Posted on 11/09/2002 2:11:31 PM PST by JameRetief
Itanium and Opteron contrasted
Letter What are the options for Intel?
By : Saturday 09 November 2002, 10:04
This appears to show both processors passing the current fastest server chip by a significant lead. However, for x86 code, Opteron looks like it will hit the market as fast as the fastest x86 processor while Itanium 3 is a joke on x86 code.
It is too bad that there are not specs available on Yamhill because I have a very strong suspicion that it was killed, not because of self-competition, but because it is not as fast as Opteron. The Instructions Per Clock on Intel chips since 386 clearly show a steady decline (100% increase for 486, 90% for Pentium, 40% for Pentium Pro, 20% for Pentium II/III, and a decline of 10% for Pentium 4). I believe that Intel is unable to extend the x86 line to a Pentium 5. That would make Yamhill nothing more than a Pentium 4 with 64 bit instructions bolted on, whereas Opteron is clearly more than just an extended Athlon.
I am still wondering why more people are not asking the obvious question: Why is it that a company with the vast resources of Intel takes three generations to get its Itanium competitive while the much smaller AMD is able to get Opteron competitive on the first try?
I think part of the answer is obvious: either VLIW is not the magic bullet that Intel thought it was or Intel's implementation is compromised enough to limit its effectiveness. Personally, I think Intel used too many registers. If they had cut their register file in half, they could have saved 3 bits and reduced the opcode length to 40. Then instead of being 1 bit short on the 5 bit predicate field they would've had all the bits they needed plus two extra.
It's not just Itanium either, Xeon is also showing problems as indicated by its latest 2MB L3 cache. Notice that both Itanium and Xeon have large, 2MB L3 caches, while Opteron has none. The only reason for a large L3 cache is bus contention and this will fail when put up against an independent bus solution like Opteron's.
Opteron’s HyperTransport is even good enough for a Cray supercomputer. Add to that, the fact that Xeon lags behind Pentium 4 in speed, while Opteron matches Pentium 4.
There are other clues as well. Intel's x86 has been showing poor scaling (the increase in speed is always less than the increase in clock) since Pentium while the latest specs on Opteron show an astounding 1:1 scaling. I still don't know how AMD managed this on a (supposedly obsolete) 8086 derivation. Clearly, AMD has some real talent in its designers. Add to this the projected heat problems that Itanium will have with increasing clock and the underwhelming performance of Hyperthreading, (2.7% and 12% increase) and it looks like next year may be quite a fight. I'm guessing that at this point AMD is working hard on its main strategy of getting Opteron out the door. In contrast, I don't believe that Intel is simply working to get Itanium 3 out the door. I'm guessing that they are working very hard on contingency plans. Redress Yamhill in another suit or perhaps create a version of Itanium with better x86 performance?
Finally, the argument that Opteron's 64 bits are unnecessary is silly. I started with a processor that had 8 bit registers; I could have made the same argument for 16 and 32 bits. Clearly, people will buy new machines at some point and the new machines will be 64 bit. I'm guessing more of them will be Opteron than Itanium.
Brent Rehmel
Email address supplied
In 1904 the Studebaker company had 45,000 workers and was the largest manufacturer of transportation vehicles in the world. They had millions in the bank. Ford had 75 workers in a rented plant in Dearborn Michigan. Studebaker got beat by an Electric plant foreman named Henry Ford. The Dodge Bros. took a big hunk of stock in Ford for supplying $12,000 worth of engines to Ford on Credit. Later it cost FORD 15 million dollars to get that $12,000 dollars of stock back from the Dodge brothers. They used the 15 mill to start Dodge cars and Trucks.
It is pretty simple. If AMD makes a better chip than Intel they will beat Intel. That should tell you a better and cheaper operating system would beat MS.
GM beat Ford with a Chevie. If you could drive a Ford you could drive a Chevie. Runing windows does not mean you can run LINUX. The Chevie just was a tiny bit better car and looked nicer. It was a bit easier start. A person who could drive could drive either.
When GM was formed by a few men to compete with Ford, GM had 1/100 the money of Ford. Ford owned his own steel mills. His own coal mines to refine the steel. He even owned his own rubber plantations and iron ore mines. GM had to buy everything. Ford owned everthing. GM won.
Gates is not invincible. It would only take the following. An operating system that was as much an MS clone as the Chevie was a Ford clone. No training needed and they both run the same software. If a chevie had not run on a highway built for fords... it would not have made it.
The competitor needs to sell for a bit less and be just a bit better. That is all it takes. NO ONE has tried to take MS on. That is not Bill Gates fault. But Linux on a PC and UNIX on an Apple are not Chevies designed to compete with a Ford.
They are attempts to compete by people who don't know how it is done.
Unfortunately this has been tried before, without success. The problem is that Microsoft produces both the operating system and the applications. If you produce an operating system that runs the Microsoft applications, e.g. "both run the same software" in your words -- why Microsoft just changes the application software so that it no longer runs on your operating system. And they can always do this. So that approach doesn't work. And because big businesses want compatible applications (the Microsoft Office suite), they will buy the MS operating system (because they have little choice.)
Linux does provide an alternative. And for certain businesses it is a viable alternative. But they don't do it by running the MS office suite. (Lindows uses WINE to run a limited set of MS apps - and it does it poorly - because Microsoft has no intention of every making it easy.) What they do is offer applications that can read in and write out MS office program file formats - thus StarOffice Writer can read in, edit, and write out Word, Excel, and Powerpoint files. Be sure that MS will ensure that the file formats change with the next MS office release.
In the long run, focusing on the ability to manipulate the MS office file formats may be sufficient to dislodge MS. And the cost of Linux is so low, its performance is so good, and its quality is so high, that it pretty much ensures that MS will never successfully capture the server market.
-----------------------------
By the time Micrsoft had finished DOS 3.x it had $300,000,000 invested in development of operating systems. Operating systems do not spring up for free. Unless somebody can steal Microsoft programming and personnel, it will cost close to a billion to compete with Microsoft.
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.