• In memory tempdb would be nice, except when you don't have enough memory.

    I'm working with a good sized database (~300+GB with indexing). Our server is a bit aged, but has 16GB available for the reporting instance (32GB in the server, 16GB reserved for cluster failover)

    Even if I managed to convince management to buy a single server with 128GB for just that system, I don't believe there would be enough space to cache data and store tempdb in memory with all the reporting/extracts running at the same time (10 extracts, up to 20 reports concurrent). With some rather large tables in the system, I just don't see how to squeeze all of that into the memory space you are advocating.

    I do see a HUGE benefit to sticking the tempdb onto an SSD. Not sure about the longevity of an SSD with tempdb beating on it all the time, but you would get big iops numbers while it is there.