Did Intel ever develop its own x86-64 a long time ago?

[Yousuf Khan]

There was a rumour not so long ago, after Intel had copied the AMD64
instruction set that Intel had already developed and abandonned its own
extensions a long time ago, so it wasn't really copying AMD. I think
either Barrett or Otellini or both implied as much in some statements,
which I can't find right now.

Anyways this old article from around the time of the Opteron's original
release, there was an interview with an Intel official. The official
stated that they had investigated adding 64-bit extensions to the x86,
but decided it was too difficult.

"Despite the advantages, converting a 32-bit machine into a 64-bit one
isn't easy. Four separate design teams at Intel examined how the company
could take one of its 32-bit chips and transform it into a 64-bit
machine, said Richard Wirt, another senior fellow at Intel. After
running simulations, all four teams concluded that such a transition
wouldn't be economically feasible, he said."
http://news.com.com/2100-1001-985432.html

Obviously the following year after this article, Intel changed its mind
completely and added x64 after all. But it does show it had no active
pre-existing research project into the 64-bit extensions, and it just
followed AMD's design lead on it.

Yousuf Khan

 

[David Kanter]

There was a rumour not so long ago, after Intel had copied the AMD64

instruction set that Intel had already developed and abandonned its own
extensions a long time ago, so it wasn't really copying AMD.

That's not a rumor.

I think
either Barrett or Otellini or both implied as much in some statements,
which I can't find right now.

It wasn't just them.

Anyways this old article from around the time of the Opteron's original
release, there was an interview with an Intel official. The official
stated that they had investigated adding 64-bit extensions to the x86,
but decided it was too difficult.

I really hope you don't believe everything you read.

"Despite the advantages, converting a 32-bit machine into a 64-bit one
isn't easy. Four separate design teams at Intel examined how the company
could take one of its 32-bit chips and transform it into a 64-bit
machine, said Richard Wirt, another senior fellow at Intel. After
running simulations, all four teams concluded that such a transition
wouldn't be economically feasible, he said."
http://news.com.com/2100-1001-985432.html

He's toeing the party line here, converting to 64b isn't that hard.
The P4 made it a bit harder...but hardly impossible. Intel probably
decided not to go down this route because they wanted folks to use IPF
and because it didn't make sense to push desktop x86 to 64b. Now, that
might have been a good differentiating feature for Xeons versus
Pentiums, but I think it's too much of a change to be feasible between
what are largely similar cores.

Moreover, you should see the large hole in your theory:
Intel might have more than 4 design teams. Oops, suddenly Wirt's
statement isn't as insightful.

Obviously the following year after this article, Intel changed its mind
completely and added x64 after all. But it does show it had no active
pre-existing research project into the 64-bit extensions, and it just
followed AMD's design lead on it.

No, they really did...a long time back. I can't speak for Wirt or any
executives, but I know people who worked on 64 bit extensions to x86
that were canned. You're obviously entitled to believe what ever you
want, but your conclusions are wrong.

DK

 

[gaf1234567890]

Wasn't that the imfamous "Yamhill" project?

There's certainly a ton of old stuff still on the web about it.

 

[David Kanter]

Wasn't that the imfamous "Yamhill" project?

There's certainly a ton of old stuff still on the web about it.

No Yamhill was Intel's reverse engineering of AMD's x86-64 spec. What
I am talking about is before Yamhill.

DK

 

[nobody]

There was a rumour not so long ago, after Intel had copied the AMD64
instruction set that Intel had already developed and abandonned its own
extensions a long time ago, so it wasn't really copying AMD. I think
either Barrett or Otellini or both implied as much in some statements,
which I can't find right now.

Anyways this old article from around the time of the Opteron's original
release, there was an interview with an Intel official. The official
stated that they had investigated adding 64-bit extensions to the x86,
but decided it was too difficult.

"Despite the advantages, converting a 32-bit machine into a 64-bit one
isn't easy. Four separate design teams at Intel examined how the company
could take one of its 32-bit chips and transform it into a 64-bit
machine, said Richard Wirt, another senior fellow at Intel. After
running simulations, all four teams concluded that such a transition
wouldn't be economically feasible, he said."

Of course it "wouldn't be economically feasible" because it would sink
Itanic - just as we are witnessing today. With all the billions of
investment sinking with it.

http://news.com.com/2100-1001-985432.html

Obviously the following year after this article, Intel changed its mind
completely and added x64 after all. But it does show it had no active
pre-existing research project into the 64-bit extensions, and it just
followed AMD's design lead on it.

Was it pre-existing Yamhill design with some tweaks to get it
compatible with AMD instruction set, or just a rush copy job (call it
reverse-engineering of AMD64 or whatever) - doesn't matter at this
point. Now AMD64 is a fact of life, and most likely Vista will take
it mainstream.

