Is x86 on its way out?

[jonneymendoza]

This is because ARM chips have become so much better in the last 10 years while x86 has been more or less stagnant with only minor improvements. So the operating systems, software and use cases have been updated to use the extra power, rather than becoming stagnant.

A core in Apple A13 is over 100 times faster than a core in Apple A4 (released in 2010).
A core in i9 10900K is almost 2 times faster than a core in Core i7 920 (released in 2008). Majority of the improvement also comes from clock speed improvements (2.93GHz vs 5.3GHz Turbo speeds).

If we've had a 100x performance improvement for x86 in the last 10 years, every OS, game, software and use-case would have been updated to use the extra processing power. In that case, your 10-year-old x86 chip would have seriously struggled as well. Stagnation is not a sign of quality or longevity, it's a sign of mediocracy, and that's what Intel has been in the last 10 years.

Intel are not the only cpu. Amd has made huge strides over the last few years

 

[jonneymendoza]

This is because ARM chips have become so much better in the last 10 years while x86 has been more or less stagnant with only minor improvements. So the operating systems, software and use cases have been updated to use the extra power, rather than becoming stagnant.

A core in Apple A13 is over 100 times faster than a core in Apple A4 (released in 2010).
A core in i9 10900K is almost 2 times faster than a core in Core i7 920 (released in 2008). Majority of the improvement also comes from clock speed improvements (2.93GHz vs 5.3GHz Turbo speeds).

If we've had a 100x performance improvement for x86 in the last 10 years, every OS, game, software and use-case would have been updated to use the extra processing power. In that case, your 10-year-old x86 chip would have seriously struggled as well. Stagnation is not a sign of quality or longevity, it's a sign of mediocracy, and that's what Intel has been in the last 10 years.

AMD says no

 

[HACO]

Intel are not the only cpu. Amd has made huge strides over the last few years

AMD has made a 2x improvement in the last few years, bringing it on par with Intel right now. Nothing in the scale of the 100x that we've seen in ARM. The scale here is so different that "oh after 10 years of doing nothing, AMD has a 20% IPC improvement this generation" and "oh Zen 3 will have a 15% IPC improvement" isn't going to make a difference.

Like it or not, x86 has been stagnant in the last 10 years compared to ARM. You could argue it wasn't 2x improvement for x86, but 2.5x if you include AMD. The point still stands.

AMD says no

If you seriously think the improvements in AMD in the last 10 years is on par with ARM's, you're living in a fantasy world.

The fact that you think AMD has "made huge strides", aka made 20% IPC improvements in 5 years which is usually the performance improvements that ARM processors usually have in 1 year, is evidence that x86 is stagnant.

I would love for x86 to be vibrant again, to see 20% year-on-year IPC improvements. It just hasn't happened in the last 10 years.

 

[deusexlumina]

I would love for x86 to be vibrant again, to see 20% year-on-year IPC improvements. It just hasn't happened in the last 10 years.

But there was that level of improvement, it just wasn't this past decade as you said.

1993, intel with the pentium at 66 MHz,

2003, amd with the fx-51. 2.2 Ghz or intel with the p4 extreme at around 3 GHz iirc.

Just randomly picked these from a list, but willing to wager that's close to x100 improvement in compute power in 10 years.

The improvements in ARM processors have been very impressive but there's no reason to believe it will not stagnate much like x86 or indeed many other forms of technology after they experience rapid growth cycles.

 

[jigger]

AMD has made a 2x improvement in the last few years, bringing it on par with Intel right now. Nothing in the scale of the 100x that we've seen in ARM. The scale here is so different that "oh after 10 years of doing nothing, AMD has a 20% IPC improvement this generation" and "oh Zen 3 will have a 15% IPC improvement" isn't going to make a difference.

Like it or not, x86 has been stagnant in the last 10 years compared to ARM. You could argue it wasn't 2x improvement for x86, but 2.5x if you include AMD. The point still stands.



