RDS VM - How much RAM and CPU core needed?

[Ertech]

What is the minimum/average RAM and CPU core requirement for Windows Remote Server 2016 - looking to have 30 simultaneous active sessions (usually running Word, Outlook, Excel, Browsing Internet)?

 

Thanks

Edited March 10, 2020 by Ertech

[CHiLL]

I have 4 CPU cores and 24GB RAM on Server 2019 - and it's not a great experience with a handful of users on, let alone 30.

[K-ITS]

I was told a while ago for RDS to allocate 2GB for each active session so 30 users = 60Gb Ram. It might actually be more now as I havent played with Server 2019.

[localzuk]

I run 2x 8 vCPU VMs, each with 16GB RAM and that works well for 30 computers. So near enough 1GB per session for basic usage.

[PotNoodleTech]

16 cores and 64GB of ram.

 

Without spending a fortune on servers, RDS provides a significantly worse user experience than normal PCs.

[HTCPCP]

ours is currently a 6 core, 12 thread 5+year old server with 12 GB RAM - going up to 64GB this week to tide it over incase we have everyone remote working. 35 users set up who use it somewhat regularly, and its not exactly a pleasant experience at the moment. Been pushing for a 32 core with 128 GB RAM expecting 24 core 64 GB to be the one we end up getting (although that is now looking like its going to be pushed back yet again, because a new IT manager has started and doesn't like having servers...)

[CHiLL]

The problem with allocating loads of cores is that the VM has to wait for that amount of cores to be free at once before it can process the instruction. Whilst that doesn't sound like an issue, I recently ran into an issue where each of our host CPUs had wait times of around 4000ms, due to each server being allocated more cores than necessary. We'd configured all servers with a minimum of 4 vCPUs, with SQL servers and RDS, configured as much as 16 vCPUs. I was up to about ~90 allocated vCPUs across our two hosts, whereas the hosts have 32 physical cores combined. Whilst cores aren't directly comparable between physical and virtual, reducing the total vCPU count from ~90 down to ~40 has massively reduced the wait time for the CPUs to about 300~600ms. Performance on the servers don't appear to be affected, but overall performance should have increase on the hosts themselves.

 

I remember a remote support company guy saying that you should allocate up to 1.5x the amount of physical cores you have. So 32 physical cores should have up to a max 48 vCPUs assigned.

Edited March 10, 2020 by CHiLL

[mavhc]

Surely everyone logging in is using the same VM though, so that doesn't matter.

 

Are people getting a remote desktop, or are you publishing the apps with RemoteApp? And are you using vGPUs?

[Ertech]

is it reasonable to say min 1GB RAM and 1vCPU per session/user? so if we have 30 active session we are looking at min 30GB RAM and 30vCPU (15 core)

[localzuk]

Surely everyone logging in is using the same VM though, so that doesn't matter.

 

it does matter, as it is applicable across the entire host, not the VM alone. So, if you have 20 VMs, all oversubscribed to vCPUs, they'll all be fighting for them - leading to those delays.

Are people getting a remote desktop, or are you publishing the apps with RemoteApp? And are you using vGPUs?

Traditional RDSH desktop here, and no, no vGPUs.

 

 

 

is it reasonable to say min 1GB RAM and 1vCPU per session/user? so if we have 30 active session we are looking at min 30GB RAM and 30vCPU (15 core)

Yeah, that seems reasonable!