AMD Pensando™ Giglio DPU
The AMD Pensando™ Giglio DPU builds on second-generation architecture with enhanced power and performance efficiency, delivering dual 200 Gbps line-rate acceleration for data center networking, storage, and security.
High-Performance Front-End Networking
Purpose-built for the high-volume data movement that training, inference, and agentic AI workloads demand, the fully programmable AMD Pensando™ DPU has been widely adopted and validated by some of the largest hyperscale data centers.
Use Cases
Multi-tenant SDN and Virtual Networking
Enable improved performance for demanding AI applications by providing essential Software Defined Networking (SDN).
Security Service Acceleration and Comprehensive Observability
Tap into granular visibility into network, storage, and security operations for fast troubleshooting, compliance, and performance optimization.
Storage Acceleration & KV-Cache Optimization
Accelerate AI storage and KV-cache access to support larger context windows, higher agent density, and continuous inference for next-generation agentic AI workloads.
Front-end Networking for AI Clusters
Leverage front-end networking to accelerate and manage data movement between compute, storage, and external networks, enabling high throughput and low latency for demanding AI workloads.
Performance Benchmarks
~1.45x
Performance over NVIDIA BlueField-3¹
AMD Pensando™ Salina DPU can help deliver more efficient network traffic management.
~2x
Performance generation over generation²
AMD Pensando™ DPUs are setting a new standard for data center throughput and scalability.
Portfolio
AMD Pensando™ Salina DPU
The AMD Pensando™ Salina DPU is fully P4 programmable and optimized for minimal latency, jitter, and power requirements. The multi-generational hardware and software stack is deployed at hyperscale for optimal front-end performance.
AMD Helios Rack-Scale Solution
Leadership Rack Performance for Hyperscale AI3
The AMD Helios Rack-scale solution design is a fully integrated AI infrastructure, combining the latest AMD Instinct™ GPUs, AMD EPYC™ Server CPUs, and AMD Pensando™ networking, designed using open industry standards enabling large-scale inference, frontier model training and fine-tuning.
AMD Pensando™ DPUs in the Spotlight
Customer Experiences with AMD Pensando™ DPUs
DXC Technology Sees TCO Savings of $65M4
DXC estimates savings in the tens of millions thanks to AMD Pensando™ DPUs in the HPE Aruba CX 10000 Series Switch, driving down costs for customers and themselves.
IBM Cloud Saw Bare-Metal and VSI TCO Savings5
Microsoft Azure Saw 100X CPS Gains6
“AMD Pensando™ helped us with CPS. Customers just love it. Now we’re world leaders in CPS performance.”
HPE Aruba: Rethinking Data Centers
The HPE Aruba CX 10000 Series Switch with integrated AMD Pensando™ technology creates a new way to build network fabrics.
Partner Ecosystem
- Infrastructure Solutions
- Storage
- Switching
Resources
Contact Us
Unlock the Future of AI Networking
Learn how AMD Pensando™ DPU technology can transform front-end networking for the modern data center.
Explore Data Center Networking
Explore the full suite of networking solutions designed for high-performance modern data centers.
Footnotes
- PEN-017: Testing conducted by AMD Performance Labs as of 15th April 2025 on the AMD Pensando Salina DPU, on a test system comprising of 2x Dual socket Xeon Dell power edge XE9680 function; Cisco 64x400G Switch; IXIA 2x400G tester as a traffic generator from Keysight; 2xAMD Pensando™ Salina DPU; 5th gen Xeon 8568 - 48 core CPU with PCIe Gen-5; Operating System Version : Ubuntu® 22.04.5 LTS; Kernel Version: 5.15.0-139-generic; BIOS version: 1.3.6 [Mitigation: Off (default), System profile setting: Performance (default), SMT: enabled (default)] AMD Pensando performs an average of 117 MPPS (millions packets per second) in AMD testing while Nvidia Bluefield-3 published performance is 80 MPPS. https://hc33.hotchips.org/assets/program/conference/day1/HC2021.NVIDIA.IdanBurstein.v08.norecording.pdf Slide 6 for ~1.45x the performance with AMD Pensando Salina DPU.
Results may vary based on factors including but not limited to system configuration and software settings.
- PEN-012A: Measurements conducted by AMD Performance Labs as of April 15th, 2025 on the current specification for the AMD Pensando™ Salina DPU accelerator designed with AMD Pensando™ 5nm process technology. The performance numbers are based on SDN Cloud Gateway pipeline in the BITW (bump-in-the-wire):
Salina (3rd gen DPU)
SDN PPS - 117MPPS
SDN BW - 782G
SDN encryption PPS - 100MPPS
SDN encryption BW - 767G
Elba (2nd gen DPU)
SDN PPS - 52 MPPS
SDN BW - 344G
SDN encryption PPS - 42M
SDN encryption BW - 288G
Salina test system configuration:
2P Intel Xeon 8568 powered production server, Ubuntu 22.04.5 LTS, Cisco N96K Series 64x400G switch, Kernel Version : 5.15.0-139-generic; BIOS version 1.3.6 ; Mitigation - Off (default); System profile setting - Performance (default);
SMT- enabled (default)
Elba test system configuration:
2P EPYC 7302 powered production server, Oracle Linux 9.7, Kernel Version: 6.12.0-201.74.2.1 el9uek_x86_64, BIOS version 2.25, Mitigation - off (default), system profile setting - performance (default), SMT - enabled (default).
