Was there a different 64-bit design for x86 from Intel?

Question

I remember reading an article in a computer magazine around the year 2000 (most likely in the German c't magazine), about the different (planned?) 64-bit extensions for the x86 family. It had an article explaining Intel's design, and an article explaining AMD's design. As I was reading those articles, I remember thinking that AMD's design sounded much more straightforward and reasonable.

Fast forward a few years, and Intel did adopt AMD's 64-bit design, much to my surprise.

Now I wonder: Did Intel design their own 64-bit x86 extension and has ditched that in favour of AMD's design or do I misremember that? I'm somewhat sure that said article did not describe IA-64 (aka Itanium, a 64-bit only architecture), but instead a x86-compatible solution. The x86-64 Wikipedia article does not mention such a design, it only mentions that Intel had adopted AMD's design with a few minor differences.

Answer 1

TL;DR: No, Intel did not want x86-64, they had IA-64

The rest was FUD, power play by Microsoft and market reality.

---

Zac67's answer may point out some important confusion, except in my recollection there was a bit more to the story than confusion and Intel creating quite some hot air about countering AMD64 with their own design. Looking back, I'm not sure how much of that was true, but in the industry it was widely seen as such, and so did I. To understand it may help to take a look at the bigger picture:

Some History

Intel had done a very similar move just the year before by introducing SSE with the (Pentium III, 1999) to counter AMD's 3dNow! (K6-2, 1998), thus when AMD released their AMD64 (aka x86-64) spec in 2000, everyone was expecting intel to show off their own thing, ofc, not compatible to AMD, as with SSE before.

Previous extensions were usually only announced shortly before fitting Hardware (CPUs) were introduced. In contrast AMD64 was announced in 1999 and publicly documented in full in 2000, while first hardware (K8) took three more years until 2003 as Opteron in Spring and Athlon 64 right for Christmas 2003.

The Early Naughties (00..05)

It further helps to remember the over all time line, as Intel entered naughties shell shocked by AMD's superior 1999 K7/Athlon CPU line that clock for clock outperformed the Pentium III and reached higher clock rates at the same time. The Pentium III was still based on a 1995 Pentium Pro (P6) architecture. It wasn't until the 2007 core architecture that Intel regained ground with the Core design, the first really new architecture - no, the Pentium 4 was an utterly failure. Great on first sight, but deadly crippled when running real world loads. Also still a direct P6 offspring (thus named P68).

Enter Microsoft

Now add another major player to this 2000 to 2005 backdrop: Microsoft

It was the time that pushed very hard into the server market. They had just released Windows 2000 Server in 1999 and started to work on Windows Server 2003, intended to be a distinct, server first product (hence the name change). Windows 2003 was also posed to support IA-64, Intel's sole 64 bit product in form of the Itanium CPU. Adding AMD64 support was seen as a possible additional target, especially as it supports upward compatibility with acceptable performance - frankly, Intels IA-32 support on IA-64 sucked.

But MS did neither have capacity nor any intention to support two different 64 bit extensions for x86 at kernel level - which is the really important part here. Having a bunch of additional instructions at user level (think MMX or SSE) doesn't matter at all, while having more registers to save and restore are minor details usually covered by the most basic HAL-functions. But AMD64 not only introduced new 64 registers and addressing, but also a new memory management model, which at least requires adapting and hardening all modules touching virtual memory - which can be a lot. This is where the cost of two different 64 bit extensions for MS were and why they wanted to avoid them as much as possible.

Intel Dragging its Feet

While Intel was in 2000 still licking the Athlon wounds and pressing ahead with a firm commitment to IA-64, AMD delivered full AMD64 documentation in early 2000 and prototype hardware during 2002 to Microsoft. Everything a software developer could wish for. At that point Intel might have still convinced themself that it's a fast fading buzz. That is until AMD presented the Opteron 64 in April 2003 - right the same time as Microsoft premiered its Windows Server 2003. Of course shown off running at Opteron based servers - also the reason why the Opteron 64 was pulled before the Athlon 64.

Intel Being Dragged

And guess what? It was a success. Server operators saw the Opteron as a great way upward at a lower cost than IA-64. After all, they were faster and more cost efficient even without running a 64 bit OS.

So Intel seeing their customers going for faster and larger AMD64 servers instead of IA-64 (or more Pentium 4), needed a stop gap measure for their server business (Xeon). So a Pentium 4 variant supporting EMT64 was introduced a good year later in June 2004. (I believe there was also a Pentium 4 version shown at the Microsoft Server 2003 event capable of doing Intel64, but I can't find any evidence)

In fact, the exact timeline of AMD64 introduction with the Prescott line paints a good picture of that 'dragging'.

• In February 2004 Intel introduced a Pentium 4 using the 90 nm Prescott core as Family 15 Model 3. This core did already was already fitted with Intel64, but disabled.
• In June 2004 the Nocona based Xeon was offered with Intel64.
• Finally in November 2004 the Prescott E0 Stepping was announced which now also enabled Intel64.

Part of this forth and back can be heared in David Cutlers 2023 recollection recorded 2023 by Dave's Garage, which Sebastian Koppehelt found.

Fudding Names

Originally Intel named the 'new' extension EM64T for Extended Memory 64 Technology, which nicely shows the whole reasoning behind adding it: the new Memory Management fitting what Microsoft implemented in Windows Server 2003 and the later High end Workstation Version Windows XP Professional x64 Edition (Not to be confused with the Itanium Version Windows XP 64-Bit Edition).

