Wednesday, September 12, 2007

AMD, Quad Core Opteron, and Korea

This has been an exciting week so far for those of us following the Processor Wars. Intel announced it is making more money than it thought it would. Korea announced (See Infoworld article) that it thinks Intel used illegal methods to make its dough, joining Japan and the European Union in that allegation. AMD released its genuine quad-core Opteron processors, to be distinguished from Intels dual-dual core Xeon processors. Preliminary reviews have already started emerging for the new AMD processors; they look amazing. Since this blog is primarily for investors, I'll be converting some tech-speak to a semblance of why this matters in reality.

First, as far as I know, all AMD executives and employees are alive and well, so the boys at Intel have been too busy trying to catch up with AMD advanced technology to watch the Sopranos. Their unethical business methods will doubtless be considered by Intel investors to be mere sharp business practices. Essentially what Intel is accused of by Japan, Europe, and Korea is limiting AMD to very small markets by illegally discouraging dual sourcing of processor chips. Intel says it never did this, and if it did it stopped doing it. And my understanding is that nothing Intel did, would, in itself, be illegal. You can price your products however you want. It is the context that matters: pricing schemes can become illegal when they are designed to create a monopoly. Intel's main defense is look, we lost market share lately, we can't be a monopoly. But they lost market share after two changes: AMD introduced 64 bit x86 processors when Intel tried to force its clients to switch to Itanium processors in order to get 64 bit, and Intel was informed by regulatory agencies that it was being investigated.

In 2006 Intel was on the ropes and responded by taking advantage of its capital position. It started a pricing war that benefited computer makers (Dell, HP, etc.) and consumers but that drove AMD to the wall. In the later half of 2006 Intel released a new processor design that showed some improvement on older designs. This happened while AMD was trying to absorb its acquisition of ATI, which simply put it in position to get its ass kicked by NVidia and Intel at the same time.

Today AMD's stock price is below its 2002 level, even though it is a much larger company now. 2007 revenues look to be roughly double those in 2002. Earnings are bad, in fact negative, because of the pricing war with Intel and the cost of the ATI acquisition.

But that is the past; the future depends on industry adoption of Barcelona technology, particularly the quad core Opteron processor for servers. Unlike when Opteron was introduced (no major players joined in the introduction), IBM, Dell, and HP all are ready with quad-core Opteron servers. These processor chips can also be plugged into AMD-based servers that have been selling in 2007 because they are pin-compatible with the dual-core Opterons. I doubt many companies will be pulling out dual-cores to put in quad-cores, but it means the motherboard designs were ready to go well in advance of chip availability. So it is off to the races.

Now we will enter the benchmark propaganda wars. Intel fan boys will claim that Barcelona sucks. AMD fan boys will say it is way superior. Both will show benchmarks to prove their point. How can that be? Well, there are two big issues (and a swarm of small ones) with benchmarks. One is optimization, the other is specialization, in this case floating point versus integer. For people who actually buy servers, who actually want the best bang for the buck, who are not locked into Intel, they are going to need to run tests on both platforms to see what works.

There is agreement that AMD has really good floating point capabilities. So that means scientific applications are going to lean to AMD. But there is a claim from the Intel camp that servers mostly just serve up data; Intel's floating point capacity will do just fine. Ten years ago I would have agreed with that, but things have changed. Visual is in. Visual requires floating point. Fewer and fewer servers are being devoted to just plain accounting or data serving.

To me the new AMD architecture looks superior for a couple of reasons, but I know that superior looking architecture does not always translate to real performance. First, the AMD memory controller is still on the server chip. Intel likes to run benchmarks that go to memory infrequently enough that its off-chip memory controllers don't appear to be a bottleneck (this is not new to Barcelona, but should become more of an issue with four cores needing to be fed memory). AMD's design will be a huge advantage in high end servers with as many as 8 chips, meaning 32 cores.

While both Intel and AMD have introduced technology to make virtualization easier, and both have some merit, I suspect AMD's architecture will work better in datacenters. If you are running a bunch of virtual machines on a multiprocessor, multicore server you also want to have multiple network connections (typically 10 Gigabit Ethernet connections), one for each virtual server. AMD's new design scales out nicely; Intel's current design, the Xeon 7300, does not.

Again, I have not personally tested the new AMD or latest Intel machines. I am analyzing what specialists are publishing.

I would say that AMD investors should take heart; Barcelona looks good. It will look even better as AMD ramps up processing speed (the first out the gate quad-cores have slower clock cycles than most of us expected). However, as long as Intel is willing to undercut AMD on pricing, you have to consider AMD a long-term investment, and a risky one at that. I own the stock; I want to see what its worth when AMD market share hits the 51% tipping point. That may never come; at best it is years away.

See also my AMD page

No comments:

Post a Comment