Scott: As a company we are focusing on our XT line - our main product - and we are really pleased with how our Jaguar system at Oak Ridge was used during the last half year and where it sits at the TOP500 list (at number 2). If it would have been number one, we would have been be happy about that too, of course. But Cray and Oak Ridge choose not to rerun their Linpack benchmark, even though people looked at it and said, actually, if we wanted it and put some tuning work into it, we could get a higher number than the Roadrunner, the number 1. But it is not that important. We rather keep working on scientific problems on the machine. I think that is a good thing.

From a product perspective Cray introduced two new products that have gone on the market into different segments: the CX-1 and the XT5m which was just launched this March. The first customer was actually HLRZ here in Germany. Just earlier this week we signed a contract with Finnish Met: our fourth weather site here in Europe. We are excited about that. The centre has two identical systems: one for operation and one for research.

Primeur: What type of research will they do on it?

Scott: Weather and climate related application research. The second system can also be used as the fail-over system. Which is just something that most operational weather forecasting sites are doing. They have a system for primary operations work and another system that is used for other work. But should the first system be down, then it can take over the forecasting.

The performance of our biggest European installation at CSCS, the Swiss supercomputer centre, is now up to 140 Tflop/s. I think it is the fourth largest system in Europe. It is interesting, because they, as most of our large customers, have upgraded several times their system. They plan to upgrade to Istanbul, the next-generation 6-core Opteron processor.

Primeur: When you say "upgrade" you mean that the racks and the power supplies and everything stays the same?

Right: You can just change out the processors. One of the things that we always tried to do in our systems, is that you can upgrade chips, memory, processors and keep the rest of your system infrastructure the same. In fact next year we come out with our first big network upgrade since we introduced the XT line. The SeaStar interconnect that Cray uses, has been used since the XT-1 in late 2004. There have been some modest enhancements to that in SeaStar 2 and SeaStar 2.1., but today it is still the same basic interconnect. Next year we are going to introduce a new interconnect called Gemini. It will provide for global address space languages. It will also keep about the same bandwidth but it dramatically improves the messages rates: about 50 times increase in the number of messages that can be sent per second and about a four times reduction in latency. So that is a pretty major network upgrade.

Primeur: And the rest of the infrastructure does not change?

Scott: Everything can be left in place: you do not even need new cables. You take out the old network card and put in the new card. We try as much as possible to allow people to scale-out and scale-up and upgrade systems, and as I said, almost all of our large customers have upgraded, some of them multiple times: 3 or 4 times. Oak Ridge National Lab in the US has upgraded about 5 different times. They started around 3 Tlop/s and they have grown to 1 Pflop/s.

Primeur: So many upgrades, is this not difficult to do in Europe? Because in large public organisations you have these public procurements. And of course if they want to upgrade a Cray system, the only one who can do that is Cray.

Scott: Sometimes people have to go for public procurement. And sometimes the original contracts have been constructed so you can upgrade mid-life. Hector in the UK, for instance, was done that way.

The big news product-wise is starting next month: we are starting to upgrade processors to the Istanbul 6-core processor. Next year we will get a new computer blade that moves to DDR3 memory which gives you a 2-4 increase in memory bandwidth and an increase in the number of cores.

We are continuing to work on Cascade which is our DARPA-HPCS programme system. That will be coming out in 2012. It will have, for the first time since the introduction of the the XT line, a whole new infrastructure: New cabinets, new cooling designs, a hybrid optical interconnect. There will be electrical links within a couple of cabinets and optical links everywhere else. And we will also develop by that time Intel processors. Now we sell Intel processors already in the CX1, the desk side system. With the next generation network that we are doing for Cascade we will have a PCI-Express based processor interface and we will also come out with an Intel based board.

Primeur: What is the largest Intel system that you have now?

Scott: The biggest one is in the CX-1, the desk side system. We also have some clusters we did for customers. But for our main product line it is AMD until we get to 2012.

So from a product perspective right now the interesting things are the CX1 and the XT5. The CX1 is the desk side system. The intent is to bring high-performance computing to a broad range of scientists and engineers. So it is available with Linux, but we are also partnering with Microsoft so it is available with HPC server as well, and we had a lot of new interest from people that traditionally have been using their desktop systems and lap tops to do their computational work and they do not really want to learn how to use a different environment. They want a familiar way of interfacing with machines and get more computing performance. It is also designed to fit into office environments: standard office power, and noise cancellation for instance. We just started selling that late last year. We have got something like 20+ different channel partners. Because Cray's sales force is relatively small and focused on high contact customer relation ships, they cannot sell this small box in large quantities.

Primeur: Yes, and if you look to companies like SGI and Sun, perhaps it is good to be not too broad in the things you do.

Scott: As a company, I almost feel guilty talking to some of my old friends that are not so fortunate because of the global recession. We had the best year ever as Cray in 2008. We paid off most of our debts. We grew 40% in business. And obviously the Jaguar system was a big success too.

Primeur: Do the long term HPCS contracts with the US government help with that?

Scott: Yes, we are more aggressive with the designs and we are able to invest a lot more in those areas than we otherwise would have been able to do so. We also worked more in software engineering: user-interfaces, compilers and libraries programming tools. We will continue to work on that contract through 2011. We have got a new organisation of custom engineering at Cray. We formed three business units to focus on. The personal systems group is focusing on the CX1, the scalable systems business unit is focusing on the main Cray MPP line including the XT5, and we formed another unit, custom engineering which is software design and we have been also doing a lot of custom storage and I/O for the systems that we sell, because people want different capabilities and set-ups in their I/O system. So we are getting involved in that as well. We are also getting leverage forth and back between the system groups and the custom engineering group.