... Huang is Nvidia's amiable patriarch, doling out equal doses of reassuring hugs and tough love. He roams the halls of company headquarters, chatting and laughing with workers, remembering the names of their spouses and asking after their children. But he has little tolerance for screwups. In one legendary meeting, he's said to have ripped into a project team for its tendency to repeat mistakes. "Do you suck?" he asked the stunned employees. "Because if you suck, just get up and say you suck." The message: If you need help, ask for it.
The first sentence cannot be overemphasized: NVIDIA employees do seem to respect good ole Jen Hsun a lot. But from my understanding, the suckage part frightens employees more than anything. The emperor has no clothes, but since he shoots the messenger, nobody’s gonna tell him! They give him the message he wants to hear: “Yes, Jen Hsun, next generation will be even better!”
What better way to operate, then, that to hide your mistakes? I talked of misinformation earlier in the editorial–well, it’s appropriate again, even more than ever. This time, though, it’s internal misinformation. Employees get so fed up they have to tell someone who understands the hypocrisy, but won’t cost them their jobs. Certainly this motivates some of the most reliable sources of the leaks on sites like NFI.
You could alternately compare today’s NVIDIA with the Renaissance Christians, when they refuted many new discoveries, such as Copernicus’. They refuted the truth, probably sometimes for personal gain, but also sometimes simply because they didn’t know what they were talking about, except that saying certain things get them killed. So the true academics learned to distribute their uncensored works underground, to share their discoveries in a manner where church-approval was not required. The parallels here to rumors boards are astounding.
Did you notice we don’t even know ATI’s R420/Loki transistor count yet, even though it seems to have recently taped-out? We knew the NV40’s months ago; seems to me ATI’s employees are much happier than NVIDIA’s.
It can be hard to believe the misinformation customers and employees get is increased by incompetence. Here’s an excellent example, the register usage and FP16 vs FP32 performance problems—time and again noted as the key failure of the NV3X architecture—remains unknown to the NVIDIA PR department responsible for dealing with its ramifications! As late as September 2003, a prominent employee in NVIDIA public relations sincerely claimed to me that FP16 is simply two times as fast as FP32. Problem is, in real games, you’ll have a hard time making it run even 45% faster. With such ridiculously high expectations when it comes to your hardware, it’s not hard to understand why nobody at NVIDIA seems worried—they got the “fastest” card on the market, after all ...