System memory bandwidth based on the number of DIMMs for DELL PowerEdge R770 servers.
Summary
The Dell PowerEdge R770 server with Intel Xeon 6 processors delivers significant improvements in DDR5 memory bandwidth for today's workloads. These 17th generation Dell PowerEdge rack servers are available in dual-socket, 2U form factors for mass market use.
In dual-socket systems, such as the 2U Dell PowerEdge R770, each socket supports 8 memory channels, for a total of 16 channels and up to 32 DIMMs, while single-socket systems, such as the 1U Dell PowerEdge R570, support 8 channels and up to 16 DIMMs. Both platforms use DDR5 RDIMMs with speeds up to 6400 MT/s in a single DIMM per channel (1DPC) configuration and 5200 MT/s in a dual DIMM per channel (2DPC) configuration.
The article compares memory throughput performance measured on these new servers with different memory configurations to previous-generation servers powered by Intel Xeon 5 processors.
Intel Xeon 6 Memory Architecture
Intel Xeon 6 processors improve on DDR5 memory support, first introduced in Xeon 5. With a disaggregated chiplet architecture with separate compute and I/O tiles, memory operates at speeds up to 6400 MT/s, which is approximately 15% better than the 5600 MT/s supported by Intel Xeon 5 processors.
Memory bandwidth test
The STREAM memory test is a synthetic benchmark designed to measure sustained memory throughput, which is the speed at which data can move to and from memory under real-world conditions. It is widely used in high-performance computing (HPC), artificial intelligence, and machine learning environments, where memory bandwidth is often the limiting factor. The STREAM test assesses how efficiently a system can handle large data transfers, which is critical for workloads that process large data sets.
First, we'll look at memory performance on the PowerEdge R770. This data covers different memory channel groups with 32GB DIMMs running at a maximum frequency of 6400 MT/s.
It is important to note that memory speed is affected by the number of DIMMs per channel (DPC).
On Intel Xeon 6 platforms:
- 1DPC configurations support speeds up to 6400 MT/s.
- 2DPC configurations typically operate at 5200 MT/s.
On Intel Xeon 5 platforms:
- 1DPC configurations support speeds up to 5600 MT/s.
- 2DPC configurations typically operate at 4400 MT/s.
The data presented here was collected with BIOS settings configured, comparing SNC (Sub NUMA Cluster) settings with and without 2DPC. The following systems were tested with 2DPC:
- The latest generation Dell PowerEdge R770 is based on Intel Xeon 6 processors, equipped with up to 16 DDR5 6400 MT/s DIMMs per socket.
- The previous generation Dell PowerEdge R760, powered by Intel Xeon 5 processors, and equipped with up to 16 DDR5 5600 MT/s DIMMs per socket.
Note:
- Fully balanced : All available memory channels are populated with DIMMs, maximizing bandwidth and capacity.
- Almost balanced : most memory channels are populated in a way that approaches balance, offering near-optimal bandwidth with somewhat reduced capacity or symmetry.

Figure 1. Trends in aggregate system memory bandwidth by DIMM count for Intel Xeon 6 processor-based PowerEdge servers using customized BIOS settings and SNC enabled.

Figure 2. Trends in aggregate system memory bandwidth by DIMM count for Intel Xeon 6 processor-based PowerEdge servers using customized BIOS settings and SNC disabled.

Figure 3. Trends in aggregate system memory bandwidth with DIMM population for Intel Xeon 5 processor-based PowerEdge servers using configured BIOS settings and SNC enabled.

Figure 4. Trends in aggregate system memory bandwidth by DIMM count for Intel Xeon 5 processor-based PowerEdge servers using customized BIOS settings and SNC disabled.
Note: Dell Intel Xeon 5/6 based servers do not support unbalanced memory configurations. DIMMs installed in unbalanced channels will be automatically disabled.
Key findings
These results demonstrate a 20.6% performance increase over the previous generation in either a fully populated or balanced memory configuration. Near-balanced configurations showed an average performance increase of nearly 21% over Intel Xeon 5. This improvement contributes to the consistently high level of system performance that customers have come to expect from Dell PowerEdge Intel-based server solutions.
This data reflects the continued advancement of DDR5 memory technology and Dell Technologies’ expertise in the world-class Intel processor platform. Regardless of the memory configuration selected for your Dell Intel Xeon 6 rack solution, enterprise-class performance remains consistent across all configurations.
The following optimized BIOS configuration was used to achieve these results:
| Settings | Value/Status |
|---|---|
|
ADDDC settings |
Disabled |
|
Corrected ECC SMI memory |
Disabled |
|
DIMM self-healing (repair after packaging) is enabled Unrecoverable memory error |
Disabled |
|
Memory paging policy |
Closed paging |
|
Virtualization technology |
Disabled |
|
NUMA subcluster |
Enabled |
|
LLC "Preliminary Selection" |
Enabled |
|
x2APIC mode |
Disabled |
|
Dell Controlled Turbocharger |
Enabled |
|
System profile |
Custom |
|
Processor power management |
Maximum performance |
|
C1E |
Disabled |
|
States C |
Off-line |
|
Energy efficiency policy |
Productivity |
|
Processor Interprocessor Bus Power Management |
Disabled |
|
PCI ASPM L1 Channel Power Management |
Disabled |
|
Thermal profile |
Maximum performance (Optimized for performance) |
|
Fan speed offset |
Maximum (+100%) |
|
Minimum speed FDan |
Custom |
|
Minimum fan speed (PWM) range X - 100 percent) |
100 |
Server configurations
| Server model | Processor |
Memory |
|---|---|---|
|
PowerEdge R770 |
Intel Xeon 6787P |
32GB 2R RDIMM 6400 MT/s |
|
PowerEdge R760 |
Intel Xeon 8592+ |
32GB 2R RDIMM 5600 MT/s |
Testing conducted by Dell Technologies on PowerEdge R770 with Intel Xeon 6 6787P processors and Dell PowerEdge R760 with Intel Xeon 5 8592+ processors. Results may vary in individual cases.
- Sub-NUMA Clustering (SNC) improves memory throughput by reducing latency through better locality awareness. When enabled, NUMA-aware workloads can achieve higher throughput because memory access is optimized within smaller clusters.
