Intel proposes x86-S, a redux of the x86 architecture

[Dr. Dro]

I haven't seen a discussion on this and it's something that caught my eye.

Intel published a whitepaper (links in the bottom of this post) which proposes the removal of certain traditional x86 architecture features and registers that date back to the original 8086 which have remained present for backwards compatibility purposes, such as 16-bit addressing and 32-bit protected mode support, ring 1 and 2 support, enforcing X2APIC - and certain changes such as making the CPU boot directly to 64-bit state after a reset command (eliminating the hoops the processor goes through during the initialization process), as well as replacing some of these features with a potentially much slower backwards compatibility system implemented through virtualization capabilities which would permit legacy operating systems designed with the current model in mind to remain functional.

Their justification is that by simplifying the ISA, resources could be devoted to more important features which are relevant to modern computers. I'm not a programmer, but this seems like a pretty substantial change (that has been a long time coming), as I understand many of these limitations already applied to some degree on 64-bit versions of Windows as it is (such as the removal of 16-bit support), but I would be interested in the thoughts of folks smarter than I am. My personal concerns, for example, how would this affect old video games that we boot in DOS mode? How would this affect hypervisors, boot times, security, reliability, etc. - I thought it was a pretty interesting read that had me thinking.



Envisioning a Simplified Intel Architecture

Direct link to the PDF

 

[Camm]

The biggest issue here would be more running directly on hardware older OS's & Software. Those calls could be moved to software (at likely pretty minimal penalty) with involvement from Microsoft & other OS communities so on the face of Intel's claims I think the proposal has merit.

 

[Dr. Dro]

The biggest issue here would be more running directly on hardware older OS's & Software. Those calls could be moved to software (at likely pretty minimal penalty) with involvement from Microsoft & other OS communities so on the face of Intel's claims I think the proposal has merit.

From what I understood, this would only affect operating systems and potentially old software that runs close to the metal (such as DOS games), from an user and even developer perspective there would be no change from how things work on 64-bit operating systems today, and a way to keep existing operating systems would be at least initially, if with some drawbacks, provided.

I've always known that modern PCs shared a lot in common with the very first ones, but I had no idea that so many of these ancient things were still fully implemented in advanced chips like these hybrid architecture Lakes we have today. From what I see, I think it's a good proposal as well.

 

[AusWolf]

Before I vote, I need to know if 16 and 32-bit games would run that way or not.

If software compatibility isn't an issue, then my vote is yes - eliminating unnecessary die real estate is a good idea.

If software compatibility is an issue, then my vote is no - I love old games, and I'm not willing to give them up, or have separate retro builds just to play them.

 

[Dr. Dro]

Before I vote, I need to know if 16 and 32-bit games would run that way or not.

If software compatibility isn't an issue, then my vote is yes - eliminating unnecessary die real estate is a good idea.

If software compatibility is an issue, then my vote is no - I love old games, and I'm not willing to give them up, or have separate retro builds just to play them.

The idea that I got is that anything that currently runs under 64-bit Windows would be unchanged under this new model, and that OS developers would be responsible for the adjustments required to support the new model, without changes to existing software that runs within the OS - but with the removal of all of these legacy features, 32-bit and older operating systems would need to be virtualized/emulated and would cease to be directly compatible as the CPU would operate exclusively in a new proposed 64-bit mode.

Accurate, full emulation of vintage PC hardware is very slow - you need an IPC brute of a CPU to run a 300 MHz Pentium II at full speed on PCem or 86Box.

 

[AusWolf]

The idea that I got is that anything that currently runs under 64-bit Windows would be unchanged under this new model, and that OS developers would be responsible for the adjustments required to support the new model, without changes to existing software that runs within the OS - but with the removal of all of these legacy features, 32-bit and older operating systems would need to be virtualized/emulated and would cease to be directly compatible as the CPU would operate exclusively in a new proposed 64-bit mode.

Accurate, full emulation of vintage PC hardware is very slow - you need an IPC brute of a CPU to get it to run a 300 MHz Pentium II at full speed on PCem or 86Box.

Ah, so if compatibility is solved through the OS right now, it wouldn't be any different with the new model, either. It's only OS compatibility that's affceted. Then my vote is yes.

I mean, running a 32-bit OS with 4 GB RAM support is already a pretty stupid idea these days. There's no need to support them any further.

 

[sLowEnd]

It's better in the long run if legacy baggage that is no longer needed is reduced

 

[Dr. Dro]

It's better in the long run if legacy baggage that is no longer needed is reduced

I agree.

Ah, so if compatibility is solved through the OS right now, it wouldn't be any different with the new model, either. It's only OS compatibility that's affceted. Then my vote is yes.

I mean, running a 32-bit OS with 4 GB RAM support is already a pretty stupid idea these days. There's no need to support them any further.

