Artificial Intelligence Computing Leadership from NVIDIA
Horizon RDSH Density
We currently have a Horizon RDSH farm that supports ~800 CCU. There are 32 servers across 2 cloud pods to accommodate the ~800 CCU at ~25 users per virtual server. The current environment runs ok and there are no graphics cards. We decided that we would venture into testing some vGPU in different use cases so we bought 8 new Cisco UCS B200 M5 blades with 2 x I6248R CPU, 512GB of RAM and a single P6 in each blade. All blades are connected to a Pure x70 all flash array. My question is, would it be possible to run the same ~800 CCU RDSH farm on these new blades and if so what would be feasible profiles for the P6s? Do we build the same number of servers and assign them the 4A profile or can we get by with a 2A or 1A profile given that the current environment doesn't have graphics at all? Since we are new to this and bought this setup for testing purposes for various applications I am just curious if it is possible to run our largest farm with vGPU and maintain our user density.
We currently have a Horizon RDSH farm that supports ~800 CCU. There are 32 servers across 2 cloud pods to accommodate the ~800 CCU at ~25 users per virtual server. The current environment runs ok and there are no graphics cards. We decided that we would venture into testing some vGPU in different use cases so we bought 8 new Cisco UCS B200 M5 blades with 2 x I6248R CPU, 512GB of RAM and a single P6 in each blade. All blades are connected to a Pure x70 all flash array.

My question is, would it be possible to run the same ~800 CCU RDSH farm on these new blades and if so what would be feasible profiles for the P6s? Do we build the same number of servers and assign them the 4A profile or can we get by with a 2A or 1A profile given that the current environment doesn't have graphics at all?

Since we are new to this and bought this setup for testing purposes for various applications I am just curious if it is possible to run our largest farm with vGPU and maintain our user density.

#1
Posted 10/02/2020 04:20 PM   
Hi ADT The B200 M5 supports 2x P6 GPUs and it's worth running that configuration so you can maximise the density. You should be looking at either 8A or 16A vGPU Profiles meaning that with 2x P6 installed, you'll have 2x (16A) or 4x (8A) RDSH VMs per Blade. Don't use anything less than 8A with RDSH VMs in a production environment. The following link is targeted at Citrix XenApp, but it relates directly to your use case as well. Have a read all the way through, as I believe you'll find the information useful in understanding what kind of configurations to go for and the pitfalls of sizing the vGPU Profiles incorrectly: https://gridforums.nvidia.com/default/topic/10042/xenapp/looking-for-advice-on-optimal-config-for-latest-gen-citrix-xenapp-vgpu-solution/ Regards MG
Hi ADT

The B200 M5 supports 2x P6 GPUs and it's worth running that configuration so you can maximise the density. You should be looking at either 8A or 16A vGPU Profiles meaning that with 2x P6 installed, you'll have 2x (16A) or 4x (8A) RDSH VMs per Blade. Don't use anything less than 8A with RDSH VMs in a production environment.

The following link is targeted at Citrix XenApp, but it relates directly to your use case as well. Have a read all the way through, as I believe you'll find the information useful in understanding what kind of configurations to go for and the pitfalls of sizing the vGPU Profiles incorrectly: https://gridforums.nvidia.com/default/topic/10042/xenapp/looking-for-advice-on-optimal-config-for-latest-gen-citrix-xenapp-vgpu-solution/

Regards

MG

#2
Posted 10/05/2020 04:06 PM   
Scroll To Top

Add Reply