Slashdot Mirror
Inside Nvidia's Testing Facilities
An anonymous reader writes "FiringSquad has up a behind the scenes look at NVIDIA's Santa Clara HQ. In addition to the usual shots of the server farm, they spend several pages talking about the Silicon Failure Analysis Lab which is the secret to NVIDIA's success as a fabless semiconductor company. They also have shots of NVIDIA's thermal analysis lab where they run the GPUs at 40 deg C and 0 deg C, and the Performance analysis labs."
I think that just means no one has posted yet. When you see that it's your opportunity to be the first to say something.
The article greatly oversimplified the compute HW setup. Nvidia has a many-thousand-node computational grid with servers across a wide variety of size tiers for different job types (mostly chip design/validation). Stuff is tested pretty extensively prior to mass purchase, and what's running a given size tier depends a lot on combinations of demand scheduling and HW vendor model rollout scheduling, both in CPUs and the boxes they sit in.
40 degrees C is a sort-of standard for "elevated ambient" testing of electronics. The point of testing at higher temperatures is mostly to ensure that heat transfer out of the chips is sufficient at that temperature to keep them from overheating. The chips themselves will likely be running at much greater temperatures internally, but as long as the heat sinks are efficient enough, the chips shouldn't overheat.
For consumer electronics, I guess the assumption is that if it's 40 degrees in your room, you're going to go find somewhere cooler to be, rather than sitting there with your PC blowing hot air on you.
In other industries, the standards are different. Many products designed for use in an automobile are tested at 50-60 degrees, which is closer to the interior temperature of a car in full sun in a temperate climate.