Intel proposes new X86-S architecture

[AlphaBeta]

https://cdrdv2.intel.com/v1/dl/getContent/776648

tl;dr:

• starts execution directly in 64-bit mode
• supports only rings 0 and 3
• 16-bit mode removed, including v86 mode
• 32-bit mode supported only for user apps (ring 3)
• segmentation for 32-bit mode gutted to the same model used in 64-bit mode
• port I/O supported only in ring 0
• string I/O removed

[ActivateYourTech]

If it is x64 only, and you can't run any 32 or 16-bit programs, then why is it called x86-S? Why not call it x64-64 since the x64 we use is actually x86-64?

[rafaelgs]

If it is x64 only, and you can't run any 32 or 16-bit programs, then why is it called x86-S? Why not call it x64-64 since the x64 we use is actually x86-64?

32-bit programs are still supported, read again…

• 32-bit mode supported only for user apps (ring 3)

[AlphaBeta]

If it is x64 only, and you can't run any 32 or 16-bit programs, then why is it called x86-S? Why not call it x64-64 since the x64 we use is actually x86-64?

For the record, x86-64 stands for 64-bit x86. x64 is a Microsoftism.

[ActivateYourTech]

Oh. I got the information from this article, which said "Under this proposal, those wanting to run legacy operating systems or 32-bit x86 software would have to rely on virtualization."

https://www.phoronix.com/news/Intel-X86-S-64-bit-Only

[vbdasc]

Will NTVDM under, say, Windows XP still work if the hardware virtualization extensions are used? This "16-bit mode removed" sounds ominous.

And why are Intel doing that? Will there be any benefit, besides trimming like, 0.05% of the chip die? Has Intel nothing better to do, in the face of abysmal financial results and the onslaught of ARM?

One of the reasons people run x86 instead of other architectures is backward compatibility. Sacrificing it in order to look progressive and relevant could prove to be a disastrous move.

[xyz]

This "16-bit mode removed" sounds ominous.
And why are Intel doing that? Will there be any benefit, besides trimming like, 0.05% of the chip die? Has Intel nothing better to do, in the face of abysmal financial results and the onslaught of ARM?
One of the reasons people run x86 instead of other architectures is backward compatibility. Sacrificing it in order to look progressive and relevant could prove to be a disastrous move.

In German there is a spell "es wird nichts so heiß gegessen, wie es gekocht wird". It fits quite well to your post.

You can be sure that Windows will provide compatibility layers and/or emulation for 32bit programs in case these are still of relevance when x86-S CPUs are market-ready. 16bit support is anyway dead since EFI and x64 Windows versions without the use of otvdm or emulation. Most people will not notice the loss of native 32bit support, after all they didn't even noticed the loss of 16bit (except for nerds like us and few industries). CPUs with x86-S will be anway so fast that they will - even with emulation - still outcompete any native 32bit x86 CPU when running the same program.
And there will be legacy PC emulators for x86-S too, similar to what x86box is today for x64 Windows.

[Hyoenmadan86]

I've heard rumors about this x86-S for almost 4 or 5 years now. Back then was supposed to be only the removal of every trace of still existing legacy platform hardware on x86 (DMA controller, Legacy IRQ controller, LPC bus to be replaced by I2O bus, ps/2 chipset support for keyboard/mouse emulation on both CSM and UEFI only environments), and kickstarting the CPU directly in 32bit mode with APIC enabled. 32bit support would still exist for both UEFI OS and applications, along with v86 mode to be used by any emulation and the Virtualization Extensions were still supposed to exist for some time more. A sort of soft transition.

Now I see they decided to rush and just snap all the legacy bits altogether. No transition at all. They must be wary on AMD plans to not only do the same, but also offer their in-development hybrid ARM/x86 architecture on their next line of CPUs, and they want to come with their "new architecture" before them, so Intel can regain the "standard" from AMD and their AMD64. Because what Intel will do isn't "x86 standard" and will need specific HAL and kernel support from OS, in the same way new AMD x86/ARM hybrid CPUs will need. Microsoft always support only one way to do x86 (last time MS ditched Intel EMT64 in favor of AMD64), and looks like Intel wants to win the race this time.

Is nostalgic to see 16bit getting removed from the CPU silicon... But is true right now v86 is useless even by using VTx/SVN virtualization extensions. Ntvdmx64 developer (which fixed and offers patches to build the original MS NTVDM from the leaked windows sources) experimented with enhancing the VDM core with VTx extensions (which in turn make use of V86 available in 32bit mode), just to find VGA/Video and sound performance is subpar from even software emulation because looks like them lack cases to support emulation of hardware in usermode. Works fine for terminal only apps, but as soon as you want to play graphic applications or games, it will work slower in the range of software only emulation.

You would think Intel would add extensions to VTx to handle the cases of virtualized v86 CPU mixed with software emulation of VGA/graphics... Instead they decided remove every trace of remaining 16bit from silicon altogether. A big shame, as emulation is still too slow when you approach last DOS gaming era. Specially if your CPU is some underpowered "i3" APU or anything in the range.

[yksoft1]

Does this mean the end of IBM PC compatibles is finally coming?
(Currently there are still some new motherboards with CSM module in their UEFI so technically those are still IBM compatibles).

[eflanili7881]

I think with this architecture, Intel finally removed last bit remains of DOS, as it states 16-bit mode removed.