RETAIL

By Chris O'Neal - HP always has a strong presence at the SC shows, so Supercomputing Online sat down with Steven Joachims, Marketing Manager, High Performance Computing, Technical Computing Division at HP. Supercomputing Online: What's HP highlighting at SC2002? Joachims: We’re highlighting our high performance computing solutions in particular around Itanium 2. We have been gathering a lot of performance data. There are a lot of announcements that have gone out about ISV partners who are aligning behind Itanium 2. Performance is excellent. It is a world leader in many cases. It is always at or near the top on any ISV application and made many performance measurements records. So Itanium 2 is a big promo for us this year. And we’re highlighting our scalability technology whether you are talking clusters or Grid computing technology. Everything in our booth is Grid connected. We’re demonstrating HP-UX Itanium 2 high performance compute cluster. Linux Itanium 2 high performance compute cluster with record-setting performance in a cost effective, high bandwidth, low latency supercomputing system. We have some hot new storage devices that are being highlighted from StorageWorks. HP StorageWorks Enterprise Virtual Array has record breaking performance with high capacity and high availability. We have got Itanium clusters that are running all over the show with a myriad of different high performance interconnects. Supercomputing Online: Please tell us about HP’s high performance interconnect strategy. Joachims: Our strategy is a little bit different from what you will find from either Sun or IBM. One of the first things we want to promote is the fact that we are adhering to open standards. We are working with partners to deliver the best-of-breed high performance computing networking technologies that you will find in the industry. Myricom and Quadrics are two of the top that we support in our product line today. In fact, one of the biggest installations, # 61 on the Top 500 list, is PNNL. They are running Itanium 2 high performance computing cluster from HP. Today, it has 256 processors and it’s going to be upgraded to 1900 processors with a peak performance of over 11 teraflops. There is an impressive demonstration on the performance they deliver. They have benchmarking Itanium against Sun, IBM, Cray and NEC. And they have got massive ISV applications and NWChem is one of them. And the Itanium system that we delivered is beating everybody. There isn’t anybody that comes close to it. IBM isn’t real happy about the performance because they got their best performance POWER4 system running the same application and it’s just a little bit behind Itanium. Sun doesn’t even come close. The Reality is that the UltraSPARC isn’t in the game in most ISV applications in high performance computing. When it comes to running ISV application like MSC.Nastran, it is one of the most popular in engineering and manufacturing. HP has with Itanium 2, PA-RISC, Alpha processors in our product family, some of the highest performance you can find on those ISV applications. So we deliver performance. When you talk about Interconnects the key question is when you take a performance of running on one system and scale that up to a cluster configuration to get supercomputing kind of performance by scaling applications up. PNNL is a great example of scaling to very high levels of performance. And you can see a lot of other vendors trailing off. The more processors you throw at a problem the less efficiency you find. The best-of-breed are the applications that many of our customers run on Itanium systems. We have all kinds of advanced interconnect R&D projects in house. We have a best-of-breed Interconnect technology partnership with Quadrics. ELAN 3 from Quadrics is highest performing interconnect that you will find in the industry. ELAN 4 is just on the horizon. It has some interesting performance characteristics including bandwidth and latency numbers that make a big step forward. They have some cool features like something that you will not find on other interconnects. You have got 2 systems that need to communicate; ELAN 4 has the ability to put information from one system into the memory of another system without interrupting it. Most interconnect technologies require processors to get involved this side and has to be interrupted and stop doing useful work and work on communication. So we are going to stick with that technology for the high end buyer who wants the absolute best performance. Compaq has been working with Quadrics for years. And they delivered some of the highest performance scalable system in the world in conjunction with Quadrics. I’ve been working with Quadrics for about year now. And our customers are very pleased with the stability and the performance that they get. One of the nice things is that when you buy a system from us, you’ve got a choice in Interconnects that you want to deploy in the system. Supercomputing Online: What do you think about 10 Gigabit Ethernet? Joachims: I think Ethernet popularity is on the rise. It has a very low cost and it works with everything and all the drivers. If you are running a heterogeneous environment, everybody supports it. It’s the lowest common denominator. And the performance isn’t bad. Now the ISVs have gotten better and better at parallelizing their applications. So the effects of latency and bandwidth are mitigated in the interconnect you choose. I’ve been told quite a few times by our application engineers. If you have a large system with thousands of CPUs and did nothing but change interconnects in that system, then the results from your applications could be different. Basically if you’re a running a big simulation like a fluid dynamic code that shows wind turbulence over an airplane wing. If you ran that model and changed interconnects and ran it again, you could get different results. The picture in model will be different. For that reason, large installations are looking for a commitment from vendor to pull things together and make sure that everything works. When you build your own computer you’re not certain what you’re going to find when you start running those applications. Supercomputing Online: What is your strategy at the new HP? Joachims: One of the key messages that we are delivering at this show is that HP strategy for high performance technical computing is as follows: We have built a business unit to focus on high performance technical computing. We have actually elevated the visibility and the level of leadership for high performance technical computing with the merger. Seven months ago when we rolled out the new HP, one of the things that we announced is that we have a high performance