For over 10 years, from the late 1980s throughout the 1990s, technical workstations were the stunningly sexy starlets of the computing world. Advancements in graphics cards and RISC microprocessors gave engineers and scientists unprecedented performance at their desks, and Silicon Graphics and Sun Microsystems became the twin-sister darlings of Silicon Valley, based on their come-hither designs and, more importantly, the walk-into-a-pole gorgeous applications that ran on them. They were beautiful, but we fell in love with them for the genius they embodied. Covetous engineers passed product catalogs back and forth, leering at the front covers to goggle at screen shots and the back pages to devour the accompanying stats. The resulting market for them was as large as that of the entire adjacent HPC industry.
But by the turn of the century, the megawatt spotlight on the glamorous technical workstation was fading. PCs and so-called “personal” workstations closed the gap in performance while offering lower prices, standard hardware and operating environments, and the resulting ability to run personal productivity applications (like word processing and email) within the same configurations. By the mid 2000s, technical workstations were considered over-the-hill, and not coincidentally, SGI and Sun had lost their bombshell appeal, having turned over the runway to fresh-faced ingenues that stacked up on datasheet measurements but somehow lacked the sassiness and sex appeal we so admired in our fading pin-up models.
Another decade later, the technical workstation is largely forgotten – a fascination from a bygone era still celebrated in the memories of experienced UNIX-heads sprouting long gray hair, as they shake their heads at the times we live in and reminisce about the good old days. To be sure, there have been innovations in graphics and in x86 processors, but the workstation – as it is still sometimes called – isn’t used for work anymore. These personal workstations are glorified PCs that are usually better suited for games than for any serious task. These dolled-up prom queens might be pretty to look at, but we don’t love them, and we never will.
The technical workstations we loved had features that PCs did not. They had multiple 64-bit processors and lots of memory, and they produced graphics effects that made you lose your train of thought every time you looked at the screen. They were dreamboxes. (Sigh.)
We sit back in our reverie, confident that we will never long for another system like that again. And it was with jaded superciliousness that I greeted this summer’s new entry-level HPC products.
First came a new base configuration for the Cray CX-1, the CX-1 LC, which not only lowered the entry price point but also established Windows as a credible technical platform. Then SGI suddenly emerged from its post-merger hangover to launch the SGI Octane III, a product clearly engineered – and named – to make us remember what the company was once capable of. And in a deep bow to the adored queen of the new graphics world, both the CX-1 and the Octane III offer the latest NVIDIA GPUs.
I attended NVIDIA’s GPU Technology Conference in San Jose prepared to see some amazing effects, and I wasn’t disappointed. The fact that they were broadcasting in real time in breathtaking 3D HD stereo was only a baseline jumping-off point for showing off their newest tricks, from photo-realistic ray tracing to an eye-popping augmented reality demo that looked more like magic than technology. Amidst all this eye candy, it is even more amazing that an HPC product would turn everyone’s heads.
Fermi, NVIDIA’s next-generation GPU computing architecture, addresses a punch list of technical shortcomings that had held Tesla back. Fermi offers double-precision performance and ECC memory. It has C++. And NVIDIA also introduced Nexus, an integrated development environment with source-level debugging, immersed in Microsoft Visual Studio. Certainly some hurdles remain (such as overcoming the latency hit inherent in moving a calculation off-chip), but the crowd of paparazzi gathering around GPU computing is growing thicker.
The only blemish on an otherwise awesome launch is that NVIDIA still seems to misunderstand where its HPC opportunities are. The demonstrations and endorsements were substantial, but they tended to meander headily through different product classes and application categories. The target markets stated in NVIDIA’s analyst presentation don’t line up to the benchmarks the company is reporting or the ISVs it is targeting. And NVIDIA’s stated total addressable market figure for 2010 – over $1.1 billion in GPU sales (not system sales, but GPU sales) for HPC applications in finance, energy, academia, government, and other supercomputing – is so ridiculous that it can only be the result of intentional self-delusion.
At the heart of the problem is NVIDIA’s oxymoronic designation for its Tesla line, the “personal supercomputer.” This term seems to go around HPC marketing teams every ten years like a new strain of positioning flu: the Try1Buy1 virus. In 1988 Apollo infected IBM with the concept, and in 1998 Apple introduced a new mutation with the Power Mac G4. Apple must have then sneezed on – guess who – NVIDIA, which launched its own personal supercomputer in 1999 to compete.
Apparently this resilient piece of message coding has a 10-year incubation cycle. NVIDIA has introduced a second-generation personal supercomputer to market, and both SGI and Cray have caught the new product class. Maybe 10 years is how long it takes to forget why it didn’t become an epidemic of success the last time. Those reasons are these, for use now and in all future census years:
- People who are looking for a supercomputer want a supercomputer, not something that fits under a chair.
- People who are looking for an HPC adoption platform don’t think of themselves as supercomputing users. In fact, half the time they don’t even think of themselves as HPC users.
- If you finally manage to break through the confusion and attract a potential buyer, you encounter an automatic purchasing roadblock. (Accounting: “No, Jim is NOT authorized to buy his own personal supercomputer!”)
To call any of these products a personal supercomputer is to forget how we might have thought of these products back in the days when desktop and deskside systems quickened our pulses. Oh my, the technical workstation is back, and there are reasons to fall in love.
In Cray’s, SGI’s, and NVIDIA’s entry-level HPC products, we have the ability to bundle best-of-class, whiz-bang graphics environments together with HPC clusters that pack enough smarts to do a wide variety of scientific, engineering and analytical tasks. Notwithstanding the excellent endorsement from Oak Ridge National Labs, the knockout opportunity is in the integration of HPC workflow from desktop to low-end cluster.
“It’s not a personal supercomputer. It is a workstation,” says Jean-Marc Talbot, CEO of CAPS Entreprise, whose HMPP Workbench compiles C and Fortran code for CUDA environments. “[NVIDIA] will have to move aggressively to convince the ISV community, and it’s the same for Cray and for SGI. That’s where the opportunity is. We need to look at what we can do to speed ISV adoption.”
If my earlier conversations with ISVs in the HPC community are any indication, getting their interest won’t be a great challenge. Many of them will look to platforms like Tesla, Octane III, and CX-1 as integrated systems that provide a smooth introduction to their products.
Insights like this have already guided Cray to soften up on the “personal supercomputing” talk and to shift the conversation to how Cray is putting the “work” back into workstation. SGI calls it a personal supercomputer, but Octane was an iconic workstation product, and clearly this heritage was on someone’s mind. Meanwhile NVIDIA is meeting one technical challenge after another, but the company hasn’t put its finger on why its messaging hasn’t thoroughly resonated yet with commercial production environments.
I do know why. I still remember the sexy allure and the mind-blowing effects. I remember the smug pride and the jealousy that seethed among the haves and the have-nots. I remember being able to tell who the cool engineers were by the systems they proudly displayed in their cubicles. I remember the smoldering lust, the burning desire, the primal need to possess such an elegant package of intelligence and sex appeal.
I know what we’ve got here. The technical workstation is back. I’m in love.