like the predictions I had...
I did a bit of high-performance programming recently, and it got me thinking: the predictions I had to make about how the compiler would transform the code are kind of like the predictions I had to make about how the CPU would run it.
Also, when I started putting the computation on multiple cores, I kind of had to imagine the programming language and microarchitecture levels at the same time. 'Cause of cache magic.
A cynical take: The article says that microarchitecture is the only meaningful difference between modern CPUs. The binary interface doesn't matter. So there's actually no reason to use anything...
A cynical take: The article says that microarchitecture is the only meaningful difference between modern CPUs. The binary interface doesn't matter. So there's actually no reason to use anything except an open instruction set.
The article claims that for performance the ISA doesn’t matter. For backwards compatibility, the ISA matters a lot. You don’t get higher performance for a binary that doesn’t run. Apple is...
The article claims that for performance the ISA doesn’t matter. For backwards compatibility, the ISA matters a lot. You don’t get higher performance for a binary that doesn’t run.
Apple is managing to switch ISA’s for the third time now (to PowerPC, Intel, and now ARM) so it can be done, but it does take a major effort and this isn’t transparent for customers, even with emulation.
There is a lot more portable software out there these days though, including all web applications.
Are there x86 processors similar to the ARM Cortex-M ones? They use an instruction set that's a subset of larger ARM chips. Maybe that's a sign that the ARM ISA has a better base for today's...
Are there x86 processors similar to the ARM Cortex-M ones?
They use an instruction set that's a subset of larger ARM chips. Maybe that's a sign that the ARM ISA has a better base for today's requirements than a similarly small x86.
The Cortex-M is a microcontroller. How do you define "today's requirements"? Microcontrollers are much simpler ICs, so this is an area where a simple risc chip design makes sense. Nothing in the...
The Cortex-M is a microcontroller. How do you define "today's requirements"?
Microcontrollers are much simpler ICs, so this is an area where a simple risc chip design makes sense. Nothing in the article contradicts that; almost anywhere you see it discussing the size requirements for cutting edge processor design such as prediction and instruction decoding caches, these are not found on any microcontrollers to the best of my knowledge.
So a microcontroller is an example of extreme size/cost optimizing and doesn't translate to a "better base" for advanced processors.
Not necessarily. That stripped down instruction set for the embedded space doesn't scale. Hence the full fat ISA. The article is about competing for modern performance considerations, and even...
Not necessarily. That stripped down instruction set for the embedded space doesn't scale. Hence the full fat ISA. The article is about competing for modern performance considerations, and even demonstrates x86 performance at low power. Microcontrollers are pretty much a race to the bottom in cost, increasing in performance largely due to process advancements imo.
Edit: although, sure, I'd personally consider RISC a nicer design than CISC, but the article details how even RISC designs aren't as simple and clean at the high end any more. There is also a quote about how VLIW has been considered "messy", but doesn't necessarily mean less efficient.
If I’m reading this correctly, the reason the distinction doesn’t matter any more is because computers have gotten so fast and chips/features so small that what used to be defining features of each architecture are now just tiny pieces.
The bigger deal seems to be how each individual chip designer (Apple, Qualcomm, Intel, AMD, etc.) design other parts of the chip and how they keep information flowing to and from the processor.
I recall awhile back an article about how modern x86 is so far removed from the original that the x86 compatability is actually holding chip capabilities back.
Open source everything and end this architecture dependent madness.
instruction sets on modern CPUs are kind of weird; it almost feels like assembly programming is not only outdated, but counterproductive; there's hardware put into CPUs to take apart your bad instructions and tear them into micro-instructions so that it can run more of them at the same time. Modern high performance CPUs have transcended the RISC/CISC paradigm and are essentially VLIW machines in everything but name. They're practically DSPs at this point, and that's before you get into multicore design.
I did a bit of high-performance programming recently, and it got me thinking: the predictions I had to make about how the compiler would transform the code are kind of like the predictions I had to make about how the CPU would run it.
Also, when I started putting the computation on multiple cores, I kind of had to imagine the programming language and microarchitecture levels at the same time. 'Cause of cache magic.
It seems like the article is saying that backward compatibility is fairly cheap and doesn't hold things back much?
A cynical take: The article says that microarchitecture is the only meaningful difference between modern CPUs. The binary interface doesn't matter. So there's actually no reason to use anything except an open instruction set.
The article claims that for performance the ISA doesn’t matter. For backwards compatibility, the ISA matters a lot. You don’t get higher performance for a binary that doesn’t run.
Apple is managing to switch ISA’s for the third time now (to PowerPC, Intel, and now ARM) so it can be done, but it does take a major effort and this isn’t transparent for customers, even with emulation.
There is a lot more portable software out there these days though, including all web applications.
NoRather, it is arguing that the talent of the designers is what matters.Are there x86 processors similar to the ARM Cortex-M ones?
They use an instruction set that's a subset of larger ARM chips. Maybe that's a sign that the ARM ISA has a better base for today's requirements than a similarly small x86.
The Cortex-M is a microcontroller. How do you define "today's requirements"?
Microcontrollers are much simpler ICs, so this is an area where a simple risc chip design makes sense. Nothing in the article contradicts that; almost anywhere you see it discussing the size requirements for cutting edge processor design such as prediction and instruction decoding caches, these are not found on any microcontrollers to the best of my knowledge.
So a microcontroller is an example of extreme size/cost optimizing and doesn't translate to a "better base" for advanced processors.
Right, so doesn't the fact that the same ISA works for both powerful and low-power chips mean that it's a nicer and more flexible design?
Not necessarily. That stripped down instruction set for the embedded space doesn't scale. Hence the full fat ISA. The article is about competing for modern performance considerations, and even demonstrates x86 performance at low power. Microcontrollers are pretty much a race to the bottom in cost, increasing in performance largely due to process advancements imo.
Edit: although, sure, I'd personally consider RISC a nicer design than CISC, but the article details how even RISC designs aren't as simple and clean at the high end any more. There is also a quote about how VLIW has been considered "messy", but doesn't necessarily mean less efficient.
Huh, yeah, I guess Thumb alone is a better measure of the ISA capabilities at that level.