Prescott may have hidden 64-bit functionality

[GP2]

Some more sheer idle speculation. Bear with me, this is a bit long, some background detail is required to develop the line of argument:


1) The number of Athlon 64s that will actually be available for sale over the next year will be rather low, most likely due to poor yields:

AMD Athlon 64 yields don’t look too brilliant
http://www.theinquirer.net/?article=11758

The Matrix of AMD processor chip sales
http://www.theinquirer.net/?article=11769


2) The Athlon 64's large size makes it uneconomical to manufacture at the present time. They need to go to 90 nm, but can't yet do so:

AMD unveils details of its 64-bit chip
http://news.com.com/2100-1006-5080640.html

Athlon's large size presents another challenge to AMD--it is comparatively expensive to produce. At 195 square millimeters, it's larger than Intel's current Pentium chips, which occupy 150 square millimeters. Intel has said its next-generation Pentium, code-named Prescott, will occupy only 112 square millimeters.

Brookwood said AMD needs to create a smaller Athlon 64 over the next six to nine months if it wants to compete on a cost basis with Intel. "This is clearly not a good long-term solution for AMD as a desktop processor," Brookwood said. "They must shrink it."

AMD said the Athlon 64 and its server sibling, the Opteron, will be moved from 130 nanometers to a 90-nanometer manufacturing process in the first half of 2004.

3) Intel Prescott to be born on the 3rd of December
http://www.theinquirer.net/?article=11700


4) Intel is way ahead of AMD for 90-nm manufacturing, and it has more manufacturing capacity (more fabs).

Intel Prescott’s 90 nano apocalypse is upon us
http://www.theinquirer.net/?article=11775


5) Intel has a cross-license with AMD, so they can use the AMD64 instruction set.


6) Despite charges that Intel is intimidating manufacturers not to build systems with the AMD chip, one of the companies that is apparently resisting this intimidation and coming out with Athlon 64 systems is none other than HP -- Intel's partner on Itanium. Which seems a bit odd.

HP lets some Athlon64 specs out of the bag
http://news.com.com/2100-1006-1018579.html

AMD Taiwan intros A64s, accuses Intel of intimidation
http://www.theinquirer.net/?article=11732

THE THIRD launch in two days of AMD’s Athlon 64 processors is currently taking place in New York New York, Taipei, with support from Nvidia partners Via and Nvidia, but with precious little signs of major vendors brave enough to say they will use the chips.

Hector Ruiz, CEO of AMD, launched an attack on Intel claiming pressure from the chip giant forced third party vendors to chicken out.

The heftiest press release in the world claimed that 60 manufacturers and system builders will use the FX and the Athlon 64 chips. What do we have? HP [emphasis added], Fujitsu and Packard Bell, and a stream of smaller people

Put it all together and what do you get?

Speculation follows:

- Prescott really does contain hidden x86-64 instructions.
- Intel reveals this in December when Prescott is released, dooming Itanium.
- HP knows this is coming, which is why they're supporting Athlon 64 and AMD even though they're Intel's partner on Itanium and even though many other companies are "intimidated" by Intel not to support AMD. Perhaps a petty way of getting even with Intel for dumping them?
- With its 90-nm manufacturing process ready to go and its larger manufacturing capacity and probably better yields, Intel can manufacture more 64-bit chips than AMD can, and it can do so at lower cost. They can undercut AMD's prices and grab most of the market share for the 64-bit desktop market.

Everybody thought Intel was napping while pursuing their futile Itanium strategy. Instead, Intel is in a position to grab the dominant position in the very 64-bit desktop market that AMD pioneered and created. Intel wins again.

[Dresdenboy]

Then let's hope that Prescott won't be a paper launch like P4EE. Even Tom's Hardwareguide had to send their P4EE sample back to Intel because other (smaller) sites needed a CPU. And since Intel want's to sell the P4EE in december we have to ask: Why? Aren't the Prescotts available then?

I will take a look again at Hans de Vries' speculations on Prescott. Maybe he did some conclusions by scaling 130nm structures of Northwood down to 90nm (by a factor of 1.44). But the real scaling is 1.29 (I compared that by examining the sizes of some parts on the die which are comparable). But since there are even no 64bit integer registers - how would they implement a 64bit compatibility or at least addresses.

A more interesting fact would be, if Prescott is able to do a SSE2 mul or add operation at once and not in 2 halves going through the pipelined units. And if it could do that - would it have the available issue ports?

And most people say that if there are 64bit extensions already implemented, they could be disabled like HT was in Willamette. So we shouldn't count on any expectations but wait instead and look, what Prescott can do.

