AMD Opteron Interlagos and VDI

[kclinden]

I am engineering my company's VDI project and I can't find much information on using the interlagos CPUs with VDI. I should be able to reduce the amount of servers requried for my solution if i use these processors instead of ones with 12 cores. What about intel vs amd for VDI?

Thanks
Kasey

[gunnarb]

All I see is this question about processors and I'm not aware of your overall goals but I do feel that I should comment on this as the question seems out of place.  So please take what I say with a grain of salt, if a customer asked me what you just asked I'd respond the same, if I saw the customers background work and I know the context of where this question is coming from I'd respond differently, so this response is coming from the fact that all I know is you are asking about CPUs in a View Forum and my response is...

I'm not sure if CPU is what you should be designing around.  RAM tends to be the bottle neck, not CPU.

Also, keep in mind good design pratice isn't about squeezing as many desktops into a server as you can.  This "might" reduce costs, but you are putting more eggs in a single basket.  That server WILL fail eventually, so even if you can put 512 VMs on a single server, should you?  This is a common argument when I'm dealing with Cisco UCS customers who can put terabytes of RAM in their hosts.  You don't necessarily want to and even when you can it jacks the per VM cost up becuase RAM is what costs the most not CPU.

Pod/Block design calls for a maximum of 128 VMs per host.  In a simple design I'd drop 256GB or RAM in that server and dual proc 6 core processors (8 cores would be better).  That should about guarantee me I'd get 128 VMs in that single host.  Of course if I really look at the cost of this type of a server I might find that its cheaper to get two servers that have 2 proc quad cores with 128GB of RAM and just aim to get 64 VMs per host.  Less eggs in one basket and normally I can buy 3 of those for every one of the big beefy hosts.

In the end CPU is rarely a concern, good scalable design is about doing VDI cost effectively, having the latest greatest processor is rarely on the design table because RAM is normally the place where the money is spent. 

My two cents: Get whatever processor you like and whatever makes the most sense from a cost stand point.  Don't choose a processor becuase you can squeeze a few more VMs on a host, design your deployment, do capacity planning and determine what you need.  Heck during that process you may find that your VMs are CPU hogs, in which case finding out the best CPU could be extermly important.  I just haven't seen this in any deployment yet (even my big UCS deployments).  Most deployments have a number "we need __Ghz __GB RAM, and __TB in storage to virtualize __ VMs" then we figure out the most cost effiective way to accomplish that.

Again, I don't know the context of your question, I only see the question the context could be that you know all of what I just said and I just sound like an ass, if so I appoligize.  If not, then I encourage you to make sure you are designing for scalablity, high availability and most importantly, for what you need.

Gunnar

[kclinden]

Mainly I want to know if there is a performance gain/decrease when using AMDs 16 core cpu. In our VDI solution well be installing around 70 servers using HP BL465c G7s with the Opteron 6176. That is a 12 core cpu. Each server will also have 256gb of RAM. Because 70 servers is a fairly large amount to add to our server environment. Reducing the server count by say 20% will allow us reduce the amount of servers admins needed, the amount of physical space, heat, and power. We have about 9000 desktops and will be virtualizing ~70% of them. That puts us at about 63 VMs per server in our current planned configuration. If we were to reduce it down to say 60 servers that puts as at ~100 VMs per server.

[gunnarb]

And there is the context! For all I know you were deploying 50 vms and concerned about subtle differences between AMD and Intel. 

I don't have an answer for you.  I know how I would get the answer in your position.  I'd leverage you large buying power to get both servers in as a demo then use View Planner (or something similar) to stress test them with one of your workloads.  This will give you the exact answer you are looking for.  I think View Planner is publicly available, if not your local VAR should have access to it.  Also I think Liquidware Labs has something better.

Not an answer I know but a suggestion, I'm not sure if you'll find someone who's stress tested that specific processor to see which gives more bang for the buck but you can always do it yourself and post a blog.  Honestly that sounds like a blast personally, but I'm a big nerd for this stuff.

Gunnar

[kclinden]

Any info on whether ther is a large difference between Intel's best vs AMD's best for destkop virtualization?

[Casper42]

Not sure there is a big diff at the VM level between the 2 vendors, I would focus on your app stack and see if there is a preference there.  Something else to keep in mind is Interlagos only adds an extra boost with vSphere 5.  Supported on 4.1 but then all you get is the normal boss from more votes and IPC improvements.  I think Sandy Bridge will end up the same.  Also you mentioned HP, go through your HP rep/SA and ask then to run your questions past the VDI shared solutions team.  Some wicked smart guys who focus on nothing but VDI.  They can escalate to "Doug" in Houston if need be as well.  He's the Product engineer who does an insane amount if testing and writes the reference architectures we publish.

[kclinden]

Thanks for the info. We are going to be using vSphere 5 and View 5.

[Casper42]

I love Auto-Correct on my phone.

Normal boss from more votes = Normal BOOST from more CORES