technical computing division within the company. The Vice President and General Manager of the High Performance Technical Computing Division for HP is Winston Prather. He has been meeting with customers and market analysts and so forth of late and doing a fantastic job. Customers in this market segment are particularly sensitive to that because there has been a lot of vendors that have come and go. A lot of vendors only focused on technical computing. What they found was it's difficult to amortize their costs. R&D is costly. The market segment is big, yet small enough that it doesn’t drive enough R&D. Vector supercomputing is a very expensive proposition to come up with. And Cray and NEC are still in that game. A lot of our business is still based on Alpha and PA-RISC technology. Itanium 2 was just rolled out in July. So that is a new technology. And of course we haven’t changed our entire portfolio over yet in our install base. We are shipping Itanium 2 systems in volume. For example, Airbus one of our large PA-RISC customers is running 75% of the computer related engineering resource is running on HP-UX and PA-RISC. Nigel Barry is their IT Architect and he is presenting at the conference. He wants to upgrade many of those systems to Itanium at some point in the future. His biggest problem is that he has numerous applications. Many customers, not just Airbus, have all these legacy applications and whether crashing cars or designing airplanes, you’re enthusiastic about being able to upgrade performance doing in box, low cost upgrades. You can just pull out boards and put in new ones. Reboot and you’re running a lot faster. But you have got the practicalities of thousands of users and you basically build your plans in over time. You break things in and make sure everything is working to the Nth degree. You’re making sure your in house applications run. So you have got to be pragmatic about moving forward with new technologies. Research centers tend to be a little less risk adverse, but still you’ve got to be concerned about the services they are providing to large users. In July, we introduced brand new Itanium 2 products. And we are beginning to ship a lot of those to the high performance computing marketplace. We have had a number of announcements have gone out about new customers. And HP-UX and TRU 64 based systems are still an enormous part of our business. The next-generation AlphaServer systems product is code named Marvel. There is a lot of public information about that. We have been very open about it and even before public announcement and shipping volumes. We have got a number of early systems that have been built and delivered to key ISV porting centers and early adopters. So we are bringing up the applications. And the thing that stands out about that product is the memory bandwidth. That is one of the important performance characteristics for high performance computing applications. Actually, the three areas that most HPC buyers look at: 1. CPU performance – You’ve got raw speed. Itanium is best of breed. Marvel is going the highest performance processor for memory bound applications. We haven’t announced the availability of the product yet. But that is on the horizon. 2. Memory bandwidth - How fast can you feed the CPU. Itanium 2 is impressive. We have STREAM benchmark that shows it sustaining 4 gigabytes/sec, which is 1 ½ to 2 times better than competitors products. 3. Interconnects - Gives you the ability to take multiples of those processors with high performance memory and scale that out to hundreds and thousands of CPUs. Our strategy is to work with third parties and deliver a complete solution. And to provide open standards and work with everybody who has promising technologies. Just look at the top 500 list, we have a lot of examples of our ability to deliver solutions for high scalability. We have a strong solution for all three areas. And nobody surpasses us in any of the three. Our strategy is to continue to enhance our PA-RISC and Alpha processor offerings. We are going to have multiple generations of both of those that will continue in the engineering labs. And to our customers that have multiple generations of those coming. We are going to Itanium as the long-term road map. We have a brand new fabrication process and brand new architecture for EV7. And the big change there is the memory bandwidth on that chip. It is 12 gigabytes/sec and nobody comes close to that. So it’s for memory bandwidth dominant applications. Alpha is going to continue to serve a large component of the high performance computing crowd for years to come. Itanium has a higher clock frequency and bigger caches. It has a healthy memory bandwidth component to it. And it just screams. It is designed to scream on floating point applications which are prevalent in the HPC market. HP jointly developed Itanium with Intel. And then we choose to leverage their fabrication and on-going engineering design teams. And we are behind the adoption of Itanium with our partners and our competitors and everyone else because we really want to see Itanium take off. Supercomputing Online: Do you have programs for porting to Itanium? Joachims: In fact there are multiples. We have made it easy for ISVs to bring up applications. We have Dozens of applications engineers in technical computing markets that work with the biggest ISVs on porting, tuning and benchmarking. So the biggest ISVs have been promoting availability of their applications and support with Itanium. They have been evaluating the performance, which is best of class. Supercomputing Online: Is there anything you’d like to add? Joachims: Compaq has had such a strong presence at the high end of the marketplace. You look at the top 10 supercomputer sites in the world; Compaq has a strong presence at the high end. There’s a professional services organization to support the designing of those large systems, installing and supporting the customer. Providing expertise with the application rollover to those high end system and that’s a part of organization now. HP has had a strong technical computing market focus predominately in manufacturing and engineering. With Superdome and rp8400, HP has been really successful at delivering systems that are very popular in the technical market. And what we have done is blend those things together into one new division for high performance technical computing. And we have got all the elements of engineering, marketing, sales, worldwide footprint with regional support offices all for high performance technical computing. We’re talking thousands of people working within HP’s HPTC division. ---------- Supercomputing Online wishes to thank Steven Joachims for his time and insights. ----------