We provide a new benchmark, called Sparqloscope, for evaluating the query performance of SPARQL engines. The benchmark combines three unique features, which separates it from other benchmarks: 1. Sparqloscope is generic in that it can be applied to any given RDF dataset and it will then produce a comprehensive benchmark for that particular dataset. Existing benchmarks are either synthetic, designed for a fixed dataset, or require a query log. 2. Sparqloscope is comprehensive in that it considers most features of the SPARQL 1.1 query language that are relevant in practice. In particular, it considers advanced features like EXISTS, and various SPARQL functions for numerical values, strings, dates, language filters, etc. 3. Sparqloscope is specific in that it aims to evaluate relevant features in isolation and as concisely as possible. The benchmark generated for a given knowledge graph consists of only around 100 very carefully crafted queries, the results of which can and should be studied individually and not in aggregation. Sparqloscope is free and open-source software and easy to use. As a showcase, we use it to evaluate the performance of six SPARQL engines (QLever, Virtuoso, MillenniumDB, GraphDB, Blazegraph, Jena) on two widely used RDF datasets (DBLP and Wikidata). The full materials and more are provided on https://purl.org/ad-freiburg/sparqloscope .