This, more than anything else, was the message that Clark Masters, executive vice president of Sun’s Enterprise Systems Products group, and Shahin Khan, the newly named vice president of Sun’s new High Performance Technical Computing business unit, wanted to get across in a conference call with the press and analyst community yesterday.

They have a point. One could argue that the initial Stanford Unix Network Microsystems RISC workstations machines that Scott McNealy, Bill Joy, Andy Bechtolsheim, and Vinod Khosla invented in the early 1980s were the seed that grew into the modern HPC computing market as we now know it: massively parallel RISC/Unix boxes and Intel/Linux clusters, with a smattering of exotic vector processing machines. Sun’s current CTO, Greg Papadopoulos, was one of the chief designers (alongside Danny Hillis) of the CM5 massively parallel supercomputer from Thinking Machines. Mark Tolliver, Sun’s chief strategy officer, worked at rival Hewlett Packard and held the top marketing job at Maspar, a defunct parallel computer maker that was eaten by Digital Equipment. Mark Canepa, heads up Sun’s storage unit, and he cut his teeth at HP, running the Apollo workstation unit HP acquired to go after Sun’s fledgling workstation business and eventually driving the HP acquisition of vector supercomputer maker Convex Computer Corp, an acquisition that made the HP 9000 Superdome possible. Sun acquired the rights to the Kendall Square parallel supercomputer patents when that company went bust, having the same thoughts.

But then SGI decided to buy Cray and that changed everything. Clark Masters led the development of the Sparc-based Cray 6400 supers at Cray Research, which SGI did not want, and when Sun bought that business in 1996, Masters turned it into the Sun Starfire Enterprise 10000 server that put Sun in the data center alongside IBM and HP and put Sun in the HPC market proper. Khan, by the way, did product marketing at Floating Point Systems and Cray Research (two companies where Masters held sway). So when Masters and Khan claim that HPC is in Sun’s DNA, they are not exaggerating as the top brass at many companies are apt to do.

Just because these executives get the HPC market does not mean that Sun has been able to get the business from HPC buyers. And that is what the new HPTC business unit is all about: bringing some development, marketing, and sales focus to Sun’s efforts so it can capitalize on all of the human and technological assets it clearly possesses. We’ve been doing reasonably well, says Masters. But I think we can do better with some focus.

While Masters shut down any talk about the revenue streams that Sun expects to get from the HPC market with its focus on the space, the company has said in the past that it gets about $2bn a year in HPC related sales of servers, workstations, software, and services. That’s about one-sixth of Sun’s total sales. Khan said yesterday that while HPC has traditionally focused on number crunching and high resolution graphics to simulate natural phenomena, the market has broadened to modeling business processes (data warehousing and the business intelligence, supply chain management, and customer relationship management software that hands off data warehouses being just one example).

HPTC is more than just servers and workstations. Sun is large enough to invest in HPTC and then bring new technologies into the mainstream, taking the cool stuff Sun Labs is doing and driving innovation, says Khan. But Sun knows that technology doesn’t sell itself really well, and that is why it is attacking certain markets with certain solutions. The new V880z personal server for visualization and the SuperCluster preconfigured HPC clusters, both announced a few months ago, are two examples of the kind of products Khan will try to fashion going forward. Khan says that Sun is working on small and medium clusters for customers who do not need teraflops of number crunching power. We certainly have noticed the uptake of Linux in this area, says Khan.

The odds are that someone at Sun is advocating the use of Linux and 64-bit Opteron processors from AMD – it may even be Khan and/or Masters – for HPC clusters. By pigeon-holing it into the HPC space, Sun can hope to deflect comparisons between Opteron machines. IBM Corp is trying to do the same thing with its support of the Opteron processors, but you can bet that if Opteron takes off, IBM will deliver full-blown commercial servers based on the chip outside of the HPC area. Expect more aggressive and complete Linux solutions soon, was all Masters would say at first about Sun’s future Linux machines, adding later that the Linux machines that Sun delivers will be full participants in the cluster interconnection and software stacks that Sun has for HPC customers.

As for Opteron, Khan showed typical Sun humor and obtuseness. The Opteron announcement improves our overall standing in HPTC because it dilutes the standing of Intel’s Itanium. He went on to say that there are going to be three different and competing 64-bit X86 instruction sets – Intel’s Itanium, AMD’s Opteron, and the skunkworks Yamhill 64-bit version of the Pentium 4 Xeon that Intel won’t even admit exists. With Itanium, Intel has the wrong architecture even if it is the right company, says Khan. And AMD has the right architecture even if it is the wrong company. That last bit was probably a tripping of the tongue, especially if Sun ships Opteron machines and does more than just endorse its software on the chips. Like many server makers, Sun would prefer that Yamhill exists, particularly since HP gets the best deals on Itanium processors because it helped create them. Yamhill would re-level the playing field for 64-bit X86 processors from Intel.

Masters says that while IBM has a per-processor floating point performance advantage compared to Sun because its Power4 and Power4+ processors can do four floating point instructions per clock compared to Sun’s UltraSparc-III, which can only do two FP instructions per clock, this will change with future Sun processors. Sun’s current machines have an advantage in I/O bandwidth and main memory capacities. And he says that in the visualization area, SGI has been the dominant supplier of machinery and has not kept up investments in this area, leaving the door open for Sun. SGI would certainly dispute this claim, but Sun is clearly going to try to eat at least some of its lunch in the visualization market.

Source: Computerwire