nVidia Grid, Tesla P40, M10 GPU over CPU Performance
While at VMWorld, I learned more about nVidia’s offerings for server GPUs. Especially in companies where running and working with graphics is important. Instead of installing an expensive desktop at a person’s workstation, the Tesla series of GPU accelerators can give you the horsepower you need.
nVidia Tesla P40 GPU
With 24 GB of GDDR5, and a memory bandwidth of 346 GB, the P40 will give performance to high-performance need. Each GPU has 47 Tera-Operations per Second, a server with 8 P40’s will deliver the same performance of 140 cpu servers.
nVidia Tesla M10
This board allows for up to 64 users at a time, or 128 with 2 boards per server. Utilizing 2560 CUDA cores (640 per GPU), programs such as AutoCAD, Photoshop, Autodesk MAYA, ArcGIS, and more will deliver without buying another desktop machine.
nVidia at VMWorld 2017 (the video)
We saw how nVidia Grid with quadro VDWS workstation can perform better than a CPU based workstation. We saw an 8K video render in VMWare’s ESXI using a 40 core with 80 Hyperthreads, without dropping frames. We also saw Solidworks Visualise, with 2 renders of an Audi A8 design. Both of these renders are being performed on one server.
Finally, we see the vRealize operations manager. This dashboard can manage Esx servers, they can also see the GPU resource utilization.
Each Tesla GPU has a dedicated decode engine for AI-based Video Services. Check out nVidia.com/trygrid to test Grid in a VMWare, or Citrix environment.