AI applications range from image and video analytics to sensor analysis and more. They can be applicable to industries such as retail, manufacturing, health care, and smart cities. Sizing the PowerEdge server for edge deployments depends on the AI applications that you plan to run.
We use intelligent video analytics from Deep Vision as an example to size the PowerEdge server. The following table provides three configurations and a sample scenario for which the configuration might be applicable:
Table 3. Recommended configurations
Scenario | 6 to 7 video processing streams for facial recognition | 10 to 12 video processing streams for facial recognition and people counting | 20 video processing streams for facial recognition, people counting, and vehicle identification |
PowerEdge server model | R7515 | R740 | R7525 |
Processor | AMD EPYC 7502P 2.5 GHz, 32C/64T, 128 M Cache (180 W) | Intel Xeon Gold 6252 2.1 G, 24C/48T, 10.4 GT/s, 35.75 M Cache, Turbo, HT (150 W) DDR4-2933 | AMD EPYC 7452 2.35 GHz, 32C/64T, 128 M Cache (155 W) DDR4-3200 |
Memory | 8 x 8 GB | 8 x 16 GB | 16 x 16 GB |
GPUs | 1 NVIDIA T4 GPU | 2 NVIDIA T4 GPUs | 4 NVIDIA T4 GPUs |
Network | Broadcom 5720 Quad Port 1 GbE BASE-T, rNDC | Broadcom 5720 Quad Port 1 GbE BASE-T, rNDC | Either of the following: - 1 GbE: Broadcom 5720 Quad Port 1 GbE BASE-T, rNDC
- 25 GBE: Mellanox ConnectX-5 Dual Port 10/25GbE SFP28 Adapter, PCIe Low Profile
|
Storage | 6 x 480 GB SAS SSDs in RAID 6 | 8 x 480 GB SAS SSDs in RAID 6 | 12 x 480 GB SAS SSDs in RAID 6 |
Trusted Platform Module | Trusted Platform Module 2.0 | Trusted Platform Module 2.0 | Trusted Platform Module 2.0 |
iDRAC | iDRAC Enterprise | iDRAC Enterprise | iDRAC Enterprise |
These configurations are a basis and your Dell sales representative can customize the server configuration, including processing, memory, and disk, to meet your requirements. Multiple factors can drive these configurations. In video analytics, these factors include type and number of video analytics modules used, number of cameras, resolutions of cameras, and the nature of the video being analyzed. For guidelines for configuring the edge systems, see NGC-Ready Recommended Configurations.
For large deployments at a single location, choose between a scale-up server configuration and a scale-out configuration (for example, a single PowerEdge server with several T4 servers compared to multiple servers each with one or two GPUs). Cost, resiliency, and power requirements of the AI application help drive the decision.
