Top things you should consider when buying new hardware for your SSAS server.
When you query a SSAS multidimensional cube your query might get split between the storage engine (raw data) and the formula engine (complex calculations). The formula engine is single threaded. So it is important to choose a CPU with high frequency. 3.4 Ghz will be much better than 2.1 Ghz.
The CPU should also have as much cache as possible.
I prefer to always have DAS (Directly Attached Storage) rather than using a SAN. At least 2 fast SSD disks in proper RAID should be perfect. Or you could consider flash memory like Fusion-IO (SanDisk). SSAS multidimensional uses Windows File Cache quite a lot.
I am not using RAID with failover. Because if the server crashes I can create a new virtual server in no-time. And then deploy the cubes to it.
You should choose as fast memory as possible. And make sure you have enough of it 🙂 SSAS Tabular reads the model into memory. So this will increase you query performance quite a lot.
I will not talk too much about NUMA since it’s quite complicated. SSAS Multidimensional should perform quite well using NUMA. SSAS Tabular might have problems on a server with NUMA. I have read some forum posts telling that a query runs faster on a Surface Pro than a high-end server. And that the problem might be because of NUMA. On of the solutions on this is to install a virtual server on top of the server. And then bind one of the sockets to this.
At one of our customer we ended up buying this server (august 2015).
2 x Intel Xeon E5-2643v3, 3.4GHz, 6-core, 20MB L3 cache
120GB Memory 2133MHz
2 x 800GB SSD drive
Please note that this server specification might not suit your needs.
You should always identify bottlenecks before you decide on a new server specification. And then scale the server for your needs.
Marco Russo has written a nice article on “Optimize Hardware Settings for Analysis Services Tabular”