Stardog has announced the launch of SKATHE, a new private GPU cloud facility accelerated by NVIDIA and located in Ashburn, Virginia. SKATHE provides a dedicated solution for processing GPU compute and is designed to enhance the performance of Stardog Voicebox, the company’s 100% hallucination-free AI data assistant.
SKATHE offers a hybrid cloud infrastructure that balances GPU and CPU workloads with CPU processing resident in AWS and Azure. This initiative comes at a time when businesses are reassessing the cost-benefit balance of cloud operations, especially following significant investments over the past two decades, and recent pushbacks against cloud spend by Fortune 1000 companies.
“We invested in SKATHE because of our dual mandate to offer customers world-class user experience while also building world-class unit economics for ourselves” said Kendall Clark, Founder and CEO of Stardog. “SKATHE lets us do both by positioning us for flexibility and profitability.”
SKATHE uses NVIDIA Triton Inference Server with NVIDIA TensorRT-LLM, supported by servers equipped with NVIDIA GH200 Grace Hopper Superchips. The integration with NVIDIA’s full-stack accelerated software and computing makes it easy and fast to deploy Voicebox’s ensemble of AI models across various deep learning and machine learning frameworks.
The TensorRT-LLM library enhances inference performance on NVIDIA GPUs, featuring advanced optimizations like dynamic inflight batching and masked multi-head attention for superior latency and throughput. The GH200 Grace Hopper Superchip integrates an NVIDIA H200 Tensor Core GPU with a 72-core Grace CPU, delivering up to six times faster inference compared to NVIDIA A100 Tensor Core GPUs, ensuring high performance and robustness.
Planning for rapid growth, Stardog anticipates the global expansion of SKATHE facilities to strategic markets, including New York City, the Bay Area, Texas, London, and Germany. The company will extend its infrastructure globally and increase compute capacity to support additional GPU-intensive tasks such as knowledge graph question answering, graph neural networks (GNN), vector embeddings, continuous fine-tuning, and model training.
“As we scale SKATHE globally, we are focusing on geographic and computational expansion,” Clark added. “We see opportunities for additional growth by expanding SKATHE into a global private GPU cloud including new compute services beyond model inference.”
Explore AITechPark for the latest advancements in AI, IOT, Cybersecurity, AITech News, and insightful updates from industry experts!
