POST OF THE DAY
Advanced Micro Devices
Re: Intel Roadmap in Shambles!

Format for Printing

Format for printing

Request Reprints

Reuse/Reprint

By eachus
May 10, 2004

Posts selected for this feature rarely stand alone. They are usually a part of an ongoing thread, and are out of context when presented here. The material should be read in that light. How are these posts selected? Click here to find out and nominate a post yourself!

The cancellation of the 90 nm Pentium 4 line of chips provides AMD with a unique opportunity to increase its market share and penetration into the enterprise market. Intel will be on hold for several months, while they create a new roadmap/chip to take on the mantle of flagship.

The Intel roadmap is not in shambles. How long have I been saying, here and on the MF Intel board that Intel will have a good year, because of Pentium-M. What they have to/had to do was start selling Banias and Dothan into the desktop and server segments.

That decision has apparently been made--finally. I expect to see Dothan desktop models at the Dothan intro on Monday, and I expect Intel to talk up Dothan as having "almost" the performance of Prescott, but without the scorching heat. In practice, everyone here should by now know that Intel doesn't make its huge profits on leading edge CPUs, it is the sales in the sweet spot that determines profits, and the sales below that provide most of the volume.

Let's say that Dothan has roughly the IPC of Hammer chips--on 32-bit code only of course. So we can roughly equate the 2.0 chip to the current 3.0 Prescott and Northwood, and the 1.7 and 1.8 GHz chips should bracket the 2.6 and 2.66 GHz Northwood CPUs. Prescott has a 112 mm� (at 90 nm of course), Banias is 82.79 mm� (0.13 �m) and Dothan
87 mm�. Banias can easily be made 6-up, Dothan is approaching the limit for 6-up mask sets, but I expect that it is 6-up as well. Prescott is at best 4-up, as are current Northwoods at 131 mm�. So Banias or Dothan is cheaper to manufacture than Prescott. Why has Intel been trying to force Prescott down the throats of non-speedfreaks?

So Intel's current competition to AMD will be Dothan. What about the future? There are two paths going forward. One is apparently a dual-core version of Dothan called Jonah. The other path is Merom and Conroe, which will be laptop and desktop versions of the same core, with some of the current desktop chip features added in. I suspect that AMD64 support will be "in there", along with Hyperthreading and SSE3, although the Hyperthreading may be desktop only. My reason for concluding this is that a year ago there was a different follow-on to Dothan called Gilo. My guess is that Merom had the 64-bit support, and Gilo didn't. Now Gilo is replaced by Jonah, which will be Intel's first dual-core consumer part. (I also expect that if Intel knows what it good for it, Merom and Conroe at the latest will include NX bit support.)

What about servers? There is obviously a wide-open opportunity for AMD there. But that is nothing new, before Intel announced that Nocona would support AMD64, the server market was pretty much wide open for AMD above the low-end. I don't see that as having changed. If 32-bits and one CPU is sufficient, Dothan will make a great server CPU. I don't know how much work it will take to come up with a dual and or quad CPU version of Dothan, but given that Jonah will combine two Dothan cores on one chip, I suspect that two Dothans with one Northbridge will work now. Killing Potomac and Jayhawk tends to confirm that.

So Intel will soon have Nocona as a stake in the 64-bit server ground, and of course, nothing here has changed the (high-end) Itanium2 track. But everything I have seen indicates you have to be a die-hard Intel fanatic to choose Nocona over Opterons. (Opterons in 64-bit mode use more power--but are faster. Nocona slows down when you use 64-bit addressing, and I can't imagine the power demand goes down.)

So I think that where the server market will be in six months is:

Low-end, single CPU: Dothan for cooler operation, Opteron/Athlon64 if you need 64-bits. Athlon64-M if you need both.

Low-end, dual CPU: Nocona if power and 32-bits only are not an issue, Opteron 2xx otherwise.

Mid-range: Forget Intel, the hard choice will be between different Opteron architectures, and OS options. Should you stack several IBM 325's, or get one four-way HP ProLiant DL585? Is Solaris better for your requirements? Or Windows 2003, or Linux? Of course, there will still be "legacy" systems around, so Sun and HP will be selling UltraSPARC and PA-RISC systems into this market as well.

By the end of this year, I expect that Opteron will dominate this market, at least in units. Note that it may be tough to figure out if that is true. Sun will be selling UltraSPARC, Opteron, and Xeon boxes. HP will be selling Opteron, PA-RISC, Itanium2, and Xeons. IBM will be selling (low-end) Power, Opteron, Xeon, and even z-series into this market segment. I could go on, but you have to get into the third tier before you encounter companies selling only one architecture.

High-end: The choice will depend on what you are doing. UltraSPARC IV will still be viable for a little while, but only as an upgrade for current Solaris systems. Same for PA-RISC. Other than that, the choice will be between Opteron, Itanium2, and IBM Power. The choices will be made in many cases by benchmarking, and choosing the least expensive system that meets the customer's needs.

I don't expect the OctigaBay to be officially available six months from now, but some users may have beta systems. If the (now-Cray) system works out as I expect, it will be a huge change in price-performance for database applications. Then drop in dual-core CPUs and it will dominate the high-end database server market.


Become a Complete Fool
Join the best community on the web! Becoming a full member of the Fool Community is easy, takes just a minute, and is very inexpensive.