Intel and University fall out in hot-tempered marital spat

TEMPERS RAN HIGH at the HPCA-10 conference in Madrid this week. Sparks flew at a panel discussion on the relationship between industry and academia over Intel's presence at the campus of one Spanish university. Mateo Valero, the charismatic director of the computer architecture department of the Universitat Politecnica de Catalunya (or UPC), during the Q&A session lamented what he saw as a one-way street in collaboration between Intel's on-campus labs and his department. The remark was directed at panel member Antonio Gonzalez, who is a professor at UPC but also runs one of the Intel labs.

Matters were not helped by the way the Intel segment of the panel, which included the moderator, eventually smothered the debate in pleasant but one-sided banter--pointedly ignoring a fuming Valero, armed with a large but not very reliable microphone, until they were able to announce that time had run out and no more questions could be taken.

Harsh Words
According to Valero, he used to encourage his students to mix with the Intel people and "tell them your secrets over lunch" to improve their chances of getting hired--but saw little openness coming the other way.

Gonzalez on the other hand argued that knowledge was flowing both ways, as students worked well-paid stints at the lab and returned to the university with valuable new experience. The current model, with Intel running a so-called closed lab, was a Good Thing according to Gonzalez, who claimed that "managers at the university are very positive about the way we work. It is a win-win."

Panel members Yale Patt and Mark Hill, weighing in for the academic side of the debate, concurred on the importance of internships but joined Valero in questioning that last point. El Patt was never one to mince words: "How does it become a win-win for the university," he wanted to know, if the students who are not employed by Intel are spilling their guts while the students who are employed by Intel say "what's mine is mine and what's yours is yours"?

Intel's Justin Rattner, also on the panel, didn't see the situation as either unique or problematic: "by putting our labs close to the universities we can communicate with a much larger section of academia and do research of interest to Intel." He described the relationship as "a uniform model, based in biological sciences where a new drug, a new method which is going to be worth billions of dollars is applied in the computer field. I've had guys come after me wanting millions of dollars for something that was going to be one tiny element in a future processor."

"All collaboration is good," quoth Valero, "even in the humanities[1]. But for us it's been almost impossible to get enough funding to attract industry attention. We need information from industry, and vice versa."

And particularly because certain companies, he said, "compete with us for conference papers. It's not true that Intel has one lab in Barcelona--they have two. One was the old Compaq lab that developed Tarantula[2] in direct collaboration with us. We have no problem with them because they do not compete with our research." But the closed lab in the same building, according to Valero, now sometimes works on the exact same subjects as the UPC's students without collaborating with them, and then enjoys unfair advantages in presenting essentially the same work to the academic community.

The implication is clear: the Intel lab is taking students' ideas but not giving anything back, harming independent research that it itself benefits from. Insiders certainly confirm some instances of this, and experience has wisened staff members at UPC about being too forthcoming with their ideas when dominant IT manufacturers come knocking.

Still, it would be unfair to read global Intel policies into these incidents. Tensions will arise when a closed laboratory and an open academic institute are expected to work together--all the more so when there are glaring conflicts of interests. As Valero pointed out, Gonzalez is both a fully paid, full-time professor at UPC and a rival Intel employee. With Intel paying Gonzalez more than the university could, Valero maintained, it is practically impossible for the former to treat the university impartially.

The camps were seated at separate tables at the banquet the next day, where Valero was presented the certificate of his IEEE fellowship, an accolade he was formally awarded half a decade ago but never received the paperwork for. Some say Gonzalez was not applauding, but both men are professionals and will surely be able to see eye to eye despite their differences. Privately they might even appreciate the passions their work inspires. It sure beats stuffy bookworms in white coats and bone-rimmed glasses.

A Threat to Science?
In the end, no knives were drawn and life will go on despite the occasional argument. Nonetheless, there are obvious risks in a poorly balanced relationhip between computer architects in academia and industry--particularly when the sheer mass of a company like Intel is involved.

Intel like other corporations in the field disposes of vastly more powerful simulation software to evaluate new microarchitectural concepts than universities do. That software, along with other key information, is carefully shielded from Intel's interns and other outsiders--often for no other apparent reason than an ingrained culture of secrecy. The universities' job is to come up with ideas, some say, and the manufacturers can worry over whether they actually work with real products. Given that situation, it's hard to imagine manufacturers not being tempted occasionally to tell researchers their ideas are no good, while secretly continuing to work on them and/or applying them in their products.

The result is a growing gap between well-equipped industry on the one hand, and on the other, academics who are left unable to validate their own work unless they commit their research to the benefit of selected manufacturers. Either would be a threat to the scientific process. Intel in particular wields a dangerous power in the market--whether it wants to or not. Its failure to disclose its key proprietary simulation software may heap considerable pressure onto the universities to join or die.

Is computer architecture doomed to disappear from the open world of academia? Will there continue to be a role for the universities beyond breeding employees and funding research for the major manufacturers? Let's hope there will be. One thing the panel agreed on quite readily is that universities can afford to look years further ahead than the manufacturers can, and both perspectives are required for continued progress.

Footnotes

1. This is a reference to Mark Hill's Ten Commandments of Poor Technology Transfer (see earlier article), a guide on how computer architecture scientists should not do their job. It disparages relationships with the industry: "relationships are for people in the humanities."

2. Tarantula was a spectacularly high-spec'ed vector co-processor for the Alpha, sadly killed by Compaq as a dowage to HP.