We are a K-12 school district and have High School students using workstations to do CAD work in Autodesk Inventor and Revit. These workstation are located in three different computer labs. We would like students to be able to continue their work when not in class which would include using a computer elsewhere in the building or while at home. We are in the process of purchasing new servers (either HP DL380 or Dell R730) and will be installing either M10 or M60 cards.
We have 200 student VMWare View desktops spun up and are also delivering applications via View and a few RDS servers. Our current thought is to purchase two M10 cards per server and every View or RDS session will take advantage of some GPU power. The question becomes, does Inventor and Revit run well with the M10 card or should we jump up to the M60?
We will not be replacing our workstations running this CAD software. Running the CAD on on occasional basis when a student is not in class is our goal. That and the fact that their are other applications like Photoshop, Google Earth, and streaming video that can take advantage of a GPU
The workstations that students use have a Xeon processor running at 3.2GHz and of course a nice graphics card. Our new servers are going to have Intel Xeon processors in them which is standard. The clock speed on most (not all) Xeon processors for servers does not go any higher than 2.6GHz. The servers will have 2 Intel E5-2680 v4 processors at 2.4GHz, 14 cores each. Each desktop will be getting two vCPUs, 4GB RAM, and running either Win7 or Win10. This along with the Tesla M10 cards is what we are hoping will run the Autodesk pieces just fine.
These CPUs won’t be optimal but for a class room scenario might be OK. But I would strongly recommend to use 4 vCPUs and at least a 1GB vGPU profile. What density do you expect for your configuration for one host?
The Autodesk products will be used via VMWware View only when students are not in class. In class they use HP workstations. This will occur in places like study halls and at home. I would expect only 20 or 30 students to be using these products via View at any one time. We will be using three hosts with two Tesla M10 cards in each host so the densities are low. We plan on giving every View desktop and RDS session some graphics horsepower via these cards but with 6 cards total and about 150 VDI sessions and maybe 30 RDS sessions the numbers should work. If I jump to 4 vCPUs I will need to add at least one more host.
I kept thinking that the Tesla M10 could only have a maximum vGPU profile of 512K. I have dug around and cannot find this spec.
This is the Grid vGPU users guide and on pages 3-7 it describes each card and the type of profiles they support.
Under the Tesla M10 section it appears that M10-1A might be the best choice. This supports a 1GB vGPU profile, a single monitor which is all students would ever use, and a max resolution of 1280x1024. This gives a max of 32 vGPUS per board. I will have 6 boards so this can support 192 View sessions with each one getting some GPU power including the ones where students are firing up Autodesk Inventor or Revit.
The other thing I could do is instead of using Intel E5-2680v4 processors at 2.4GHz jump to something else. These processors give me:
2 CPUs per server x 3 servers x 14 cores = 84 cores total
If I jump to Intel E5-2698 I do drop down to 2.2GHZ but the core count goes up
2 CPUs per server x 3 servers x 20 cores = 120 cores
This might let me devote more 4 vCPUs per desktop. The other real wildcard here is how I deliver the Autodesk products. Instead of adding them to my desktop image, I could deliver them as an application via my RDS farm. The conventional wisdom has always been that delivering applications via RDS uses less resources that delivering a full Win7 desktop
Hi,
as I said. Please use CPU clock frequency >3GHz. For sure you will need to overbook the CPU when using 4vCPUs but that should be fully OK. The M10-1A profile is only for RDS. You should use M10-1B (vPC license) or M10-1Q (vWS license) depending if you need Quadro feature set.
If you choose M10 cards you can consider the following sizing:
3x Host with 2M10s each, using M10-1B or 1Q = 32VMs per board = 64VMs per Host = 192VMs (VDI) in total.
As you mentioned you would like to leverage also a RDS host you could use one GPU in Passthrough/8A mode for the RDS host means 184VM (VDI) and 1 RDS host with M10-8A so you’re set.