Regards,
DB

[GP2]

Why Intel's Prescott will use AMD64 extensions
http://www.theinquirer.net/?article=11781

This is just a short opinion piece, but they seem convinced that Microsoft twisted Intel's arm and forced the issue.

Of course, Intel loss of face could still turn out to be a big economic win if they churn out a 64-bit chip in larger numbers and at higher profit margin than AMD can, for the reasons speculated about in my prior post. Not to mention the higher clock speed... they could certainly end up eating AMD's lunch.

[nucleon]

Actually I think AMD is laughing all the way to the bank. Well depends how much they are getting from Intel for their 64bit code.

Side note: http://www.theinquirer.net/ is a good site isn't it?

Intel has licenced AMD's 64bit extensions, so as long as the agreement is based on how many units intel sells. AMD should be laughing all the way to the bank as MS have now forced intel to use AMD's 64bit code.

So AMD is only dependant on how much the market wants 64bit. It could be very irrelevent, if the market chooses intel over AMD. Market buys AMD - AMD wins, market buys intel AMD wins.



-- Craig

[GP2]

Originally posted by nucleon
Intel has licenced AMD's 64bit extensions, so as long as the agreement is based on how many units intel sells. AMD should be laughing all the way to the bank as MS have now forced intel to use AMD's 64bit code.

Yes but AMD had to license SSE2 from Intel, among other things.

I think there are various cross-licenses, so the main cost to Intel is simply their pride.

[nucleon]

Originally posted by GP2
Why Intel's Prescott will use AMD64 extensions
<snip>
Of course, Intel loss of face could still turn out to be a big economic win if they churn out a 64-bit chip in larger numbers and at higher profit margin than AMD can, for the reasons speculated about in my prior post. Not to mention the higher clock speed... <snip>

I think intel will find it harder to get the clock speeds with supporting AMD 64 extensions. The register file has had a huge increase 2-4x the IA-32 size. Registers can take up a fair bit of real estate on the CPU die - clock skew anyone. Who knows - 90nm may have solved those issues.

Intel has relied on pure clock speed to get it over the line. AMD is trying to get the most per clock cycle. I think intel may have some trouble with AMD's 64bit code.

But the above is pure speculation - so who knows what will happen. But it's fun from the sidelines. :)

-- Craig

[nucleon]

Originally posted by GP2
Yes but AMD had to license SSE2 from Intel, among other things.

I think there are various cross-licenses, so the main cost to Intel is simply their pride.

If intel got AMD's 64bit extensions for the price of SSE2 - that's a bargain. I'm sure if AMD knew MS wouldn't support intel's desktop 64bit instruction set, AMD would have charged through the roof to let intel get 64bit instruction set.

As events have unfolded, intel would have been up the creek without AMD's 64bit extensions. We would have had itanium code on the desktop. That would have been miserable for all the desktop apps.

-- Craig

[QuintLeo]

AMD and Intel have had several cross-licence deals going all the way back to the 8088 - IBM insisted on a second-source for that part, and Intel picked AMD to be that second source.

What a lot of folks forget is that AMD has been making microprocessors ABOUT as long as Intel has - AMD's 2900 series dates from the same timeframe as the 8080, or possibly the 8008.

Seems like every innovation one of them does, the other has the option to use or do a modify-and-use routine on, due to the licensing stuff.


As far as the current Athlon64/Opteron performance vs. 32 bit stuff - consider the 80386 to 80486 transition, then the 80486 to Pentium transition. The newer parts were NOT any faster than the "fastest of the old" untill the compilers and such caught up with the new optimizations....

[Dresdenboy]

Somehow it even looks like Intel is using the Inquirer to spread some info (and misinfo) to distract people from AMD. I think the best signs for their troubles are:

1) Prescott delayed to december

2) The release of a P4EE where only a few were available to reviewers - not being packaged as P4EE but in form of remarked Xeons. Tom's Hardwareguide wrote that they had to send their sample back to Intel that other sites and magazines could run benchmarks on it.

3) P4EE will be available in significant numbers around xmas - what about Prescott then? Maybe the main Prescott launch time will be Q1.

And the higher clockspeed of Prescott wouldn't help much if their would be some 64bit compatibility. Instead it would give the same picture like we see now with 32bit applications - a CPU with high clock frequency vs. one with lower clock frequency but higher IPC, lower mem latency and better scaleability for MP systems.

It will be an interesting time.

[Dresdenboy]

a P4 EE test sample as used by Tom's Hardwareguide:
http://www.informatik.uni-rostock.de...ges/cpu_p4.jpg

It simply doesn't look like a ready product..