Now when those CPU finally arrived, Intel choose Intel64 as marketing name - obviously to counter AMD64 being mainly used. Over time public simply choose to call it x64 or likewise - today morphed into x86-64.

Conclusion

Intel had in 2000 no intentions to build a 64 bit x86. Their server focus (and that's what 64 bit was seen for) was at 100% IA-64 aka Itanium. They created a notable level of FUD - and maybe some half founded development as support - to stop AMD from gaining ground in server applications with their AMD64. Mostly felling how inefficient and dead end a 64 bit rucksack for a low end technology like IA-32 is.

Microsoft in turn was as well geared to IA-64 support in addition to 32 Bit x86 (IA-32). Supporting a 64 Bit x86 as well was fine to them. Except there was no time nor willingness to support two different 64 bit extensions. AMD delivered documentation early on and customer ready hardware right in time (2003) for Windows Server 2003 to shine. Intel could only react to jump onto the already rolling train. No room for any development of their own.

Answer 2

Even before Intel management saw which way the wind was blowing, Intel CPU architects had built "our own internal version of x86-64" into Pentium 4. This is from a Quora post by Bob Colwell, who was Intel's Chief x86 Architect before he retired in 2000, having worked on P6 (PPro through PIII) and Pentium 4.

Bob Colwell on Quora in 2023

Intel’s Pentium 4 had our own internal version of x86–64. But you could not use it: we were forced to “fuse it off”, meaning that even though the functionality was in there, it could not be exercised by a user. This was a marketing decision by Intel — they believed, probably rightly, that bringing out a new 64-bit feature in the x86 would be perceived as betting against their own native-64-bit Itanium, and might well severely damage Itanium’s chances. I was told, not once, but twice, that if I “didn’t stop yammering about the need to go 64-bits in x86 I’d be fired on the spot” and was directly ordered to take out that 64-bit stuff. I decided to split the difference, by leaving in the gates but fusing off the functionality. That way, if I was right about Itanium and what AMD would do, Intel could very quickly get back in the game with x86. As far as I’m concerned, that’s exactly what did happen.

This phrasing I highlighted suggests that Intel corporate management was (early on at least) solidly opposed to any 64-bit extension to x86.
(At that point, presumably in the late '90s before the first P4 release in Nov 2000, so well after the P7 project Eugene Styer's answer mentioned, cancelled in 1994 to focus on IA-64 as the 64-bit future.)

A few news articles like https://www.techspot.com/news/105222-intel-could-have-beaten-amd-x86-64-transition.html have been reporting on this in the last couple days as AMD engineer Phil Park found it while researching x86-64 history and quoted it on twitter

Anyway, Bob Colwell retired from Intel in 2000, well before 90nm Prescott / Nocona in 2004 was the first Intel 64 CPU. It takes a long time to get a CPU design from initial development to tape-out + release, but presumably he was talking about Willamette or at the latest Northwood (released 2002).

Initial Prescott steppings shipped with 64-bit support disabled for some reason. I don't think this is what Bob was talking about, though, since this was after AMD's x86-64 CPUs came to market and were eating their lunch. Nocona is the Xeon version of Prescott, and Intel loves to only enable some features in the same silicon when it's sold as a more expensive Xeon (server/workstation) instead of a less expensive client (desktop/laptop) CPU.

---

Prescott did significantly re-architect things, like widening the integer ALUs to two 32-bit stages instead of two 16-bit stages, allowing 64-bit integer addition. (But also increasing add eax, ecx latency from 0.5 cycles to 1 cycle it seems, see comments and answer on an SO question. A Chips & Cheese article on Prescott-era Netburst unfortunately doesn't discuss the ALU latencies.)

I'm curious how x86-64 on Willamette / Northwood would have worked. Maybe the adders were wider and their high halves are part of what got fused off? Or maybe 64-bit operand-size instructions would decode to 2 micro-ops?

Answer 3

Intel worked on a 64-bit extension to the Pentium prior to developing the Itanium, but the article does not indicate any details of the Intel design. AMD64 wasn't publicly announced until 1999, so I suspect that this is not the design the OP remembers.

From What’s Up With Willamette? (Part 1) (the Wikipedia article on the Itanium links to this article):

In early 1993 Intel’s Santa Clara processor design team had just finished off the P5 project (Pentium) and started work on the P7. . . . The P7 was a powerful 64-bit x86 compatible successor to the P6 envisioned to have around 20 million transistors or nearly four times as many as the Pentium Pro. In some ways Intel’s original P7 project conceptually resembles AMD’s K8 “sledgehammer”, the 64-bit successor to the K7 Athlon.

The P7 progressed only far along enough for Intel’s engineers to realize that extending x86 to 64 bits, and staying competitive with RISC processors, would be challenging to say the least. . . . In 1994 the Santa Clara team dropped all work on the 64 bit x86 processor design called P7 and started on the first implementation of the new IA-64 architecture arising from the Intel-HP alliance, a processor later known as Merced.

Answer 4

You must be referring to either the IA-64 architecture used on the Itanium - an entirely new and incompatible ISA - or Intel 64 (initially Clackamas Technology/CT, then IA-32e or EM64T), Intel's adaptation of AMD64, an incremental approach from x86 or IA-32.

There are some differences between Intel 64 and AMD64, especially in lower-ring implementations, that the magazine might have explored (does sound like c't's Andreas Stiller). Together, AMD64 and Intel 64 are often referred to as x86-64.

That Intel worked on another 64-bit successor to x86 is not likely, or at least not in common knowledge - at least not at that time, when the Itanium had already been selected.