Azure Synapse vs Databricks

[Phani1]

Hi team,

Could you kindly provide your perspective on the cost and performance comparison between Azure Synapse and Databricks SQL Warehouse/serverless, as well as their respective use cases? Thank you.

[mhiltner]

Hey @Phani1 ! 

I have had a few cases where we have seen huge improvement when moving to SQL Serverless. Processes that used to take hours in Synapse went to minutes with Serverless. Unfortunately i don't have cost benchmarks. 

About use cases, Databricks can be used for streaming, while Synapse has no support for Spark Structured Streaming. Also, Synapse's dedicated pool is very different from Databricks Serverless SQL, as DP involves proprietary storage and has no separation of compute and storage. 

What worries me the most about Synapse today is that although it is a great no-code tool, it no longer receives new features, so it is expected to become outdated soon. For example, Synapse does not offer row level security or column masking, nor structured streaming (as mentioned before) and will not get these features as it is in end-of-life. 

[Witold]

Agree with @mhiltner, it doesn't make sense to compare it with Synapse, as it's literally dead. You most likely want to compare it to Fabric instead. Fabric is highly under development, but IMHO it still lacks behind other Data/AI solutions. No catalog layer (like in unity catalog), no Serverless (in Fabric you pay for idle time) or proper AI tooling (although ML Studio is not bad tool) integration. But this might of course change in the next month.

[Suncat]

I'm curious, how confirmed is this the synapse being outdated??? Do you know a date range when will be officially announced that its end of support please???

[Witold]

@Suncat 

There hasn't been any major changes for than a year: https://learn.microsoft.com/en-us/azure/synapse-analytics/spark/apache-spark-version-support E.g. I don't believe we will see support for Spark 3.5 at all. At least, apparently it's supported until Q1 2026.

Another thing, DP-203 is being retired.

This kind of make sense, because MS is moving everything to Fabric.