Why chip speed is old news

If you think clock speed is the most important measure of a processor, IBM's Bernie Meyerson wants you to reconsider. Meyerson, who heads research and development efforts for Big Blue's semiconductor group, says processor chip speed is old news. Instead, he's focusing IBM's work on other areas, such as on-the-fly reconfiguration that will let chips automatically adapt to different jobs. He's also proposing a new, more open way for IBM to interact with its customers.

The open approach will involve IBM disclosing more technical information than before about its chips, and it will also mean that the company will rely more heavily on the hardware and software development skills of third parties. But if the approach is successful, it will enable IBM to create a much bigger ecosystem around its Power processors. It also may boost prospects for the company's chip operations. CNET News.com recently spoke with Meyerson.

Q: IBM is attempting to make the Power processor architecture more open. What exactly does that mean?
A: What IBM is doing here is addressing a fundamental shift in the landscape of technology, making changes in Power, making changes in our strategy. Over the last one or two years, we have begun to see a significant shift in what I call the trajectory of basic technology. Basic technology simply is not driving performance at the rate it did previously.

Does that mean performance stops?
It means that the rate of performance enhancement is becoming impacted. We simply made transistors so small that as we continue to attempt to make them smaller, it requires a huge amount of innovation to get added performance. If you were getting an improvement in technology of perhaps 30 percent a year, and suddenly that rate drops to half, the rate at which your systems improve would also drop by that amount. We revealed a strategy for the coming generations (of IBM Power) which is very, very different from the typical strategy in this industry. We applied techniques we learned through all the years of working in solutions to the area of systems and the underlying hardware and software.

Just so that we are clear, what do you mean by "solutions"?
The end result that people care about. When a system is purchased, and people are looking at transaction processing capabilities, that is an end result. They are not looking at whether the clock frequency of the microprocessor is 8 percent higher. Where you might previously have announced simply a higher clock frequency, that, frankly, is not going to be the major driver of performance. What we have done is to say, "We will allow the optimization of all aspects of the system simultaneously," which is what I refer to as holistic design.

That's a term that gets thrown around a lot in the tech world. What's your definition?
Holistic design means that you don't only worry about the bits and molecules that make up individual transistors. You also worry about the way in which they have been carved into a circuit. You worry beyond that about the way those circuits have been assembled into an architecture that winds up on a chip. You then continue right up the stack to how they are packaged, how they fit in the system and, even more importantly, how they interact with the system software.

Power consumption is becoming a critical factor, limiting performance.

Power consumption is becoming a critical factor, limiting performance. You have to have a means by which you proactively and holistically address that by allowing your actual system to command changes in the chip to optimize the utilization of energy. It is really an extraordinary event. I am hoping that people really understand the sort of discontinuity we are talking about here.

What would be practical?
We are even building in the capability for the chip to physically morph, if required. For instance, you spot an excessive number of fails occurring in the memory--we have techniques in software that recognize those errors. But if it turns out that for whatever reason, one segment of the chip drives an extreme amount of correction, one can easily envision the system autonomically issuing a command to remove that segment. We are talking about moving off to an entirely different plane. This is very different from, "Here is the chip, here is what it does, and that is the end of it."

At the same time, you are proposing a more interactive way of developing electronic devices, whether large like a supercomputer or small like a handheld.
This is basically going by the premise that I have sort of lived by for a long time, which is: The world is full of incredibly bright people. What happens when we say, look, here is the Power architecture, here is the Power core, here is how you hook up and work with it, here are the tools you need to do so?

If you have a designer out there and this designer is the brightest person out there but has never been able to engage, because we had closed systems--when you throw open a system, and you make it readily available, this person says, "Gee, I can build a core that operates and communicates on behalf of this chip. I can list it in the library as being fully compatible with the Power architecture, which means that anybody who builds a Power chip could license this (intellectual property) from me for some amount of money."

Then, when companies are building (system-on-chip processors), they can actually assemble a chip where they use our core, somebody else's (input-output system) and somebody else's memory controller, because they have all been built to a common standard.