If you seriously think the improvements in AMD in the last 10 years is on par with ARM's, you're living in a fantasy world.

The fact that you think AMD has "made huge strides", aka made 20% IPC improvements in 5 years which is usually the performance improvements that ARM processors usually have in 1 year, is evidence that x86 is stagnant.

I would love for x86 to be vibrant again, to see 20% year-on-year IPC improvements. It just hasn't happened in the last 10 years.

I suppose it depends how you look at it. How much faster is an EPYC.2 system over an Opteron?

 

[HACO]

But there was that level of improvement, it just wasn't this past decade as you said.

1993, intel with the pentium at 66 MHz,

2003, amd with the fx-51. 2.2 Ghz or intel with the p4 extreme at around 3 GHz iirc.

Just randomly picked these from a list, but willing to wager that's close to x100 improvement in compute power in 10 years.

The improvements in ARM processors have been very impressive but there's no reason to believe it will not stagnate much like x86 or indeed many other forms of technology after they experience rapid growth cycles.

Please follow the context of the discussion, I was responding to a user who was claiming that their 10-year-old x86 chip is still very capable while their 10-year-old ARM chip wasn't. I explained the reason.

Now about what you said, I never really claimed otherwise and I don't disagree. I'm pretty sure the ARM improvements will slow down as well. But the thing is, 10 years ago x86 was orders of magnitude ahead of ARM when it came to performance. It's not like that anymore. The gap has pretty much disappeared and there is competition now.

Who will have the fastest core in 5 years? 5 years ago the answer to that was obvious. Right now it's "I don't know" and that's all I'm saying.

I suppose it depends how you look at it. How much faster is an EPYC.2 system over an Opteron?

I'm pretty sure a lot faster. Mostly due to SSDs and other technologies. The IPC improvement across those 10 years have not been impressive (75-80%).

That's what happens when there's no competition and Intel sits on their ass doing nothing for more than 5 years, and AMD is being treated like some kind of paragon of progress for just showing up to work. It's great that AMD is back, but no need to oversell it, the x86 scene over the last 10 years has been abysmal. It might get better now that there's competition both between Intel and AMD, and from ARM, I certainly hope so.

 

[jigger]

Please follow the context of the discussion, I was responding to a user who was claiming that their 10-year-old x86 chip is still very capable while their 10-year-old ARM chip wasn't. I explained the reason.

Now about what you said, I never really claimed otherwise and I don't disagree. I'm pretty sure the ARM improvements will slow down as well. But the thing is, 10 years ago x86 was orders of magnitude ahead of ARM when it came to performance. It's not like that anymore. The gap has pretty much disappeared and there is competition now.

Who will have the fastest core in 5 years? 5 years ago the answer to that was obvious. Right now it's "I don't know" and that's all I'm saying.



I'm pretty sure a lot faster. Mostly due to SSDs and other technologies. The IPC improvement across those 10 years have not been impressive (75-80%).

That's what happens when there's no competition and Intel sits on their ass doing nothing for more than 5 years, and AMD is being treated like some kind of paragon of progress for just showing up to work. It's great that AMD is back, but no need to oversell it, the x86 scene over the last 10 years has been abysmal.

I see. So how would you to define progress? SSD’s have nothing to do with X86.

 

[HACO]

I see. So how would you to define progress? SSD’s have nothing to do with X86.

Yeah, so your point about "how much faster a modern system is compared to an old one" isn't really about x86 when most of the speed improvements come from other things like SSDs. Like I said, IPC improvements of 75-80% over 10 years is no reason to celebrate.

I define progress as getting better at doing what you do. AMD has done this in the last few years while Intel has not. Both of them though have massively dropped the ball in the last 10 years, allowing an alternative architecture which used to be 50-100x slower to catch up and become competitive.

 

[jigger]

Yeah, so your point about "how much faster a modern system is compared to an old one" isn't really about x86 when most of the speed improvements come from other things like SSDs. Like I said, IPC improvements of 75-80% over 10 years is no reason to celebrate.

I define progress as getting better at doing what you do. AMD has done this in the last few years while Intel has not. Both of them though have massively dropped the ball in the last 10 years, allowing an alternative architecture which used to be 50-100x slower to catch up and become competitive.

You miss the point. You seem to only consider IPC performance and that too relevant to metrics like performer per rack.

Intel was cheating for years, but that is another topic.

 

[HACO]

You miss the point. You seem to only consider IPC performance and that too relevant to metrics like performer per rack.

I'm aware they've become better at slamming a ton of cores into server chips, almost entirely because the processing node shrinks allows it to be cooled down and power delivery to work. Again, not x86 architecture related, but silicon related.

Architecture in the end, is all about IPC.

 

[jigger]

I'm aware they've become better at slamming a ton of cores into server chips, almost entirely because the processing node shrinks allows it to be cooled down and power delivery to work. Again, not x86 architecture related, but silicon related.

Architecture in the end, is all about IPC.

I disagree. IPC is one aspect of the industry. Other metrics are as, or more important.

 

[HACO]

I disagree. IPC is one aspect of the industry. Other metrics are as, or more important.

Industry, yes of course. CPU architecture? Nope.

 

[jigger]

Industry, yes of course. CPU architecture? Nope.

It’s the industry that must be won over before anything can change.

 

[HACO]

It’s the industry that must be won over before anything can change.

That's philosophy now and off topic

 

[jigger]

That's philosophy now and off topic

Maybe, but it’s a pretty broad topic.

 

[Cyber-Mav]

Arm may have hit its limit. Usually each generation they get 20 percent ipc gain with big drop in power consumption. But going from a77 to a78 seems have the lowest ipc gain from all their generations.
https://en.wikichip.org/wiki/arm_holdings/microarchitectures/cortex-a78

Link above says 7 percent ipc gain. Could be they have hit the wall.

 

[pc-guy]

hang on, isn't ARM processors need a load of sub-processors to do what the equivalent x86 processors do?

in a very simple way isnt ARM a super-duper adding machine, but x86 is a complex calculator able to do all sorts of maths functions in comparison.

the best packaged CPU surely is a combination of what ARM cpus are good at (high IPC) and what x86 are good at (complex task performances) - dont ask me how they can bring the two together as they talk different set of language. but if it can be done, that would be good right

 

[HACO]

Arm may have hit its limit. Usually each generation they get 20 percent ipc gain with big drop in power consumption. But going from a77 to a78 seems have the lowest ipc gain from all their generations.
https://en.wikichip.org/wiki/arm_holdings/microarchitectures/cortex-a78

Link above says 7 percent ipc gain. Could be they have hit the wall.

That’s because A78 isn’t the high performance design anymore, that’s Cortex-X1 which is 30% faster than A77.

This is change in branding. More here: https://www.anandtech.com/show/15813/arm-cortex-a78-cortex-x1-cpu-ip-diverging

hang on, isn't ARM processors need a load of sub-processors to do what the equivalent x86 processors do?

in a very simple way isnt ARM a super-duper adding machine, but x86 is a complex calculator able to do all sorts of maths functions in comparison.

This whole “ARM is RISC but x86 is CISC” hasn’t been true for over a decade. All modern processors are RISC and all of them use microcode to translate their input ISA into their own internal instruction sets. This includes Intel, AMD, Apple, Samsung, ARM and others.

Outside of very niche use cases, there really isn’t a generic advantage to x86 or ARM. They can both be utilised for very complex tasks.

the best packaged CPU surely is a combination of what ARM cpus are good at (high IPC) and what x86 are good at (complex task performances) - dont ask me how they can bring the two together as they talk different set of language. but if it can be done, that would be good right

There’s no reason that this can’t work (other than licensing), but also no reason that this would be any useful. You can design ARM or x86 cores that are tailored to specific use cases and use them in heterogenous CPU designs. We already have such products in Big.Little ARM architectures (which will come to x86 as well).