NVIDIA and Taiwan’s Hon Hai Precision Industry Co (Foxconn) have increased efficiency at key stages of assembling a rack-mount AI platform NVIDIA GB300 NVL72using specialized robots, reports Taipei Times. In the post “Machines That Make Machines” NVIDIA explained how the company, with support from Foxconn, trained robots to assemble critical server components, increasing production consistency and workflow efficiency.
The company said the assembly success rate has exceeded 90% and said the company has focused on assembling the GB300 test trays used to test the functionality and performance of NVIDIA computing systems before shipping to customers. The key tasks for the robots were installing conductive busbars and connecting connectors.
One of the main challenges in installing the power bus was teaching the robots to grab it, move it, place it in place and secure it with screws at 16 points. The robots also had to learn how to manage two large (DC-SCI) and two small (MCIO) connectors with connected cables, which must be carefully inserted into the sockets on the server board, and the tolerances for this are very small.
Image source: NVIDIA
According to test data, the new automation mechanism achieved a DC bus installation success rate of more than 95%, with an operation completion time of 160 seconds. This is still below the targets of 99.5% and 124s, respectively. In the case of connecting connectors, the success rate of the operation was 90–95%, and the average time was 40 s per cable. In this case, the target execution time is no more than 72 s for all four cables.
NVIDIA emphasized that when assembling test trays for the GB300, automation systems are needed that are prepared for “uncertainty,” since rapid changes in development cycles and relatively small production volumes make the creation of assembly lines with strictly defined parameters impractical – robots must be sufficiently “adaptive” and at the same time reliable to respond to changes in the assembly process. It is emphasized that the technology is not yet ready for mass implementation, since there are problems with coordinating robotic hands when connecting connectors, etc.
Foxconn is one of NVIDIA’s key manufacturing partners. Last year, the company announced it was working with the latter to build a smart factory at the Houston site. in Texas using humanoid robots controlled by NVIDIA’s AI to produce AI servers for NVIDIA itself. Foxconn expects to expand the use of similar robots to other production facilities in Texas, Wisconsin and California.
Foxconn accounts for more than 40% of the global AI server market. However, she remains optimistic about the future of the industry. At an investor conference in August, it said shipments of its rack-mount AI systems would more than double from the previous year.
Sources:
