What is best use of 1400 gb SGA (2 rac nodes 768gb each)

Question

currently using 11.2.0.3.0 on unix sun sever with 2 RAC nodes each 8 UltraSPARC-T1 cpus (came out in 2005) four threads each so oracle sees 32 CPUS very slow(1.2 gb).  Database is 4TB in size on regular SAN (10k speed).
8gb SGA.
New boss wants to update system to the max to get best performance possible  Money is a concern of course but budget is pretty high,  Our use case is 12-16 users at same time, running reports some small others very large (return single row or 10000s or rows).  reports take 5 sec to 5 minutes, Our job is get the fastest system possible,  We have total of 8 licenses available so we can have 16 cores.  We are also getting a 6tb all flash SSD array for database.  we can get any CPU we want but we cant use parallel query server due to all kinds of issues we have experienced (too many slaves, RAC interconnect saturation etc, whack-a-mole).  sparc has too many threads and without PS oracle runs query in single thread. 
we have speced out the following system for each RAC node
HP ProLiant DL380p Gen8 8 SFF server
2 Intel Xeon E5-2637v2 3.5GHz/4-core cpus
768 gb ram
2 HP 300GB 6G SAS 15K drives for database software
this will give us total of 4 Xeon E5-2637v2 cpus 16 cores total (,5 factor for 8 licenses) and 1536 ram (leaving ~1400 for sga).  this will guarantee an available core for each user.  we intend to create very very large keep pool around 300 gb for each node that will hold all our dimension tables.  this we hope will reduce reads from the SSD to just data from fact tables.,
Are we doing a massive overkill here?  the budget for this was way less than what our boss expected.  will that big an sga be wasted will say a 256gb be fine.  or will oracle take advantage of it and be able to keep most blocks in there.
will an sga that big cause oracle problems due to overhead of handling that much ram?

Answer 1

Current System:
===========
a. Version : 11.2.0.3
b. Unix Sun
c. CPU - 8 cpus with 4 threads => 32 logical cpus or cores
d. database 4TB
e. SAN - 10k speed disk drives
f. 8gb SGA
g. 1.2 gb ??
h. Users --> 12-16 concurrent and run reports varying size
i. reports elasped time 5 sec to 5 mins
j. cpu license -->8
Target System
===========
a. Version: 11.2.0.3
b. HP ProLiant DL380p Gen8 8 SFF server
c. RAM --> 768 GB
d. 2 HP 300GB 6G SAS 15K drives for database software
e. large keep pool -->90 gb to  hold all dimension tables. 
f.  SSD to just data from fact tables
g. SGA -->256gb
Reassessment of the performance issues of current system appears to be required.Good performance tuning expert is required to look into tuning issues of current application by analyzing awr performance metrics . If 8GB SGA is not enough,then reason behind so is that queries running in the system are not having good access path to select lesser data to avoid flushing out of recent buffers from different tables involved in the query. Until those issues are identified , wherever you go, performance issue wont be going away as table size increase in future , problem will reappear.Even if the queries are running with more FULL Scan , then re-platforming to Exadata might be right decision as Exadata has smart scan , cell offloading feature which works faster and might be right direction for best performance and best investment for future.Compression (compress for OLTP) could be one of the other feature to exploit to improve further efficiency while reading the lesser block in lesser read time.
Investment in infrastructure will solve a few issue in short term but long term issue will again arise.
Investment in identifying the performance issues of current system would be best investment in current scenario.