 

[YKhan]

Of course it "wouldn't be economically feasible" because it would sink
Itanic - just as we are witnessing today. With all the billions of
investment sinking with it.

LOL! Sure that's one way to interpret that.

But I think what they were referring to was that they were looking to
extend the x86 architecture out to 64-bit even before they embarked on
the Itanium project. And then after they did all of the feasibility
studies, they decided that Itanium was the better 64-bit route to
follow.

Was it pre-existing Yamhill design with some tweaks to get it
compatible with AMD instruction set, or just a rush copy job (call it
reverse-engineering of AMD64 or whatever) - doesn't matter at this
point. Now AMD64 is a fact of life, and most likely Vista will take
it mainstream.

No, Yamhill was initiated after AMD64 came out. It then got renamed to
Clackamas, then CT, then IA-32e, and finally EM64T. It was always after
AMD64, but the only thing mysterious about it was whether it was going
to follow AMD64 exactly, or was it going to venture out in a different
backward-compatible direction. I got the feeling that they were
venturing into a different direction with Yamhill, but then Microsoft
got word of it, and told them to change it back to AMD64, and thus it
got renamed to Clackamas.

Yousuf Khan

 

[chrisv]

No Yamhill was Intel's reverse engineering of AMD's x86-64 spec. What
I am talking about is before Yamhill.

In the end, didn't Intel trade SSE(X) for AMD64?

 

[Yousuf Khan]

I really hope you don't believe everything you read.

You are really going to have to be more specific here. How far did Intel
get in its design? Did it define new instructions, new memory models,
lay down circuits, etc.? Or did it just run some preliminary
simulations? From what I can see here, all they ever did was run some
simulations and then declared it a lost cause.

He's toeing the party line here, converting to 64b isn't that hard.
The P4 made it a bit harder...but hardly impossible. Intel probably
decided not to go down this route because they wanted folks to use IPF
and because it didn't make sense to push desktop x86 to 64b. Now, that
might have been a good differentiating feature for Xeons versus
Pentiums, but I think it's too much of a change to be feasible between
what are largely similar cores.

Actually it was a pretty major feat. I certainly didn't think it was
possible, either. They increased some things, and reduced other things;
up until that point the x86 design was just a continuous series of
adding stuff in, but never removing things. I didn't see them doubling
the number of general purpose and SSE registers. Removing the segment
mechanism. And they did this while still maintaining backward
compatibility in the 32-bit modes. I knew if they kept the segment
mechanism, they'd have a lot of trouble, but I never imagined they could
actually decide to get rid of it.

Moreover, you should see the large hole in your theory:
Intel might have more than 4 design teams. Oops, suddenly Wirt's
statement isn't as insightful.

All Wirt said was that they assigned 4 design teams to work on this
specific project. They're not going to assign all of their design teams
to work on this project. And the 4 teams just did simulations, and then
stopped. We don't know how detailed of a simulation they did.

No, they really did...a long time back. I can't speak for Wirt or any
executives, but I know people who worked on 64 bit extensions to x86
that were canned. You're obviously entitled to believe what ever you
want, but your conclusions are wrong.

But you don't know how far along the design process these guys got into.

Yousuf Khan

 

[Yousuf Khan]

In the end, didn't Intel trade SSE(X) for AMD64?

I think they didn't have to trade anything for anything. Intel's cross
license allowed it to take AMD64, while AMD's cross licence allowed it
to take SSE. However, AMD's cross license didn't allow it to take
Intel's FSB, therefore Intel's cross license didn't allow it to take
Hypertransport, either.

Yousuf Khan

 

[David Kanter]

You are really going to have to be more specific here. How far did Intel

get in its design? Did it define new instructions, new memory models,
lay down circuits, etc.? Or did it just run some preliminary
simulations? From what I can see here, all they ever did was run some
simulations and then declared it a lost cause.

Unfortunately, I cannot be much more specific in public...since a lot
of this information could get someone fired or in trouble.

Actually it was a pretty major feat.

I don't really agree. I think the path to 64b was pretty obvious, and
both AMD and Intel probably filled in the dots (so to speak) in their
specs. I bet if you talked to smart engineers from AMD and Intel
in...let's say 1997 about a hypothetical 64b x86 they would come up
with very similar notions about what to do.

I certainly didn't think it was
possible, either. They increased some things, and reduced other things;
up until that point the x86 design was just a continuous series of
adding stuff in, but never removing things. I didn't see them doubling
the number of general purpose and SSE registers. Removing the segment
mechanism. And they did this while still maintaining backward
compatibility in the 32-bit modes. I knew if they kept the segment
mechanism, they'd have a lot of trouble, but I never imagined they could
actually decide to get rid of it.

