We currently have 2 SQL Server 2005 Enterprise machines in a mirrored environment running on Dell R815 servers. These machines consist of:

Four AMD Opteron 6176 Processors (12 cores at 2.3Ghz), for a total of 48 cores.
512GB DDR3 1033Mhz RAM
10GB Ethernet SAN connected LUNs for MDFs, Logs, TempDB, Backups, etc.

We are looking at upgrading to SQL Server 2008 R2 Enterprise or SQL Server 2012 Enterprise.

I understand all the features that Microsoft presents in their standard comparison guides, however it seems to me there is likely a lot of other smaller features that are not present in these official guides that might make a DBA's life easier in SQL 2012.

Can anyone suggest killer features that would sway towards SQL Server 2012 over 2008 R2?

As a result of Microsoft changing their licensing scheme from socket-based to core-based, we are looking at a huge expense to license all 48 cores. To combat this unexpected rise in price, we are considering swapping these CPUs out for faster CPUs with a lower core count. We are considering using 4 x Opteron 6204s, which are quad-core and run at 3.2GHz. Our current systems are typically running at around 25 to 30% utilization; does anyone have experience dropping from 48 cores at 2.1Ghz to 16 cores at 3.2Ghz (or similar), and can we expect a perceivable decrease in speed? I know that's a loaded question, and am expecting a fair amount of flack - however it just helps to think this through by getting the help of other more experienced DBAs.

  • 1
    columnstores for BI/DW? Aug 31, 2012 at 16:45

1 Answer 1


Typically the features that you don't hear about from marketing are the ones that don't bring in all the money (e.g. pushed as "enterprise feaures" in order to sell Enterprise Edition). I answered a similar question here that provided a list of my favorite new features in 2012, that also aren't limited to Enterprise Edition:

What are Objective Business Reasons to Prefer SQL Server 2012 over 2008 R2?

(I could list them here but I don't think duplication makes sense. And there are some other good answers on that page as well.)

The only difference would be you could add to the list the things that have been added from 2008 on: date/time data types, data compression (Enterprise), backup compression (Standard+), some syntax improvements (multi-row values, inline declare + assign, +=/-=, etc). These are things you would get, though, whether you went 2005 -> 2008 R2 or 2005 -> 2012. Guess I knee-jerked listing those because you won't get them if you stay where you are.

In addition, moving to 2012 now instead of 2008 R2 buys you a longer lease toward end-of-life / end-of-mainstream support. However if you are using CAL licensing you might consider 2008 R2, because this option isn't available anymore in 2012 Enterprise. And core licensing can be more expensive if your CPUs have more than 4 cores per socket. They recently updated the "How to Buy" part of the site; you should stay on top of that and keep in good touch with a licensing rep. They're not perfect (and we can have a different discussion about this) but they're the only ones who can give you plausible deniability should anything they sell you end up putting you out of compliance, as long as you haven't lied to them. :-)

If you're going to take advantage of CPU-intensive things like compression, then having faster, fewer cores may be tolerable overall especially since it will make your SQL licensing (quite likely the most expensive part of your solution) that much cheaper. More, slower cores will be more expensive in terms of licensing, and spreading the work over more cores won't necessarily do any better for an OLTP-type workload - since many tasks cannot be parallel anyway. Also, AMD cores are subject to a core factor, which means the pricing is adjusted. This means you pay less per AMD core than per Intel core, but this is probably reflected in the performance, as well (at least if my conversations with Glenn Berry are any indication). No discount like this is in effect for 2008 R2 (at least AFAIK) since you're paying by the socket. Personally, I would rather spend the money on the better cores, but at 48x + 1.33x for licensing I'm not sure I could sell the bean-counters on it. :-)

But there's no way anyone here can tell you which way to go. If you can get some spec hardware in both configurations that would be ideal - you really should test your full workload cycle to see if fewer, faster processors are better for your workload - or at least close enough that they are justified by the licensing savings. On modern hardware I'd expect that to be a better option but I am not going to sign anything that guarantees it. :-)

  • Yeah at IE3 last week Glenn Berry showed that two machines with fewer, faster processors on two sockets are much better overall both in terms of OLTP performance and cost (and cost both in terms of SQL licensing and overall cost - 10s of thousands of dollars less) than a more powerful, four-socket machine. Of course that dictates the need to be able to spread your workload across multiple machines, which may or may not be possible in your case. Aug 31, 2012 at 17:22
  • 1
    Yep, just note that the secondary needs to be licensed and has to be on a WSFC node (so the underlying OS must be Enterprise as well). This isn't a replacement for log shipping to a cheaper server, lower edition, etc. :-) Aug 31, 2012 at 17:25
  • Way too application-dependent. You can do logical partitioning, distributed partitioned views, do all writes on one but favor reads on a copy, etc. Aug 31, 2012 at 17:27
  • SQL Server does not have to be clustered in order to host a readable secondary. But the operating system must be a cluster node. (My T key is dying.) Aug 31, 2012 at 17:29

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.