Elastic looks to speed up Jina Model Inference by 8x using NVIDIA RTX GPUs

Elastic reports early performance gains from migrating Jina API and Elastic Inference Service workloads from NVIDIA L40 GPUs to NVIDIA RTX Pro 6000 Blackwell Server Edition GPUs, with some hosted models already moved without requiring customer changes. Production and controlled tests showed up to an eightfold reduction in 99th-percentile embedding latency, more than fourfold faster reranking, and a 3.7-fold increase in OCR page conversion speed, while small reranking requests running alongside large jobs saw substantially lower latency. Jina’s compact reader, OCR, embedding, and reranking models are designed for search and agentic AI workflows, handling multilingual and multimodal content, document structure, and result ranking. Elastic emphasizes that shared-service capacity depends not only on model size but also on mixed request patterns, and says it will continue migrating Jina API and EIS models to the newer GPUs.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.