The common standard is the Power Architecture?
Exactly. Build around the common architecture so that you have all the hooks and tools. But it requires you to be open--that is the key.

That must mean publishing lots of technical data for people to use. Are you saying you are going to make it easier for them to do that?
We are going to make it automatic. You could literally go out on the Web and download a design kit.

You literally go out on the Web and acquire this essentially for nothing.

We are talking about giving a level of data, essentially openness, that has never been available before and literally making it free. You literally go out on the Web and acquire this essentially for nothing.

In terms of chip design, rather than spending years, many man-years, of effort developing some kind of unusually efficient communications chip, you can go out on the Web and acquire it from somebody who is part of an ecosystem that has all of these intellectual-property elements already developed for it.

How does this make a chip design more open? I can see where it makes it easier for people to work with IBM. But IBM still controls the processor architecture.
This is as close as you can get without corrupting the ecosystem, which has to be stable for people to participate in it. That means that you have to define what I will call, for lack of a better word, interfaces. Unless everybody agrees how things talk to one another, you cannot develop something to work with what I am developing, because we have not agreed how we will talk to each other.

The architecture, therefore, must be protected, because that is what establishes how you communicate. So, yes, we have to lock down the architecture so that it cannot be randomly disrupted. Otherwise, imagine that somebody came along and altered the core architecture along with the core instruction set. It would invalidate all the prior intellectual property and destroy the ecosystem.

We have a unique situation we have to deal with. That does not mean, by the way, that we are not looking, for instance, at some sort of governance model. But it is very important that we protect the partnerships we are forming here, because there are people out there doing this sort of work, and the only way to make sure that we protect those partnerships is to absolutely ensure the integrity of the instruction set to which the people are designing their product. There is that constraint, but that is actually not a constraint for the purposes of hiding something. Rather, you simply cannot create an ecosystem without stability.

Is there a way that IBM will incorporate feedback? One thing about openness is that people will ask for different things. Will you take the top few ideas and actually build them into the architecture, if it makes sense?
That is exactly what I was referring to earlier. We are looking at the issue of putting together a governance body that will address exactly those questions. What we did was we put together a Web site at which people can actually download all of the tools and access the necessary hardware to begin the process of working in this area.

I did talk about governance at an earlier conference.

To believe that you can literally do everything better than everyone else is fundamentally foolish.

I was describing the potential development of what I will call a marketplace. We have no plans ourselves to go and create it. We do not even want to create the impression that we are trying to control it. We are not. We are really trying to be very much in the opposite direction: This thing is completely open and self-assembles.

What's in it for IBM?
It is just a fantastically powerful means to make progress. Remember, there are things that we do better than anyone else on this planet. I truly believe that we have the most innovative technologists in the world at the level of system design, software design. However, to believe that you can literally do everything better than everyone else is fundamentally foolish. There is a tremendous benefit to opening up an ecosystem and saying, "Bring me the best of the best, and I will incorporate it."

If you believe that you have a compelling value equation in a large part of the market space, then what you are missing, essentially, are ancillary pieces of it where people could bring value. When they bring value, everybody benefits.

Does it help IBM to sell more chips, get more business for the foundry service, et cetera?
I would characterize it as essentially a boost across the entire spectrum of the businesses we address. At the one extreme, the hardware side, this could clearly be a significant benefit. However, if you are at the other extreme, look at our services business--it would absolutely have the same benefit.

Does it help IBM to get into new markets where people are looking for lower-cost computers?
What this does is basically enable you to build full, custom solutions, which is to say highly optimized solutions. It will obviously open up markets that cannot be addressed through standard chips today.

It sounds as if this process will apply to future chips.
Oh, no. Power5 actually rolls out a lot of these capabilities. It is dynamically reconfigurable and has tremendous capabilities. But what I was referring to as being in the future is where the chip will autonomically morph. That is something that will take a couple of years to actually roll out.