# Cerebrium Serverless CPU and GPU infrastructure for teams deploying real-time AI workloads. - Findings24: - Site: - Preços: Freemium - Categorias: Plataformas de inferência de IA, Implantação de modelos de IA, Nuvem de GPU ## O que é Cerebrium? Cerebrium is a serverless infrastructure platform for deploying and running AI applications on CPUs and GPUs. It supports voice pipelines, large language models, image generation, video workloads, and training jobs, with compute billed by the second. Teams supply their application code or Dockerfile, while Cerebrium handles container orchestration and scaling across clouds and regions. The platform combines deployment, monitoring, and regional routing to help developers operate production AI services without managing Kubernetes or planning fixed compute capacity. ## O que Cerebrium pode fazer? ### Elastic compute and snapshot restores Scale CPU and GPU containers with demand. Memory and GPU snapshotting accelerate workload restores and reduce startup overhead; startup time varies by workload and configuration. ### Existing code and container deployment Deploy from an application entry point or Dockerfile without rewriting code or adding platform-specific decorators. Private Docker images, secrets management, CI/CD, and gradual rollouts support production delivery. ### Multiple request patterns Expose REST, WebSocket, and streaming endpoints, or run asynchronous jobs. Concurrency controls and batching support workloads with different throughput and response requirements. ### Runtime observability Inspect request logs, metrics, scaling events, and system performance in real time. Native OpenTelemetry support connects workloads to an existing monitoring stack. ### Regional deployment and isolation Select deployment regions for data residency requirements. Workloads run in gVisor-based isolated environments, with multi-region failover routing constrained by deployment requirements. ### Co-located voice components Run speech-to-text, language models, and text-to-speech on co-located CPU and GPU infrastructure to reduce network latency between pipeline components. ## Casos de uso ### Telephone and conversational voice agents Build inbound or outbound voice applications with LiveKit or Pipecat. Published examples include Twilio integration and agents combining transcription, language generation, and synthesized speech. ### Language model and retrieval services Serve language models with vLLM or TensorRT-LLM, and deploy embedding and reranking APIs for search workflows. Examples also cover vision-language models using SGLang. ### Image generation and interactive video Run Stable Diffusion XL image generation, digital avatars, and video pipelines with on-demand resources suited to variable request volumes. ### Model training experiments Execute GPU training scripts and hyperparameter sweeps, including a published Llama 3.2 workflow with Weights & Biases. ## Perguntas frequentes ### How is usage priced? Compute is billed by the second, with separate rates for GPU, CPU, and memory. Storage costs $0.05 per GB per month, with the first 100 GB free. Listed GPU rates include $0.000222 per second for L4 and $0.000944 per second for H100. ### Which subscription plans are available? Hobby has no monthly platform fee, with compute charged separately, and allows three deployed apps and five concurrent GPUs. Standard costs $100 per month plus compute and includes unlimited apps and 30 concurrent GPUs. Enterprise uses custom pricing and offers volume discounts and additional support. ### Can AWS or Google Cloud credits pay for Cerebrium usage? No. Cerebrium bills separately from AWS and Google Cloud, so credits from those providers cannot be applied. ### Can a deployment obtain guaranteed GPU capacity? Cerebrium offers guaranteed capacity for bursty workloads through a minimum monthly spend commitment, without requiring continuously paid infrastructure reservations. Terms depend on the deployment and requested compute. ## Informações rápidas - **Category**: Serverless AI infrastructure - **Primary users**: Developers and teams operating production AI applications - **Hardware options**: CPU compute and GPUs including T4, L4, A100, H100, H200, and B200 - **Listed regions**: us-east-1, eu-west-2, eu-north-1, ap-south-1 - **Website**: https://cerebrium.ai