AMD EPYC™ 9006 Series Servers for the Agentic AI Era

Powered by up to 256 cores per socket and up to 1.6TB/s of memory bandwidth, MSI server platforms with AMD EPYC™ 9006 Series processors scale AI agents and workflows across cloud, enterprise, and HPC environments.

AMD EPYC™ 9006 Series Servers for the Agentic AI Era

Key Platform Advantages

Expanded Memory Bandwidth - icon

Expanded Memory
Bandwidth

Up to 16 channels of DDR5 / MRDIMM-12800 memory deliver up to 1.6TB/s per-socket bandwidth, helping reduce bottlenecks for AI agents, HPC simulations, and data-intensive cloud workloads.

Next-Generation I/O - icon

Next-Generation
I/O

PCIe Gen6 supports up to 64Gbps per lane and up to 128 lanes per socket, doubling CPU-to-GPU bandwidth while expanding I/O headroom for networking and NVMe storage.

Efficiency for Dense Infrastructure - icon

Efficiency for
Dense Infrastructure

Higher performance per watt helps data centers scale workload density while planning around power and thermal constraints.

SP7 and SP8 Platforms for Different Workload Demands

MSI offers AMD EPYC 9006 Series-based platforms across SP7 and SP8 designs, giving customers workload-optimized choices for AI, cloud, HPC, enterprise, storage, and virtualization deployments.

SP7 Platforms :

Maximum Density for AI, Cloud, and HPC

Up to 256 Cores
Up to 512 Threads
1.7x Per-socket Perf
vs. AMD 9005 Series
CD281-S4091-X2

CD281-S4091-X2

  • 2OU2N Open Compute ORv3 21” server, DC-MHS architecture
  • Single AMD EPYC™ 9006 SP7 processor, up to TDP 600W, per node
  • (16) DDR5 DIMM slots, 16 channels (1DPC), up to 8,000 MT/s RDIMM and 12,800 MT/s MRDIMM, per node
  • (1) PCIe 6.0 x16 OCP3.0 slot (NCSI supported) per node
  • (2) 2280/22110 PCIe 4.0 x2 NVMe M.2 ports per node
  • Per node:
  • (1) PCIe6.0 x16 FHHL single wide slot
  • (1) PCIe6.0 x16 FHFL single wide slot, supporting max 160W PCIe card
  • (8) Front E3.S 1T PCIe 6.0 x4 NVMe per node
CD270-S4091-X2

CD270-S4091-X2

  • 2U2N Core Compute AMD SP7 server, DC-MHS architecture
  • Single AMD EPYC™ 9006 SP7 processor, up to TDP 600W, per node
  • (16) DDR5 DIMM slots, 16 channels (1DPC), up to 8,000 MT/s RDIMM and 12,800 MT/s MRDIMM, per node
  • (1) PCIe 6.0 x16 OCP3.0 slot (NCSI supported) per node
  • (2) 2280/22110 PCIe 4.0 x2 NVMe M.2 ports per node
  • (1) PCIe6.0 x16 FHFL single wide slot
  • (8) Front E1.S PCIe 6.0 x4 NVMe per node
D4091

D4091

  • DC-MHS SDNO-A335 HPM
  • 335mm x 210mm
  • Single AMD EPYC™ 9006 SP7 processor, up to TDP 600W
  • (16) DDR5 DIMM slots, 16 channels (1DPC), up to 8,000 MT/s RDIMM, and 12,800 MT/s MRDIMM
  • (1) PCIe 6.0 x16 OCP 3.0 slot (NCSI supported)
  • (2) PCIe 6.0 x16 MXIO Multi-Trak expansion slots
  • (6) PCIe 6.0 x8 MCIO-8i connectors
  • (1) PCIe 4.0 x4 MCIO-4i connector
  • Management Port DC-SCM2 slot for MGT3 module with Aspeed AST2700 BMC


Cloud Infrastructure

Cloud Infrastructure

Run more virtualized workloads and tenants per server, improving compute density without expanding rack footprint.

High-End AI Inference

High-End AI Inference

Keep large models fed with data at the rate high-throughput inference demands, reducing bottlenecks as AI agent workloads scale.

HPC

HPC

Accelerate parallel, compute-intensive workloads across scientific computing, simulation, and research environments.

SP8 Platforms :

Balanced Architecture for Enterprise and Storage Workloads

Up to 128 Cores
Up to 256 Threads
Up to 128 PCIe Gen6 Lanes
Planned CD281-S4081-X2

CD281-S4081-X2

  • 2OU2N Open Compute ORv3 21” server, DC-MHS architecture
  • Single AMD EPYC™ 9006 SP8 processor, up to TDP 400W, per node
  • (16) DDR5 DIMM slots, 8 channels (1DPC), up to 8,000 MT/s RDIMM and 11,200 MT/s MRDIMM, per node
  • (1) PCIe 6.0 x16 OCP3.0 slot (NCSI supported) per node
  • (2) 2280/22110 PCIe 4.0 x2 NVMe M.2 ports per node
  • (1) PCIe6.0 x16 FHFL slot per node
  • (1) PCIe6.0 x16 FHHL slot, per node
  • (12) Front E3.S 1T PCIe 6.0 x4 NVMe per node


Enterprise IT

Enterprise IT

Run day-to-day business applications efficiently, without over-provisioning for workloads that don't require maximum core density.

Virtualization

Virtualization

Run more VMs per host with the memory and I/O headroom to improve consolidation ratios.

Database

Database

Keep data moving at the rate transactional and analytical workloads demand, supporting consistent query performance as data volumes grow.

Enterprise AI has moved from experimentation to production, and agentic AI will place even greater demands on compute, memory bandwidth, efficiency and system-level orchestration. With 6th Gen AMD EPYC Server CPUs, AMD is helping customers scale enterprise AI with the performance and efficiency needed to support increasingly complex data pipelines, AI services and intelligent applications across the modern data center.

Dan McNamara

Senior Vice President and General Manager, Compute and Enterprise AI, AMD

Planning Your Agentic AI Infrastructure?

Find the right AMD EPYC 9006 Series platform to support AI agents, data pipelines, and modern data center workloads.

Talk to a Server Expert