There isn't a limit or performance metric we don't think we can improve by an order of magnitude when prioritized! If you expect to brush up against a limit or are limited by performance by an operation, contact us. Often can be fixed in days.
turbopuffer excels at | turbopuffer may not currently be the best fit for |
---|---|
Large scale (1B+ documents/vectors) with lots of namespaces (tens of millions) | Large namespaces (100M+) |
Naturally sharded data (e.g. B2B where each tenant's data is isolated in its own namespace) | Low scale, free tier |
Cost-effectiveness | Aggregation (e.g. group by, sums, explore clusters, ...) |
Fast cold starts | Single-digit millisecond latency (tpuf is currently low double digits) |
Reliability | Extensive 1st-stage ranking (we encourage generating a candidate set with hybrid search and refining/re-ranking further in your own 2nd stage) |
Hybrid search (BM25 + vector search) | Built-in re-ranking (we encourage you to do it in your own application) |
Support from DB Engineers | Built-in embedding (ditto) |
Deploy into your VPC (BYOC) | Open Source |
Metric | Max seen in production | Production limits (current) | Production limits (soon) |
---|---|---|---|
Max documents (global) | 50B+ | Unlimited | |
Max documents (per namespace) | 180M | 100M | 1B+ |
Number of namespaces | 15M+ | Unlimited | |
Max dimensions | 10,752 | ||
Max inactive time in cache | ~3 days | Contact us for custom | |
Write rate (global) | 150,000 doc/s | Unlimited | |
Write rate (per namespace) | 5,000 doc/s | 5,000 doc/s | |
Max upsert batch request size | 256 MB | 256 MB | |
Max write batch rate (per namespace) | 1 batch/s | 1 batch/s | 4 batches/s |
Queries (global) | 2K+ queries/s | Unlimited | |
Max queries/second (per namespace) | 100+ queries/s | 100+ queries/s | 10,000 QPS |
Vector search recall@10 | 90-100% | 90-100% | Configurable |
Max attribute value | 8 MiB | 8 MiB | |
Max attribute name length | 128 | 128 | |
Max attribute names per namespace | 256 | 256 | |
Max namespace name length | 128 | 128 |