Regions & Egress

Currently, we default all traffic to gcp-us-central, but we can spin up other regions in hours on request. We already run in us-west1 (Oregon) and Montreal for select customers.

Network latency is not included in query performance, and should be added to estimate total roundtrip latency. You can expect gcp-us-central to have the following approximate latency to other common regions:

us-central -> us-central  ~0.5ms
us-central -> us-east     ~25ms
us-central -> us-west     ~40ms
us-central -> eu-west     ~100ms
us-central -> eu-central  ~110ms
us-central -> tokyo       ~120ms
us-central -> sydney      ~180ms
us-central -> singapore   ~200ms

Typically, hosting providers charge you ~$0.1 per GB (150 vectors @ 1536 dimensions) of traffic leaving their network. This is called an egress internet fee. It is a common misconception that as long as your vendor is within the same region or zone, you will not pay egress fees.

Unless you set up VPC peering or a Cloud interconnect, that is the fee you will pay, regardless of whether you are on GCP or AWS, as the IP will go through the public Internet.

For sufficient scale, we can set up peering or expose regional/zonal IPs to you, to reduce your and our egress fees, typically to ~$0.02 per GB with some fixed costs. However, due to turbopuffer's cheap pricing, this is likely not something you will need to bother with until large scale.

1M   vector writes @ 1536 dimensions: 6  GB, $0.6
10M  vector writes @ 1536 dimensions: 60 GB, $6
100M vector writes @ 1536 dimensions: 60 GB, $60
1B   vector writes @ 1536 dimensions: 6  TB, $600

These fees are charged to you by your cloud provider on your Cloud bill. We are also charged egress fees by our provider to serve you query results.

Email us
© 2024 turbopuffer Inc.
Privacy PolicyTerms of service
SOC2 Type 1 certified