r430 PowerEdge CPU Options

[NPDPdev]

I have recently purchased a PowerEdge R430 and had a few questions.

The configuration I ordered contains one e5-2630v4, but I would like to upgrade it upon arrival. I am looking at getting either another e5-2630v4 with 10 cores (bringing the server to 20 cores spread across 2 CPUs at 2.2GHz (3.1 boost).

My question is: Should I just get the $120 2630v4 or should I sell the one that is in there and buy an e5-2698v4 for $400-500 with 20 cores clocked at 2.2GHz (3.6 boost)?

My knowledge tells me that the singular 20 core is best, especially since I will eventually be able to upgrade to two of them (and that is what I eventually want regardless of what I do now). I also believe that it is better to have all of the cores on one CPU, since there is less performance lost when the CPUs both have to communicate between each other and the chipset.

I am planning to use this server to offer up virtualized computing to customers and so I need the core count to be higher than ten. Also, if I am running something like ESXi or linux with Docker, how many system resources should I allocate to the host OS/hypervisor. I was thinking about reserving two cores and 8gb of ram, or does ESXi do this on its own and not really need dedicated processor cores to manage these instances?

 

Edit:

The server also contains 32GB of ram, I believe DDR4@2133MHz. I plan on offering package deals of 4GB of ram per server core, so I would need to upgrade to 80GB (probably going to get 96 because it is an even multiple of that 32 that is already in the system). If I run only 1 CPU, that would give me just 6 RAM slots, meaning I would have to purchase 16GB sticks, and possibly having to sell the RAM that is already in the system, since these cheap ebay listers usually just stuff these servers full of small capacity sticks. I suppose regardless I should buy higher capacity DIMMs if I am planning on eventually upgrading the server to have 40 cores with 192Gb RAM or more.

[jaslion]

Before you buy anything I'd say first get things up and running. Adding hardware later is easy.

[NPDPdev]

4 minutes ago, jaslion said:

Before you buy anything I'd say first get things up and running. Adding hardware later is easy.

I agree, however shipping alone will be 1-2 weeks (next Friday or later), and so I would really like to have the new CPU ordered and shipped by that time to avoid having to wait even longer.

[jaslion]

1 minute ago, NPDPdev said:

I agree, however shipping alone will be 1-2 weeks (next Friday or later), and so I would really like to have the new CPU ordered and shipped by that time to avoid having to wait even longer.

I assume you are starting a business? If so the first thing you always need to do is see if the current hardware is good enough before you make the wrong choices.

[Electronics Wizardy]

One thing to note is using one cpu often disables some pcie slots, so how many slots are you using?

 

4 hours ago, NPDPdev said:

am planning to use this server to offer up virtualized computing to customers

What are your customers running? How big are these vms?

 

Normally Id stay far away from this, there are many companies that are much better at doing this than you can ever be.

 

4 hours ago, NPDPdev said:

, how many system resources should I allocate to the host OS/hypervisor

Normally you don't give resources do the host.

 

Normally you over provision the vms, so there is more vcpus than physical cores or threads on the host, as most vms are idle.

 

Id just use the stock cpu for now.

[NPDPdev]

4 hours ago, anodos said:

I'd get the second E5-2630. If a CPU or CPU slot goes bad, you can pull one and probably still get it to boot. Having two CPUs for the same load will spread the heat around better than pummeling a single CPU.

 

Instead of upgrading the CPU later, spend the money on another server instead. With only one server, the customers are just down hard until you fix it. With two, you can migrate them to the working server while you fix the broken one. A minimum of three servers should be your goal, provisioned such that you can fully empty the broken one of customers without overloading the other two.

 

If you upgrade the RAM, keep the DIMMs you pull out instead of selling them. DIMMs go bad and you'll want spares.

Thank you for the insight. I am fully planning on running a minimum of three of near-identical PowerEdge r430s. I will probably use one of them as a personal server w/o important things on it, so that if needed, I can migrate clients to it as a failsafe. I suppose I will keep the DIMMs, however I am planning on using 32gb ones, as that is the only way to achieve max capacity on the server, so I would prefer to keep it uniform even in the event of a failure. I will likely keep them around for troubleshooting though, as keeping spare 2133MHz 32gb DDR4 DIMMs is a bit out of my price range lol.