Quad E5-4610 or quad Opteron 6376?

[CougTek]

Both the Xeon E5-46xx and AMD Opteron6xxx are about to be updated. A look at the current generation brought me a question. For a relatively affordable (emphasis on "relatively") quad socket server, what would bring the most performance, a quad E5-4610, or a quad Opteron 6376? I'm not talking about gaming here, purely typical server loads (database read/write, virtualization and perhaps HPC). On paper, the AMD solution seems like it would trump Intel's offering as it has more cores. AMD's IPC is much lower though and there only one FPU per two cores. Intel Xeon E5-4610 : 6 cores/12 threads 2.4GHz, 95W TDPAMD Opteron 6376 : 16 cores/16 threads 2.3GHz 115W TDPA quad-CPU server with 128GB of RAM would cost some 40% less with the Opteron chips. I ask because I might need a modest test server separated from our production setup in medium term. We would put many various stuff on it so it has to performance well all around. Reliability isn't paramount since it would be an internal test station. I haven't found benchmarks comparing both (only their higher-end counterparts). I don't even know how they would fare at FAH!

 

[CougTek]

I don't know why, but text formating has been screwed and I cannot correct it with the editing tool (won't let me save). I've tried with both Epic Browser and Firefox. Must be some weirdo thing with our network.

 

[CougTek]

Found my answer. A Xeon E5-4610 is about on par with a Xeon E5-2630 (100MHz difference between both). So both are more or less equal at virtualization, while Intel's Xeon beats the Opteron more often than not at other workloads despite having less than half the cores. So once the E5-4610 v2 comes out, they just won't be in the same league (unless the upcoming AMD Warsaw CPU is much better than what I expect).

 

[Mercutio]

Do you really want to be testing on hardware dissimilar to what you're using in production?
As I understand it, Opterons are generally in the same place as mainstream desktop CPUs. The performance per core doesn't hold up, so AMD is selling on the basis of providing a lot of low-cost cores and delivering a cheaper overall product.

I can't find any decent numbers to compare either but going through several benchmark sets for both CPU series I think the E5-4600s are around 20% faster per core, but I'd still say that makes the 6376 a bargain.

 

[CougTek]

Do you really want to be testing on hardware dissimilar to what you're using in production?

When everything is running inside virtual machines, having different architectures for the host doesn't matter IMO.

 

[Handruin]

When everything is running inside virtual machines, having different architectures for the host doesn't matter IMO.

It can matter if you use esx server and decide down the road that you want to use the vmotion or fault-tolerant features of esx. For vmotion to move VMs live from machine to machine it needs common CPU architecture (mainly the register and features on the chip) to be similar. If you think you'll ever decide to go down that road then you need to make sure the architecture types are fairly similar otherwise you have to apply a compatibility mask lowering the CPU register functionality down to the lowest common denominator. If one is far superior to the other, you lose the ability to use those features.

 

[Handruin]

In case you want to know more if this applies to your environment...otherwise I'll butt out.

http://kb.vmware.com/selfservice/mi...nguage=en_US&cmd=displayKC&externalId=1003212

 

[CougTek]

We use Hyper-V 2012 R2 most of the time. We also have one or two ESXi systems, but we won't use vMotion on those. Since this would be for a testing environment, backups would be done manually to a NAS.