Although a x86 processor, Larrabee is a different undertaking than Intel’s other chip efforts, offering a multi-purpose many-core unit that can handle graphics functions as well as more broader computing ones. Performance graphs and more detail can be found in Intel’s official 2008 Siggraph paper on Larrabee, entitld ‘Larrabee: A Many-Core x86 Architecture for Visual Computing’ can be found here. The first versions of Larrabee are expected to arrive later this year with the first commercial products powered by the chip, including video cards, pegged to roll out by early 2010.
What will the products that emerge out of the Larrabee project provide that is not already on the market?
One of the key things that Larrabee architecture gives games companies is a tremendous amount of additional flexibility and programmability – things that they’ve been asking for many years. And that provides them with the ability to do things and innovate in very different ways than they’ve been able to before. First and foremost is going to be a very performing graphics and throughput architecture, and it’ll provide all of that through direct X and Open GL. The majority of games and usage of Larrabee architecture will be through that. And there will be a smaller per centage of games developers that are going to go out and innovate on top of what we provided through what we call the Larrabee native interface.
Nvidia and ATi currently represent something in the region of 98 per cent of the GPU market. How will you convince the industry you provide a better alternative?
I think a big part of that is that we’ve been working with a lot of game developers. All of the input and design of Larrabee comes with the involvement of various software firms in the industry. It’s very much driven by feedback form the industry, telling us the types of thing they would like to do, and how can they do them. We’ve tried to incorporate that in.
That 98 per cent is what Larrabee is targeted at. And we are not new to the graphics market either. We are the biggest graphics vendor on this planet, most of the graphics components out there are Intel integrated graphics. Obviously you’re right, if you look to the graphics discreet market it is them, but its not that we’re entirely new to this world.
Larrabee has the ability to be more than graphics. It’s designed for a throughput architecture. Now that doesn’t mean that it can do everything in the world, but there are many things that are applicable top Larrabee. Graphics is probably the predominant one in the over all industry if you will. We’ve talked with other ISVs, medical imaging sections, oil, energy and gas, general image processing, and the financial market as well. There is applicability of the Larrabee architecture to those segments.
How important will relationships with games developers be to the project?
Very important. We’ve had a very heavy involvement in the graphics market overall for quite some time. We’re being very open and customer orientated by listening to the ISVs, talking to them and asking them what it is Larrabee needs to do, with regards to the general architecture of it as a lot of what is behind Larrabee architecture relies on the software infrastructure of it, allows us to simply change capabilities very fast. We don’t necessarily need to rev the hardware architecture and literally change the software architecture. We wanted to change the software rasteriser, we wanted to change the pixel shading logic processing, we can do that in software, which gives us a much faster turnaround. A lot of that is being driven very closely with the software companies that we’re working with.
What elements of Larrabee can be utilised by games developers?
All of them. It really comes back to this core. In the simplest case games developers will simply be able to use Larrabee like a DirectX and an OpenGL card. Theoretically they shouldn’t have to do anything to utilize Larrabee as a DirectX or OpenGL card. And that probably will be the majority of the usage of Larrabee architecture for them. For some additional people who want to go about there and innovate we give them a very flexible interface that slows them to program all the way down to the metal.
And that flexibility is extremely powerful in that it can decide how they want to do rasterisation, decide how they want to do load balancing, how they want to do all different kinds of things. We’ve pulled out a couple of examples such as the regular Z-buffer as a way to do really powerful more realistic shadows. And we call out order independent transparency as another, which is something developers have been asking for a long time. But it’s very hard to do on current GPU hardware.
In what ways will programming for many-core processors change how games are developed?
I don’t think Intel could say all the different ways it is going to change. I think it would be arrogant of us to say we know exactly how it’s going to happen. We have some ideas and we share them with the games developers, but we spend a lot of time listening to what they think. We talk to them about the architecture and the software that we have, and then ask them a lot of questions like how would you use it and does it do the things that you want it to do. And so I think that we have some very good ideas, the software developers are going to come up with a lot of different ones. And I think the industry in general and those developers are learning what they can really do beyond what they’ve done in the past.
I think for software developers it’s one of the bigger inflection points, in terms of programming for gaming and graphics. Graphics used to be, before we had graphics accelerators, completely leaned on CPUs. And it was completely flexible – people decided they wanted to do rasterisation, they wanted to do ray tracing, they wanted to global illumination, but during that time all those graphics were not real time. The processors were too slow and the algorithms were so complex that you couldn’t do it in real time. With the push to get graphics towards real time came hardware accelerators. But in order to make this hardware accelerators work and get real time graphics they made a lot of concessions. They implemented a fixed function pipeline. So they said ‘ok rasterisation has to happen this way, texture look up has to happen this way, shading has to happen this way’. Since then the standards have evolved to add more and more levels of programmability, with pixel shading etc. But now we think Larrabee allows us to come full circle and have complete programmability, and still have all the processing power to do everything that we want to do.
Does Intel want to see Larrabee used in the next generation of games consoles?
It’s definitely something we would want to discuss with the console vendors, and hope that the architecture that we’re providing is something that is very compelling for them and be interesting.
Games developers already complain that multi-threading and exotic processors currently on the market put pressure on their teams; why should they care about Larrabee?
Larrabee is going to be a DirectX and OpenGL solution. That means that for the majority of developers, if their resources are already strained, should not be a significant challenge for them at all. This is like any other graphics solution. And it will be performance competitive. But there are a few leading game companies out there that are saying ‘we want to push the envelope, we want to push the industry further.’ And they’re the ones that are going to work and experiment more on the Larrabee native interface as they go.
In order to use Larrabee you don’t have to use this new native interface. You can continue to use it just like an OpenGL or DirectX solution. But if what that ability to go a little bit further to try things that you haven’t tried before, or that haven’t been possible before, that’s what he third interface provides for you.
Do you hope that developers see Larrabee as multi-function processor and not just a graphics one?
Yeah I think the Larrabee architecture will be seen in that way. That’s something that we’re looking at. Like I say, we’ve talked to developers beyond just the gaming and graphics segment to make sure this is largely a throughput architecture of which gaming and graphics is probably the largest priority of throughput segments, if you will.