1

Hi there

We are currently looking for new Servers for a 3 node Cluster. We have an OCA-Config with Gen10-Plus Servers but this week the new Gen11 are available. Now I want to configure a Gen11 Cluster with DL365 Gen11 Nodes to get a Quote for it from HPE. Biggest Question is about the CPU Configuration. We need a total of 768GB RAM (12x64GB RAM) and 64 Cores for High Density Virtualization Workloads. Now I need to choose between the following configurations:

  • 2 x AMD EPXC 9354, 32C 3.25 Ghz

  • 1 x AMD EPYC 9554, 64C, 3.1 Ghz

  • 2 x AMD EPYC 9374F, 32C, 3.85 GHz (Depending on price - if the qoute is not much higher then the 9354 - I think this is the preffered solution)

How is it with Memory Bandwidth/Performance on the new DDR5-Platform. Is there still more performance when using 2 CPUs instead of one?

Thank you

3 Spice ups
2

There are probably a ton of folks and websites that can speak on performance metrics. I just rebuilt a DC with single socket high core CPU’s due to licensing. A lot of the licenses were still based on CPU socket as opposed to cores and it allowed me to get more bang for the buck.

I went from a C7000 with dual cpu gen8 blades to a regular rack. Was always a full blown citrix VDI environment.

The existing infrastructure was a little antiquated so I don’t really have a solid benchmark, just an interesting thing that may or may not help you on the quest = )

3

BTW, it should be EPYC 9354 (not EPXC)

What kind of “High Density Virtualization Workloads” are you running ?

There is always licensing concerns

  • Windows OS licensing for VMs are based on the host CPU core count (min of 16)
  • DB licensing such as SQL server and/or Oracle is based on the vCPU count
  • Certain applications such as agentless AV, backup software or VMware Licensing can be based on number of Physical CPU or socket count

Then there are always performance concerns such as Bus throughput (within mobo for RAM, CPU, storage etc), NIC throughput and/or even storage performance of one host (or in your case 3 host) as compared to 2 or more (or many, probably 6) hosts ?

There are always different schools of thought between AMD vs Intel Xeon CPUs and also different comparison points (heat, power consumption, prices etc), but should always look at the solution as a whole when it comes to a virtualized environment as it is no longer about a VM, a host or the CPUs alone.

4

Any of those cpu’s would be fast. Really boils down to your licesning for prodcuts, Veeam is moving away from socket licenses but vmware and a number of other vendors. Look at this beast

HPE ProLiant DL385 Gen11 Data sheet

WOAH GEN11 MONSTER

I don’t think you would see any performance issues. I usually prefer to use 2 cpus if that is an option.

@HPE

5

I would look at it from the actual use case standpoint as performance will not be an issue with DDR5 - HPE Proliant 11 supports some of the very intensive workloads and is really robust in performance.

6

If you’re running a workload, like VDI, then that would justify the dual CPU on the HPE Gen11 Proliant servers. We purchased the HPE Gen10 servers for our VMware Horizon View environment and we had no performance issues. Like others have stated, the biggest issue for our group in the past was how the software manufacturers, like VMware and Veeam, license by CPU or Cores, that would impact how many CPUs were purchased.

7

DDR5 supports more bandwidth than its predecessor, DDR4, with 4.8 gigabits per second possible, but not shipping at launch. DDR5 has about the same latency as DDR4 and DDR3. DDR5 octuples the maximum DIMM capacity from 64 GB to 512 GB. DDR5 also has higher frequencies than DDR4.

Please check below link :-

https://www.hpe.com/in/en/servers/proliant-servers.html?jumpid=ps_k397wcd4s_aid-520074550&ef_id=CjwKCAiAkrWdBhBkEiwAZ9cdcKpn6r-lOM8S7wAlChpAVZRU4Tg1ORTOmnOyzn3I1ASncvcGmrr6gRoClCcQAvD_BwE:G:s&s_kwcid=AL!13472!3!638310119508!p!!g!!hpe%20servers!19094940889!149715037171& ;