Keeping segments in 64b would have been retarded. I wouldn't have done
it, and if I could figure that out, I'm sure Intel and AMD could.

All Wirt said was that they assigned 4 design teams to work on this
specific project. They're not going to assign all of their design teams
to work on this project. And the 4 teams just did simulations, and then
stopped. We don't know how detailed of a simulation they did.

His statement also doesn't rule out that the fact that another team
might have gone a lot farther. Let's just pretend for a second,
suppose one of the teams found that it was a great idea to go to 64b
and would improve performance, cure cancer, end world hunger etc.

Now, suppose someone said: "this will kill Itanium" and Intel therefore
decided x86 should be 32b as a result. Do you expect someone from
Intel to say "We decided not to pursue x86 for marketing reasons
relating to IA64?" in public? Maybe 20-30 years from now, but not in
2003.

But you don't know how far along the design process these guys got into.

I know a bit; not a lot, but I can't really share a lot of that
information.

DK

 

[Yousuf Khan]

His statement also doesn't rule out that the fact that another team
might have gone a lot farther. Let's just pretend for a second,
suppose one of the teams found that it was a great idea to go to 64b
and would improve performance, cure cancer, end world hunger etc.

Now, suppose someone said: "this will kill Itanium" and Intel therefore
decided x86 should be 32b as a result. Do you expect someone from
Intel to say "We decided not to pursue x86 for marketing reasons
relating to IA64?" in public? Maybe 20-30 years from now, but not in
2003.

Oh, I see, you're saying that perhaps a fifth team that went unmentioned
did not come to the same conclusion and Wirt is just mentioning the
majority opinion amongst the teams?

Yousuf Khan

 

[David Kanter]

Oh, I see, you're saying that perhaps a fifth team that went unmentioned

did not come to the same conclusion and Wirt is just mentioning the
majority opinion amongst the teams?

Yes, that's one possibility. Also, I believe Wirt said something along
the lines of "it wasn't economically feasible". That is more of a
marketing/product management decision, rather than a technical one.

So perhaps each time came up with a whole spec, and an estimate on the
effort required and potential performacne. Then some marketing folks
said: "This is going to get ugly when we have to position IA64i and
IA64x, let's not do this".

Again, this is mostly guessing, not hard fact. However, they
definitely did look at 64b x86...some of the folks at Intel have
commented on comparing their initial implementation to AMD's
implementation, and then their reverse engineering of AMD's
implementation.

DK

 

[George Macdonald]

Yes, that's one possibility. Also, I believe Wirt said something along
the lines of "it wasn't economically feasible". That is more of a
marketing/product management decision, rather than a technical one.

Both people interviewed are "senior fellows" and both have technical
managment roles, Wirt in software & solutions, Rattner in microprocessor
and architecture research... and are also described as "Two of the
company's top researchers". I think it's fair to assume that this was
technical policy at the time... just one year before EM64T *and* while
Intel was simultaneously vehemently denying the existence of the Yamhill
skunk-works. At the very same time, a month or so later, analysis of die
litho photos clearly showed the bit-sliced 64-bit implementation in the
Prescott. This paints a picture of a very confused company.

So perhaps each time came up with a whole spec, and an estimate on the
effort required and potential performacne. Then some marketing folks
said: "This is going to get ugly when we have to position IA64i and
IA64x, let's not do this".

Again, this is mostly guessing, not hard fact. However, they
definitely did look at 64b x86...some of the folks at Intel have
commented on comparing their initial implementation to AMD's
implementation, and then their reverse engineering of AMD's
implementation.

Where is this "commented on comparing..."?... more secret missives which
might get someone sacked? I'd be very interested to see what the
comments amount to.

 

[chrisv]

I think they didn't have to trade anything for anything. Intel's cross
license allowed it to take AMD64, while AMD's cross licence allowed it
to take SSE.

I don't believe the "old" cross-licenses extended to these new
features, though. For example when Intel added MMX, AMD responded
with a different solution (3D-now). AMD finally adapted SSE, of
course, but lagged Intel by a generation.

Suddenly AMD is at the same version SSE as Intel, and Intel is using
AMD 64-bit extensions. So I'm thinking they negotiated the trade I
mentioned.

 

[Rob Stow]

I don't believe the "old" cross-licenses extended to these new
features, though. For example when Intel added MMX, AMD responded
with a different solution (3D-now). AMD finally adapted SSE, of
course, but lagged Intel by a generation.

Suddenly AMD is at the same version SSE as Intel, and Intel is using
AMD 64-bit extensions. So I'm thinking they negotiated the trade I
mentioned.

