There may be some misconceptions.
A relationship in SQL Server enforces referential integrity (an order cannot have a customer ID that doesn't exist). It does not automatically create an index on the Foreign Key, so it has per se no impact on performance.
But in most cases it is a good idea to define an index on a foreign key, to improve performance.
A relationship that you define in Access on linked tables does neither. It cannot enforce referential integrity (that's the server's job).
It is merely a "hint" that the tables are related via the specified fields, e.g., so that the Query Builder can automatically join the tables if they are added to the query design. (copied from here)
So you should
- Create the relationships in SQL Server to avoid inconsistent data. ("But my application logic prevents that!", I hear you say. Well, applications have bugs.)
- Create indexes on foreign keys where appropriate to avoid performance problems.
- If you are working with queries in the Access frontend, additionally define the relationships there.
Ideally you should have a test server where you can yourself define the relationships, and just send the finished SQL script to IT.