Home > Servers > Rack and Tower Servers > AMD > White Papers > Workload-Based DDR5 Memory Guidance for Next-Generation PowerEdge Servers > Introduction to test scenarios
To better understand the impact of different memory configurations, we selected five different workloads to measure their performance, performance per watt, and performance per dollar in both 1-socket and 2-socket versions. We used the following workloads and benchmarks to test performance:
Table 1. Workloads and benchmarks used for testing
Workload | Selected benchmark |
System memory (RAM) – memory bandwidth | |
CPU-based AI inference | |
Computer vision and 3d rendering | |
Structured data management and analytics (medium) | |
Web serving and client tier |
The study looks at benchmark performance and bottlenecks for balanced configurations, using 12 DIMMs per socket, and near-balanced configurations, with 8 DIMMs per socket and with 6 DIMMs per socket, with 1 DIMM per channel and different memory sizes. We ran the benchmarks on the following systems:
Figure 4. PowerEdge R7615
Figure 5. PowerEdge R7625
The test results aim to give insights into how different workloads based on available benchmarks affect server behavior. Consider the recommendations in this paper along with your requirements to right-size your servers.
For direct comparison, because PowerEdge R7625 supports only 1 DIMM per channel (1DPC), we only look at 1DPC configurations.
We used default BIOS configurations for this test.