The current cross-licensing agreement happened after the MMX vs
3D-Now mess, but it predates SSE2, SSE3, and AMD64. Not sure
where the original SSE stands in that regard.

An interesting thing about the agreement is that neither side
needs to disclose new features to the other until they start
selling x86 chips using those features. This gives the
"inventing" side a brief window in which to capitalize on
temporary advantages - such as what Intel did with SSE3.

 

[David Kanter]

Yes, that's one possibility. Also, I believe Wirt said something along

Both people interviewed are "senior fellows" and both have technical
managment roles, Wirt in software & solutions, Rattner in microprocessor
and architecture research... and are also described as "Two of the
company's top researchers". I think it's fair to assume that this was
technical policy at the time... just one year before EM64T *and* while
Intel was simultaneously vehemently denying the existence of the Yamhill
skunk-works. At the very same time, a month or so later, analysis of die
litho photos clearly showed the bit-sliced 64-bit implementation in the
Prescott. This paints a picture of a very confused company.

Not really, I think there was just a hard 'party line', and everyone
followed it up until it was changed. However, there were efforts to do
64b x86 long before that party line was formulated (and was possibly
the catalyst for such a formation).

Where is this "commented on comparing..."?... more secret missives which
might get someone sacked? I'd be very interested to see what the
comments amount to.

If you look on the internet you should be able to find some of them.

DK

 

[George Macdonald]

I don't believe the "old" cross-licenses extended to these new
features, though. For example when Intel added MMX, AMD responded
with a different solution (3D-now). AMD finally adapted SSE, of
course, but lagged Intel by a generation.

Suddenly AMD is at the same version SSE as Intel, and Intel is using
AMD 64-bit extensions. So I'm thinking they negotiated the trade I
mentioned.

http://contracts.corporate.findlaw.com/agreements/amd/intel.license.2001.01.01.html

 

[George Macdonald]

The current cross-licensing agreement happened after the MMX vs
3D-Now mess, but it predates SSE2, SSE3, and AMD64. Not sure
where the original SSE stands in that regard.

An interesting thing about the agreement is that neither side
needs to disclose new features to the other until they start
selling x86 chips using those features. This gives the
"inventing" side a brief window in which to capitalize on
temporary advantages - such as what Intel did with SSE3.

Hmm, that's an interesting detail I don't see in the confidentiality
obliterated version of the agreement - I'm curious where you got that
from?? Some of such items would likely be seen in patent applications
before chip sales though - no?

 

[Rob Stow]

Hmm, that's an interesting detail I don't see in the confidentiality
obliterated version of the agreement - I'm curious where you got that
from??

A few months before SSE3 showed up in the Opterons I simply
e-mailed AMD to ask
- when SSE3 was going to show up in Opterons ?
- why the delay, given the well publicized cross-licensing with
Intel ?

In response, I got directed to a forum at AMD where those
questions had already been asked and answered.

Some of such items would likely be seen in patent applications
before chip sales though - no?

Would a patent on SSE3 be detailed enough to let AMD to implement
it in their chips in a way that would be instruction-set
compatible with Intel's implementation ?

 

[YKhan]

Both people interviewed are "senior fellows" and both have technical
managment roles, Wirt in software & solutions, Rattner in microprocessor
and architecture research... and are also described as "Two of the
company's top researchers". I think it's fair to assume that this was
technical policy at the time... just one year before EM64T *and* while
Intel was simultaneously vehemently denying the existence of the Yamhill
skunk-works. At the very same time, a month or so later, analysis of die
litho photos clearly showed the bit-sliced 64-bit implementation in the
Prescott. This paints a picture of a very confused company.

We can let Rattner and Wirt get away with simply "toeing the company
line" when it came to not mentioning the Yamhill project, which would
come later. That's because we're not so interested in Yamhill in this
case. Yes, we know Yamhill actually did come true, it's now part of
recent history. Yamhill was a response, but it wasn't self-initiated.

But we're more interested in the older history. Was there ever an x64
project at Intel prior to AMD64's appearance (and of course, prior to
Yamhill)? Wirt says, "Well kinda, but we didn't get too deep into it."
This is where there is a mystery. They're using purposefully vague
phrases like "cost ineffective", which doesn't say how serious they
were about it, and how far they got. It also doesn't say how
technically feasible it would've been, because "cost ineffective" could
either mean in relative terms (cost ineffective against IA-64), or
absolute terms (way too expensive to modify existing 32-bit designs).

Where is this "commented on comparing..."?... more secret missives which
might get someone sacked? I'd be very interested to see what the
comments amount to.

I'm actually a little surprised that anyone would care at Intel
anymore. Afterall, EM64T is already with us, and therefore Intel's x64
is a done deal. So coming clean about the distant past would be nothing
more than clearing the air.

Yousuf Khan