Question 1

We optimize cold start already. Why Cedana?

Accepted Answer

Because Cedana is complementary: it executes at the kernel level and leverages your existing model and kernel optimizations.

Question 2

Why should we try this now?

Accepted Answer

Inference is becoming increasingly stateful, complex, and expensive to serve, as we explain in The Era of Stateful Inference.

Question 3

Does this work with NVIDIA Dynamo?

Accepted Answer

Cedana seamlessly integrates with Dynamo and provides additional capabilities: balancing prefill/decode with faster latency and higher granularity, and increasing reliability for higher SLA compliance.

Question 4

How long will this take to integrate and try out?

Accepted Answer

Integration happens within minutes. We work with you on a testing process to give you high confidence it works and you can scale.

Question 5

Will it work with our autoscaler and orchestrator?

Accepted Answer

Cedana works with Kubernetes and Kueue with a Helm chart. We have seamless integrations for SkyPilot and Slurm.

Question 6

How does it handle our serving framework (vLLM, SGLang, Triton)?

Accepted Answer

Cedana captures at the OS level, so the serving framework is irrelevant to the checkpoint layer. We show resume across vLLM and SGLang on the call.

Question 7

We've built our own warm-pool optimization. Why Cedana?

Accepted Answer

Most platforms have. Most result in overprovisioning as they scale across the model catalog, and this gets worse as model sizes grow. By accelerating cold starts that take 10 minutes or more to seconds, the warm-pool equation completely changes.

Question 8

What about compliance and data handling at our scale?

Accepted Answer

Cedana operates on GPU memory state inside your existing security boundary. It is SOC 2 compliant, ships on-premise, and supports fully air-gapped deployments.

Question 9

Does it handle distributed multi-GPU inference?

Accepted Answer

Yes. Distributed multi-GPU coordinated checkpoint is the differentiating capability. We show an 8-GPU model-parallel job restored as one unit on the call.

Maximize tokens per GPU.