December 9th, 2019 by Michael Rink
Diamanti Announces Kubernetes Platform with GPU Support
Today, Diamanti announced their first enterprise platform with GPU support for running containerized workloads under Kubernetes. The new platform is targeted at the artificial intelligence and machine learning (AI\ML) markets. Diamanti was founded in the early 2010s and provides open-source Kubernetes platforms.
Despite having already been shipped to early access customers, Diamanti has told us surprisingly little about their new platform. We know it is based on their standard x86 server platform. The two big additions to the platform are the inclusion of their Diamanti Spektra hybrid cloud Kubernetes management software and full support for NVIDIA's NVLink cross-connected GPU cards. We covered Diamanti Spektra last month. There have been no major changes since then and it is still in technology preview. NVIDIA NVLink is a high-speed direct GPU-to-GPU interconnect and despite its obvious advantages, is rarely supported by hardware manufacturers. I am pleasantly surprised that Diamanti has gone to the trouble of sourcing a board that does support it. NVIDIA NVLink offers some very obvious benefits for the AI\ML markets Diamanti is targeting. NVIDIA NVLink is a high-speed, direct GPU-to-GPU interconnect. By allowing the GPU hardware accelerators to talk directly to each other without eating up CPU resources, NVLink drastically increases the effectiveness of hardware acceleration provided by having multiple GPUs for parallel processes, which most AI\ML applications can take advantage of. It's really encouraging to see Diamanti taking the next step past providing just bare-bones GPU support in its platform.
Currently in early access