Yeah, and by the time a processor with this ISA rolls out, it would be a particularly niche case. For Intel to publish this asking for feedback, it's probably something they're accounting at the design stage of a processor that is still in the far future.

 

[BoboOOZ]

I haven't seen a discussion on this and it's something that caught my eye.

Intel published a whitepaper (links in the bottom of this post) which proposes the removal of certain traditional x86 architecture features and registers that date back to the original 8086 which have remained present for backwards compatibility purposes, such as 16-bit addressing and 32-bit protected mode support, ring 1 and 2 support, enforcing X2APIC - and certain changes such as making the CPU boot directly to 64-bit state after a reset command (eliminating the hoops the processor goes through during the initialization process), as well as replacing some of these features with a potentially much slower backwards compatibility system implemented through virtualization capabilities which would permit legacy operating systems designed with the current model in mind to remain functional.

Their justification is that by simplifying the ISA, resources could be devoted to more important features which are relevant to modern computers. I'm not a programmer, but this seems like a pretty substantial change (that has been a long time coming), as I understand many of these limitations already applied to some degree on 64-bit versions of Windows as it is (such as the removal of 16-bit support), but I would be interested in the thoughts of folks smarter than I am. My personal concerns, for example, how would this affect old video games that we boot in DOS mode? How would this affect hypervisors, boot times, security, reliability, etc. - I thought it was a pretty interesting read that had me thinking.



Envisioning a Simplified Intel Architecture

Direct link to the PDF

I watched a good Intel presentation about how an Intel CPU boots in Windows, and indeed the complexity required to support all those ancient standards is mind-boggling (it starts with loading 8-bit instructions, then 16 and so on). It would be a welcome step to debloat the x64 processors and come a bit closer to the cleanliness of ARM.

 

[AusWolf]

Yeah, and by the time a processor with this ISA rolls out, it would be a particularly niche case. For Intel to publish this asking for feedback, it's probably something they're accounting at the design stage of a processor that is still in the far future.

They'd better account for it. I don't care if it's 2023, 2033 or 2043, a good game is a good game, and just because something was released in 1999, it doesn't mean I don't want to play it again.

 

[Dr. Dro]

They'd better account for it. I don't care if it's 2023, 2033 or 2043, a good game is a good game, and just because something was released in 1999, it doesn't mean I don't want to play it again.

Yeah, man. I've seen people calling it an Itanium moment but it just seemed wrong to me. It's really thought out and currently existing x64 software wouldn't be at all affected by this change, or at least shouldn't... I sure hope no Windows 64 bit software is making use of 32-bit ring 0 but hey you never know these days

Peepos that voted no, what are your thoughts and concerns on it? I've not been able to see a negative when it came to concisely supporting a modern 64 bit operating system.

 

[nageme]

More discussion:

Intel Exploring x86S Architecture, Envisions an Unadulterated 64-bit Future

Intel has published a highly involved and extensive whitepaper on the topic of streamlining its CPU architectures, most notably by focusing on a purely 64-bit specification, and consequently dropping legacy 32-bit operating modes (as well as 16-bit!). Team Blue's key proposal states: "This...

www.techpowerup.com

 

[Frag_Maniac]

If they do end up using such a thing, I certainly hope if it in any way adds more bloat to the OS, you can opt out of using it. I mean I am not really one to revisit games with really old graphics anyway. Certainly not 16 bit, and I wouldn't even be bothered by being limited to 64 bit.

At my age, with the way my vision is slowly degrading, I prefer the best image quality I can get.

 

[rdos]

This initiative assumes everybody wants to run Windows or Linux 64-bit versions. I also think most of these proposed changes will not make much of a difference in the transistor counts of the CPU. Intel processors already have horrible segmentation support, and so their processors are unlikely to have much if any hardware related to segmentation. I find it more likely this is some marketing ploy, and perhaps relates to licensing and the fact that AMD seems to produce better processors.

They can remove legacy bloats from 64-bit operating systems by adding the new prosed features to boot application processors directly to long mode WITHOUT breaking legacy operating systems. Most of the bloat in Windows and Linux is due to poor design choices in the operating systems, not legacy functions of the processor, which they don't use anyway.

 

[Dr. Dro]

This initiative assumes everybody wants to run Windows or Linux 64-bit versions. I also think most of these proposed changes will not make much of a difference in the transistor counts of the CPU. Intel processors already have horrible segmentation support, and so their processors are unlikely to have much if any hardware related to segmentation. I find it more likely this is some marketing ploy, and perhaps relates to licensing and the fact that AMD seems to produce better processors.

They can remove legacy bloats from 64-bit operating systems by adding the new prosed features to boot application processors directly to long mode WITHOUT breaking legacy operating systems. Most of the bloat in Windows and Linux is due to poor design choices in the operating systems, not legacy functions of the processor, which they don't use anyway.

