Indexing Pipeline tends to vary greatly in runtime on same machine

[demongolem-biz2]

I am working on a shared GPU machine which generally speaking I am the only one who is running processes on it. Each time I kick it off, the indexing pipeline seems to take a different length of time, perhaps 10s for one run and 80s for another run of the same documents.

I have this code

indexing_pipeline = Pipeline()  
indexing_pipeline.add_node(component=converter, name="TextConverter", inputs=["File"])   
indexing_pipeline.add_node(component=preprocessor, name="PreProcessor", inputs=["TextConverter"])  
indexing_pipeline.add_node(component=document_store, name="DocumentStore", inputs=["PreProcessor"])  

Are there reasons proper to the Haystack code base which can explain such a great swing?