Are x86 CISC based processors doomed !?

[GoldenLag]

5 minutes ago, moderategamer said:

So yes although the chiplet method provides better speeds than two separate cpu's it's ultimately not that useful.

Its useful when you want more cores without risking yields. Which is the entire point of chiplets, active substrates and more. 

 

Every core doesnt neccesarly have to communicate with eachother if the software and sqedualler is designed to handle it. And in Serveruse that is not much of an issue due to custom sqedualler.

 

But ARM will run into the same issues as x86 if it tries to do all the things x86 does. Maybe at a lesser extent it will be limited, but it ultimatly has to overcome the same challanges. 

 

Im not saying a new architecture wont take over, but a lot of requirements needs to be met before that happens.

 

Apple is prpbably the first on to move to full arm because they make their own chips and only need to keep a very limited backwards compatibility. 

[moderategamer]

1 minute ago, GoldenLag said:

Its useful when you want more cores without risking yields. Which is the entire point of chiplets, active substrates and more. 

 

Every core doesnt neccesarly have to communicate with eachother if the software and sqedualler is designed to handle it. And in Serveruse that is not much of an issue due to custom sqedualler.

 

But ARM will run into the same issues as x86 if it tries to do all the things x86 does. Maybe at a lesser extent it will be limited, but it ultimatly has to overcome the same challanges. 

 

Im not saying a new architecture wont take over, but a lot of requirements needs to be met before that happens.

 

Apple is prpbably the first on to move to full arm because they make their own chips and only need to keep a very limited backwards compatibility. 

True I did make a point of stating that ARM chips have to play by the same rules just for the record I'm merely stating that ARM has less of a footprint limitation meaning it can physically have more cores without the need of trickery. It is useful for separate cores to communicate however but to what degree is debatable. I'm not saying AMD's solution is bad I know nothing about it and after all they are the chip designers so they obviously know more than me and they wouldn't do it if there was no benefit to it. However I do feel like it is kind of a hack to boast higher core counts when in actuall fact they have just put two cpu's on a single chip. 

[GoldenLag]

7 minutes ago, moderategamer said:

However I do feel like it is kind of a hack to boast higher core counts when in actuall fact they have just put two cpu's on a single chip.

You mean 2 or more chips in a single substrate. I mean it behaves like a single CPU. Its kinda like Intel making their coffelake architecture longer by adding 2 more cores, except AMD makes extea chips. (Intel dies get longer and longer the more cores they add on coffelake)

 

9 minutes ago, moderategamer said:

I'm merely stating that ARM has less of a footprint limitation meaning it can physically have more cores without the need of trickery.

Which depends entirely on whatbis included in the cores. For the moment the ARM cores are smaller. But with AVX units and possibly extra execution units they will get larger. Yes they are smaller, but dont think they wont get big when you start adding a lof of custom instruction sets and units

[Drak3]

4 hours ago, PrinceNorris said:

Are there any advantages of using x86 nowdays? We have x64 for pretty much anything not mobile, ARM for anything mobile.

x64 is an extension of x86.

[Sauron]

5 hours ago, PrinceNorris said:

Are there any advantages of using x86 nowdays? We have x64 for pretty much anything not mobile, ARM for anything mobile.

x64 is shorthand for x86-64 or amd64 - it's still an extension of the x86 instruction set. You are using x86 right now.

[Mira Yurizaki]

9 hours ago, moderategamer said:

It would seem that x86 chipsets are stagnating getting to the limits of what they can achieve through die shrinks and with the evermore popular parallel computing tasks it's looking like ARM RISC based chips could be the future with it's low power low transistor design. x86 instruction sets have become bloated and often complex instructions take more than 1 cycle to complete whilst ARM instructions often run in a single cycle. ARM chips are catching up with their x86 counterparts and with large companies like Microsoft and apple making a push for ARM does this spell doom for Intel and AMD x86 chips or will they prevail . It will take a complete over hall of current software to make such a change possible but it's looking likely that this task will be undertaken. 

 

Here is an interesting Video that gives some food for thought about this, what are your thoughts is it the end of an era of power hungry x86 CISC based computing or do we need to put away the tinfoil hats for when the aliens come?.

The current implementations of x86 compatible processors are really a RISC back-end with a CISC compatible front-end. Even people found when developing RISC that  if you used simple CISC instructions that performance would improve. But also note that this was back in the 80s, I'm pretty sure Intel, AMD, and other x86 stakeholders have found ways to mitigate much of the penalty for more complicated instructions.  Also the x86 architecture relied on various other front-end features to improve performance, like OOE, branch prediction, etc. Some of which are relatively new to ARM because implementing them in hardware up until then was prohibitively expensive power-wise.

 

Besides that, I don't think the instruction set architecture (ISA) itself is what makes a processor better or worse. Rather, it's the implementation of it. Ever wonder why Apple's SoCs dominate the performance field even though other ARM compatible processors can't touch it? It's because of how Apple implemented the architecture. Similarly, how AMD implemented x86 during the Athlon/Athlon XP days compared to Intel's Pentium 4 allowed them to make processors that were much more efficient.

 

So I'm led to believe if you know what you're doing, the complexities of a given ISA have minimal impact.

