InfluxDB 3 includes a Last Value Cache. It returns latest‑value queries in under 10 ms. The feature targets monitoring workloads and dashboards.
Key points
Feature: Last Value Cache in InfluxDB 3
Performance: latest‑value queries under 10 ms
Sources
- Original post
- Linked resource - Linked in the post
Miles training framework built on Modal's GPU clusters
Miles provides a flexible, scalable foundation for teams to run their own training workloads. The training framework is built on top of Miles and runs on Modal's on-demand GPU clusters. The sandboxing infrastructure uses Kubernetes and can handle millions of concurrent rollouts.
Key points
Training framework: built on top of Miles, runs on Modal's on-demand GPU clusters, uses serverless GPUs to flexibly scale inference workloads
Sandboxing infrastructure: built on Kubernetes, able to handle millions of concurrent rollouts
Sources
WEKA NeuralMesh announced to reduce idle GPU costs
WEKA NeuralMesh is a storage and memory platform for AI and HPC workloads. It claims to keep GPUs fed by moving data at memory speed. The platform advertises up to 6× data reduction with less than 5% write overhead and up to 13× write throughput.
Key points
Data Reduction: up to 6× reduction on AI/ML data with under 5% write overhead
Performance: up to 13× write throughput, 1.9× reads and 3× lower latency
Sources
- Original post
- WEKA® NeuralMesh™ - Linked page
Clef-omni launch and Clef-flash price update
Clef-omni is released with audio, video, text, and image input. Cloudflare teams use it to find PII, detect malicious domains, and close spam issues. Clef-flash pricing is now $0.038 per million input tokens.
Key points
Feature: Clef-omni adds audio and video input, plus text and images.
Pricing: Clef-flash costs $0.038 per M input tokens.