HP confident on HPC future
Wants to be the ‘HP’ in HPC, not the ‘dot’ in dot-matrix
Blog A quick meeting with HP at SC11 confirmed that the company is feeling good about their HPC achievements and prospects for the future. HP is the second biggest HPC vendor on the most recent Top 500 list with 141 systems (28 per cent). However, they’re still behind market leader IBM, which has a 44 per cent share with 223 total systems.
The HP picture is worse when you look at a comparison of system size and performance. IBM has 26 of the top 100 systems, while HP has only seven. Of course, one of those seven includes the 1.19 petaflop NEC/HP TSUBAME 2.0 system that’s #5 on the list, which isn’t too shabby.
So why is HP smiling about their HPC chances? First, according to the company, they don’t measure their HPC success by appearances on the Top500 list. They (correctly) assert that there’s plenty of profitable HPC business to be had at smaller customers with smaller-than-Top500 systems.
HP’s HPC strategy has been to stick with deals that don’t require heroic feats of design and engineering, or anything that isn’t off-the-shelf on the hardware side. So no custom processors, interconnects, or exotic accelerators for HP.
They also intentionally eschew deals that look iffy on the margin front – which will take them out of the running for many, if not most, of the deals in the top 250 or so.
HP’s HPC folks are smiling because of what they see happening in the HPC market. Accelerators (like NVIDIA’s GPUs) are a game changer for HP. Hybrid systems (CPU + GPU) are increasingly coming into fashion as the ecosystem of enabled apps and industry expertise quickly increases.
With the right workload, hybrids can run rings around traditional CPU-based designs in terms of raw performance, performance per watt, and price/performance too.
Hybrid systems today fit very well in HP’s wheelhouse. GPUs come in packages that are easy to bolt on to typical servers via PCIe. The aforementioned robust GPU ecosystem means that HP doesn’t have to do any trail breaking when it comes to programming or other engineering-type stuff to make it all work together.
What this means is that HP can play to their strengths in high volume manufacturing, packaging, and distribution. As one of the very largest x86 system vendors in the world, HP gets rock-bottom prices on components. Their Pod Works factory, co-located with their major assembly center in Houston, builds entire data centers to order in their own carrying case (a shipping container-like box).
HP’s PODs (Performance Optimized Datacenters) hit directly at two factors that supercomputer types want most: 1) Lots of bang for the buck and 2) Speedy delivery. With a POD, labs don’t have to build or buy new raised floor space to house their gear. The PODs are self-contained and highly energy efficient – lowering operational costs.
Acquisition costs are lower because HP builds the PODs right next door to HP’s main assembly facility. With everything they need right on site, it’s much faster (and cheaper) to build, test, troubleshoot, and then deliver a POD than it is to ship lots of parts to a customer location and build the system on site.
I think HP’s recently announced Project Moonshot (their effort to radically reduce energy use and foot print, while increasing performance) will play a large role in their future HPC offerings. As part of this program, HP inked a deal with server ARM chip designer Calxeda to produce extremely low power scale-out offerings for lightweight webby apps. It won’t take long for HPC-types to take these chips for a test drive, experimenting with them on their own and combined with GPUs or other accelerators. HP could be the first major vendor in on this trend.
The mainstreaming of hybrid HPC systems gives HP a shot at putting a lot more systems on the Top500 list. Their heft in x86 servers overall gives them the sales channel reach and economies of scale to price their boxes at levels that will be uncomfortable for at least some competitors. PODs give them the ability to deliver huge amounts of gear quickly at a low cost, requiring the customer only to come up with some extension cords and lots of parking lot space.
I don’t expect to see HP hit the pinnacle of the Top500 list, but I do expect to see them become a larger presence on the list, particularly in the top 100. The other vendors won’t be standing still, of course. And the same conditions that make HPC favorable for HP also make it an easier market for others (Dell, and a wide range of second tier players) to attack. ®
Sponsored: Hyper-scale data management