 

EDIT: I'd also recommend reading the top answer at https://stackoverflow.com/questions/2679882/why-is-x86-ugly-why-is-it-considered-inferior-when-compared-to-others

 

EDIT 2: Another thing to take note, this time from AnandTech (Emphasis added):

 

Quote

Back in the 1980s up to the 1990s, this became one of the major reasons why RISC was rapidly becoming dominant as CISC ISAs like x86 ended up creating CPUs that generally used more power and die area for the same performance. However, today ISA is basically irrelevant to the discussion due to a number of factors. The first is that beginning with the Intel Pentium Pro and AMD K5, x86 CPUs were really RISC CPU cores with microcode or some other logic to translate x86 CPU instructions to the internal RISC CPU instructions. The second is that decoding of these instructions has been increasingly optimized around only a few instructions that are commonly used by compilers, which makes the x86 ISA practically less complex than what the standard might suggest. The final change here has been that ARM and other RISC ISAs have gotten increasingly complex as well, as it became necessary to enable instructions that support floating point math, SIMD operations, CPU virtualization, and cryptography. As a result, the RISC/CISC distinction is mostly irrelevant when it comes to discussions of power efficiency and performance as microarchitecture is really the main factor at play now.

 

[Stefan Payne]

14 hours ago, moderategamer said:

It would seem that x86 chipsets are stagnating getting to the limits of what they can achieve

Yes, because lack of competition for the last 10 years or so.

Quote

it's looking like ARM RISC based chips could be the future with it's low power low transistor design.

Yes but not because of its "low power low transistor design", its because its open and can be licensed by anyone who can afford it.

And then 3rd partys can make their own processors.

Either based on the Standard design.

Or just implement the Instruction set and make a compatible processor with their own technology. 

For example AMD could modify Ryzen so that it runs 

 

Quote

x86 instruction sets have become bloated and often complex instructions take more than 1 cycle to complete whilst ARM instructions often run in a single cycle.

Yes, but isn't it the same with ARM??
And that some instructions aren't done in one cycle is for good reason sometimes...

AMD K6-2 has shown what happens if you "shorten" instructions...

 

14 hours ago, GoldenLag said:

Well. For them to actually be in trouble. ARM would need to be able to emulate x86 and AMD64 and beat traditional x86 CPUs. 

Otherwise there wont be any adoption. 

No, software needs to be adopted to the instruction set.

And ARM is already dominating.

And from what you hear, Apple is very likely to move away from Intel towards other things because there are more sources for ARM Chips than x86 because of the Patent Situation.

 

Apple can du their stuff themselves, buy a Samsung Chip, buy a Qualcom Chip or let AMD do one for them.

[bloodthirster]

Reading this thread hurts my brain (other than that post by @Mira Yurizaki).

[moderategamer]

1 hour ago, bloodthirster said:

Reading this thread hurts my brain (other than that post by @Mira Yurizaki).

Feel free to go somewhere else...

[straight_stewie]

13 hours ago, Stefan Payne said:

And from what you hear, Apple is very likely to move away from Intel towards other things because there are more sources for ARM Chips than x86 because of the Patent Situation. 

^^ Just to add to this, all throughout Apples history they've been willing to use vastly different architectures than their competitors.

The Mac came out with a Motorala 68000, 3 years after the IBM PC had been released with an Intel 8088.

After that, Apple used PowerPC all the way up until 2006 or 2007, when they started including Intel processors in some machines.

It's important to remember that this relatively frequent changing of core hardware is only made possible by their tight control of the Apple ecosystem: Whereas PCs have hardware manufacturers and software manufacturers that all have to be on the same page, Apple is a "one stop shop", responsible both for the machine architecture and the hardware. This can afford more flexibility when it comes to what hardware their machines use.

[Mira Yurizaki]

2 hours ago, straight_stewie said:

The Mac came out with a Motorala 68000, 3 years after the IBM PC had been released with an Intel 8088.

After that, Apple used PowerPC all the way up until 2006 or 2007, when they started including Intel processors in some machines.

It's important to remember that this relatively frequent changing of core hardware is only made possible by their tight control of the Apple ecosystem: Whereas PCs have hardware manufacturers and software manufacturers that all have to be on the same page, Apple is a "one stop shop", responsible both for the machine architecture and the hardware. This can afford more flexibility when it comes to what hardware their machines use.

The thing with going from PowerPC to x86 is that Apple was part of the PowerPC's development, so they intimately knew what was going on with that hardware. On top of that, because they helped develop it, I'm pretty sure that means they had access to all of the secret sauce that made the processor tick and neither IBM nor Motorola could say anything about Apple using this knowledge to make the transition easier.

 

I don't think transitioning from x86 to ARM will go that smoothly, especially considering that Intel tried to sue Microsoft (and maybe ARM?) for trying to write an x86 emulator on ARM on the grounds that Microsoft was leveraging Intel's trade secrets or whatever. Pretty sure if Apple tried the same thing, Intel would also try to sue them. Though if they make their app ecosystem almost completely reliant on the App Store and demand there be iOS versions of the app as well, then they could easily ditch x86.

[BuckGup]

Judging by my sources working in the HPC industry basically 1 out of 3 computers are now ARM based. It's cheaper, faster, and can be specifically made for their use case. Also EPYC is being used a lot more than I expected. Intel needs to capitalize on the GPU market next year to stay afloat.