Kare Lochson:This differentiates us from the mainframe technologies that do the same on price, and from all others in that we can run a standard operating system. No patches to the operating system are needed. It is Linux straight out of the box. All the changes are to the boot loader only. Then we can run an absolutely standard single image operating system. Theoretically it can scale to 4000 nodes. But that is not realistic to do in practice. But it is very scalable. And we can also provide addressing up to the full addressing range of the AMD processor: up to 256 Tbyte. It is fantastic to work with, because of the really enthusiastic reception from research and computer people.
Primeur magazine:Who are the typical customers?
Kare Lochson:Of course this can be used for many kinds of computing, but we have focused on the HPC industry. As a small company coming with something new, we experience that the creative minds are in HPC.
Primeur magazine:And are they interested in HPC? Because 4000 nodes is modest. The conference people are talking about Exascale, and they are not doing it for less then a billion cores.
Kare Lochson:There is a lot about HPC that is on a small scale. And also the Exascale system will not be a single image system. But you can use one technique that is what I call fat-nodes. You can do a 100, perhaps 200 nodes and you cluster that into a large system.
But our main focus in on the somewhat smaller systems.
Jon Snilsberg:And we believe there is an underserved middle-market of HPC. We are talking about the big supercomputers, and we see the many core, single node systems and there is something in between there that the market really needs. And I think the price-point of the shared memory system of our competitors is one of the inhibitors of the market to take off.
So there are existing markets that will play to begin with, and then we will open up a bigger market in the middle.
Primeur magazine:Do you provide complete systems?
Jon Snilsberg:We provide add-on cards. Our sales model is based on not competing with our customers. We want to sell on the component level and we want to enable the whole HPC community to work with this technology and build different kinds of systems, because it is very flexible.
Primeur magazine:And where are you today?
Kare Lochson:We are just running some prototype demo at the moment. We start shipping this Autumn.
Jon Snilsberg:But we have all the features of the final product. We have used an FPGA for the prototype, and have mirco-coded every functionality of the chip. The systems that are running right now have exactly the same features as the product that will be the final product but with less performance. So we are very confident that when the chip appears this Fall it will work.
Primeur magazine:Can you mention some typical customers?
Jon Snilsberg:There are two pilot projects we are currently working on and I think they are representative of the markets that we are going to address. One is a university: the University of Oslo. They are building a test system right now with our prototype cards of 16 nodes and interconnects. And their customer is an oil company: Statoil in Norway who are building a 32 node test system. And this is considered as an interesting technology for the whole oil industry due to the need of simulation of oil reserves.
Kare Lochson:The user at the university is a general university data centre compute power provider. In the oil industry they have several large memory applications. Data from oil fields are quite extensive. So that is one area that we are focusing on. That is our large memory and you need to have a large data set in memory. We can provide that.
Jon Snilsberg:We want to build this company brick on brick and stone on stone. Because there is a an enormous amount of opportunity out there. But we want to make it work. We want to make our customers happy as we go along, and make sure step by step.