What are the limits (size of database, latency,…) between free and paid pinecone accounts?
1 Like
With the free account there is a limit of 1M vectors and typical latency for that would be <100ms. There is a usage estimator that allows you to find expected latencies and costs for different sized indexes once you are in the free tier. There is no strict upper limit to the size of indexes once you’re using standard (we currently have indexes active in the 1B+ range)
Very basic question, just a beginner today, trying out on the free instance. I get this message when upserting vectors generated from openai: {“code”:3,“message”:“Vector dimension 1536 does not match the dimension of the index 128”,“details”:}
Is it the correct solution here to create an index with 1538 dimensions, and if so, is my first experiment not suitable for the free account and I need to start making purchases? (I am too early in the process to commit, but am attempting to create a proof-of-concept.)
Hi @robert,
When you create your index, you specify the dimensions you’ll use. These won’t change after the index is created, and every vector you upsert has to match this dimensionality. Otherwise, the math used to match one vector to another won’t work.
If you find you have to use different dimensions, you’ll need to create a new index. On the free tier, you can have one index at a time, so you’ll need to delete your first one. I recommend you create a collection from it if there’s data you don’t want to use.
Pinecone supports indexes with up to 20,000 dimensions, regardless of tier. The only thing that changes with tiers, as far as storage goes, is the ability to use more than one pod per index. Since pods are what actually store the data, the more pods your index has, the more vectors it can hold.
With 1536 dimensions, you’ll be able to store about 2.5 million vectors with an s1 pod or about 500k with either p1 or p2 pods. If you grow beyond that, you’ll need to upgrade to add pods to your index.
2 Likes