Case Study: Kodiak doubles AI training speed with Lambda

Kodiak doubles AI training speed with Lambda in a week

Kodiak, a company developing autonomous trucking technology, faced a challenge scaling the AI training for its next-generation foundation model, GigaFusionNet. Their existing on-premises infrastructure was insufficient, and using a hyperscaler cloud would have required a complex data migration and offered inadequate networking performance, with months-long wait times for access. They turned to vendor Lambda for a solution using its NVIDIA HGX H100 1-Click Clusters.

By implementing Lambda's clusters with high-performance InfiniBand networking, Kodiak was able to begin training in just one week. The solution provided the necessary GPU memory, inter-node bandwidth, and fast data streaming required for their massive dataset. This allowed Kodiak to double its AI training iteration speed, enabling the team to test hypotheses twice as fast and significantly accelerate their development cycle for safer autonomous vehicles.

KODIAK