Well, with Microsoft having abandoned 32-bit Windows entirely - Windows 10 version 22H2 Is the last 32-bit compatible version of Windows, and hardware manufacturers following suit (Nvidia has not updated their 32 bit graphics driver since 2018, AMD slightly before that), and Linux having competent AMD64 support since what's basically forever now, that's pretty much not an issue imo. It practically is what you are going to be running on a modern system. It's a non issue, and they are providing virtualization-based support to retain compatibility.

It places emphasis on what a modern operating system is designed around, streamlines startup procedure and does away with some potentially vulnerable attack surfaces. Works for me, I suppose.

 

[JoeTheDestroyer]

My personal concerns, for example, how would this affect old video games that we boot in DOS mode?

Most dos games don't run on modern hardware anyways, timing and hardware (sound cards, etc) issues. This is why DOSBox exists, which would not be affected by this change.

Before I vote, I need to know if 16 and 32-bit games would run that way or not.

Per the document: "The 32-bit submode of Intel64 (compatibility mode) still exists"
There would be no effect on 32-bit Windows applications running on 64-bit Windows.

As for 16-bit Windows applications, support has been dropped for them in Windows 11, so there's no benefit to far future hardware (when Win10 is too outdated to use) continuing support. At that point, emulation would suffice.

Accurate, full emulation of vintage PC hardware is very slow - you need an IPC brute of a CPU to run a 300 MHz Pentium II at full speed on PCem or 86Box.

For any software that would be affected by this change, that speed is more than sufficient.

---

Good proposal, as far as I'm concerned.

 

[Panther_Seraphin]

I agree that trimming some of the bloat of x86 away would be beneficial from a power consumption/complexity and may open up new avenues for development locked down currently due to legacy behaviour requirements.

However, what makes me a little dubious is that every time Intel has tried to introduce/update the architecture since the original x86 from the 8086 they have decided to do something stupid/confusing/anti competative/profit mongering vs just doing something for the betterment of computing

iAPX432
Itanium
XScale (Abandoning ARM just as it starts booming)

So Should x86 be slimmed down/refocused/improved? Yes
Do I trust Intel to be the leaders/do the right thing? Not as far as I can throw them

 

[Dr. Dro]

Most dos games don't run on modern hardware anyways, timing and hardware (sound cards, etc) issues. This is why DOSBox exists, which would not be affected by this change.


Per the document: "The 32-bit submode of Intel64 (compatibility mode) still exists"
There would be no effect on 32-bit Windows applications running on 64-bit Windows.

As for 16-bit Windows applications, support has been dropped for them in Windows 11, so there's no benefit to far future hardware (when Win10 is too outdated to use) continuing support. At that point, emulation would suffice.


For any software that would be affected by this change, that speed is more than sufficient.

---

Good proposal, as far as I'm concerned.

I am aware of DOSBox, and use it regularly myself. But there are some purists that keep a modern DOS (like FreeDOS) to run games on the bare metal, this wouldn't be possible anymore as I understand. btw, 16-bit was actually dropped way back in XP64... none of the 64-bit versions of Windows support them, as NTVDM was removed.

I agree, it is a good proposal IMO

I agree that trimming some of the bloat of x86 away would be beneficial from a power consumption/complexity and may open up new avenues for development locked down currently due to legacy behaviour requirements.

However, what makes me a little dubious is that every time Intel has tried to introduce/update the architecture since the original x86 from the 8086 they have decided to do something stupid/confusing/anti competative/profit mongering vs just doing something for the betterment of computing

iAPX432
Itanium
XScale (Abandoning ARM just as it starts booming)

So Should x86 be slimmed down/refocused/improved? Yes
Do I trust Intel to be the leaders/do the right thing? Not as far as I can throw them

Any steps Intel take would need to be agreed upon by AMD and possibly the VIA/Centaur folks as well. All x86 vendors would need to do so... and remember, this is also potentially just coming up now because the time that x86's most critical patents are going to expire is coming up fast.

 

[JoeTheDestroyer]

But there are some purists that keep a modern DOS (like FreeDOS) to run games on the bare metal, this wouldn't be possible anymore as I understand.

*shrug*

btw, 16-bit was actually dropped way back in XP64... none of the 64-bit versions of Windows support them, as NTVDM was removed.

But is still available up through Win10 in its 32bit variant. Win11 is the first to drop support entirely (presumably as an artifact of them dropping 32bit variants).

I only just stumbled across this, but apparently there's a third-party hack to get ntvdm working even on 64bit windows. I suspect this still relies on VM86 mode, which the proposal would remove.

 

[Panther_Seraphin]

Any steps Intel take would need to be agreed upon by AMD and possibly the VIA/Centaur folks as well. All x86 vendors would need to do so... and remember, this is also potentially just coming up now because the time that x86's most critical patents are going to expire is coming up fast.

Historically this isnt what has happened previously.

Someone releases something (MMX, 3DNOW!, SSE etc) and if the market adopts it en mass it tends to be a case then that the opposition licences it.

That is why we have x86-64 being the dominant 64bit processor as the market didnt want to move away from x86 at the time as Itanium didnt offer anything beneficial at the time and none of the supposed benefits came to fruition before x86 either surpassed IA-64 or x86-64 became adopted enmass.

 

[eidairaman1]

Historically this isnt what has happened previously.

Someone releases something (MMX, 3DNOW!, SSE etc) and if the market adopts it en mass it tends to be a case then that the opposition licences it.

That is why we have x86-64 being the dominant 64bit processor as the market didnt want to move away from x86 at the time as Itanium didnt offer anything beneficial at the time and none of the supposed benefits came to fruition before x86 either surpassed IA-64 or x86-64 became adopted enmass.

Before it was x86-64 it was AMD64 intel had to license it from AMD

 

[Dr. Dro]

Historically this isnt what has happened previously.

Someone releases something (MMX, 3DNOW!, SSE etc) and if the market adopts it en mass it tends to be a case then that the opposition licences it.

That is why we have x86-64 being the dominant 64bit processor as the market didnt want to move away from x86 at the time as Itanium didnt offer anything beneficial at the time and none of the supposed benefits came to fruition before x86 either surpassed IA-64 or x86-64 became adopted enmass.

Very different market conditions now. There's an established x86_64 architecture and user base. If Intel unilaterally changes things, it's shooting their own feet. Speaking of 3DNow!, I'm still surprised AMD didn't revive that branding for the 3D V-Cache CPUs. Would be win.

 

[eidairaman1]

3D Now was an instruction set that was dropped in 2010 because no one utilized it.

 

[agent_x007]

Historically this isnt what has happened previously.

Historically, it never happened before.
There were only attempts to drop it as a whole (Itanium), or adding more stuff on top of old things (64-bit). NEVER core features (like 8-bit/16-bit) that were previously "mandatory", were ever dropped from "x86". Because of this alone, any comparisons to earlier attempts at changes are invalid.

Also, just because those features are dropped doesn't mean you suddenly can't use old hardware.
It's going to take MANY decades before hardware supporting "old x86" dies out (ie. everything produced until today + maybe one or two generations more [only those which don't take x86S into account]).
It probably will have some quirks, but I bet you will be able to run Windows 12 (13/14?) on Raptor Lake or Windows 7/10 on at least a couple of next few generations of AMD and Intel platforms.

Lastly, nobody forces you to buy new stuff, so if you don't like it - don't buy it.
If you work as IT programmer, sorry to hear you have to change your code, but move on with times (you are payed to do this, right ?). Either way, I hope programmers will feel better about this knowing that what they feel right now is probably what 80s and 90s programmers felt like when Intel released 286/386, then Pentiums, or Hyper threading with AMD's 64-bit and multi core CPUs at mid 2000s mark

 

[Dr. Dro]

Historically, it never happened before.
There were only attempts to drop it as a whole (Itanium), or adding more stuff on top of old things (64-bit). NEVER core features (like 8-bit/16-bit) that were previously "mandatory", were ever dropped from "x86". Because of this alone, any comparisons to earlier attempts at changes are invalid.

Also, just because those features are dropped doesn't mean you suddenly can't use old hardware.
It's going to take MANY decades before hardware supporting "old x86" dies out (ie. everything produced until today + maybe one or two generations more [only those which don't take x86S into account]).
It probably will have some quirks, but I bet you will be able to run Windows 12 (13/14?) on Raptor Lake or Windows 7/10 on at least a couple of next few generations of AMD and Intel platforms.

Lastly, nobody forces you to buy new stuff, so if you don't like it - don't buy it.
If you work as IT programmer, sorry to hear you have to change your code, but move on with times (you are payed to do this, right ?). Either way, I hope programmers will feel better about this knowing that what they feel right now is probably what 80s and 90s programmers felt like when Intel released 286/386, then Pentiums, or Hyper threading with AMD's 64-bit and multi core CPUs at mid 2000s mark

Even then I'm not sure this is a concern at all. You cannot deploy such old OSes on new PCs, the oldest you can get on a 12th/13th Gen platform is 32-bit Win10 and I don't think anyone actually uses that, too much left on the table, no support for modern GPUs. The last Nvidia driver for 32-bit Windows 10 was the one that retired Fermi, 391.35 all the way back in 2018.

If one needs native 16-bit support they can likely do okay with some old Pentium III or 4 system