Results may vary by system configuration and other factors.
- Calculations by AMD Performance Labs in June 2025, based on the projected memory capacity/ bandwidth and scale up/out bandwidth specifications of AMD Instinct™ MI455X 72xGPU “Helios” AI Rack vs. the publicly announced NVIDIA “Vera Rubin” 72xGPU “Oberon” Rack. Server manufacturers may vary configurations, yielding different results. MI350-045A
Calculations by AMD Performance Labs in September 2025, based on the FP8/FP4 datatypes and the projected specifications for AMD Instinct™ MI455X 72xGPU “Helios” AI Rack vs. Publicly announced specs for the NVIDIA “Vera Rubin” 72xGPU “Oberon” AI Rack. Actual results based on production silicon may vary. Server manufacturers may vary configurations, yielding different results. MI350-046B
- Jayant Tulsiani, DXC VP Cloud and Infrastructure Services: “Previously what required hundreds of virtual and physical firewalls to meet our segmentation and compliance requirements, is now being delivered natively inline on the platform, which has been transformation to the TCO model of our business, resulting in projected TCO savings of ~$65M” -- published on amd.com: https://www.amd.com/en/solutions/data-center/data-center-ai-premiere/industry-support.html
- Satinder Sethi, GM, IBM Cloud Infrastructure Services: “By collaborating with AMD and utilizing their programmable DPU, and turnkey SDN and security services software, IBM Cloud has been able to accelerate to market our award-winning secure high-performance “bare-metal servers for VPC” offering for VMware, Red Hat and SAP workloads, driving significant network throughput improvements for customers, and realizing 30% TCO savings from greater workload density and preserved CPU cores.” -- published on amd.com: https://www.amd.com/en/solutions/data-center/data-center-ai-premiere/industry-support.html
- Gerald De Grace, Lead, Microsoft Azure Accelerated Connections: “Our collaboration with AMD using the AMD Pensando DPU based smart appliance and full-stack AMD software in production for Accelerated Connections has resulted in 100X improvement in CPS performance over our existing solution and will extend to broader tiers within the Azure network in the future.”
Footnotes
- PEN-017: Testing conducted by AMD Performance Labs as of 15th April 2025 on the AMD Pensando Salina DPU, on a test system comprising of 2x Dual socket Xeon Dell power edge XE9680 function; Cisco 64x400G Switch; IXIA 2x400G tester as a traffic generator from Keysight; 2xAMD Pensando™ Salina DPU; 5th gen Xeon 8568 - 48 core CPU with PCIe Gen-5; Operating System Version : Ubuntu® 22.04.5 LTS; Kernel Version: 5.15.0-139-generic; BIOS version: 1.3.6 [Mitigation: Off (default), System profile setting: Performance (default), SMT: enabled (default)] AMD Pensando performs an average of 117 MPPS (millions packets per second) in AMD testing while Nvidia Bluefield-3 published performance is 80 MPPS. https://hc33.hotchips.org/assets/program/conference/day1/HC2021.NVIDIA.IdanBurstein.v08.norecording.pdf Slide 6 for ~1.45x the performance with AMD Pensando Salina DPU.
Results may vary based on factors including but not limited to system configuration and software settings. - PEN-012A: Measurements conducted by AMD Performance Labs as of April 15th, 2025 on the current specification for the AMD Pensando™ Salina DPU accelerator designed with AMD Pensando™ 5nm process technology. The performance numbers are based on SDN Cloud Gateway pipeline in the BITW (bump-in-the-wire):
Salina (3rd gen DPU)
SDN PPS - 117MPPS
SDN BW - 782G
SDN encryption PPS - 100MPPS
SDN encryption BW - 767G
Elba (2nd gen DPU)
SDN PPS - 52 MPPS
SDN BW - 344G
SDN encryption PPS - 42M
SDN encryption BW - 288G
Salina test system configuration:
2P Intel Xeon 8568 powered production server, Ubuntu 22.04.5 LTS, Cisco N96K Series 64x400G switch, Kernel Version : 5.15.0-139-generic; BIOS version 1.3.6 ; Mitigation - Off (default); System profile setting - Performance (default);
SMT- enabled (default)
Elba test system configuration:
2P EPYC 7302 powered production server, Oracle Linux 9.7, Kernel Version: 6.12.0-201.74.2.1 el9uek_x86_64, BIOS version 2.25, Mitigation - off (default), system profile setting - performance (default), SMT - enabled (default).
Results may vary by system configuration and other factors. - Calculations by AMD Performance Labs in June 2025, based on the projected memory capacity/ bandwidth and scale up/out bandwidth specifications of AMD Instinct™ MI455X 72xGPU “Helios” AI Rack vs. the publicly announced NVIDIA “Vera Rubin” 72xGPU “Oberon” Rack. Server manufacturers may vary configurations, yielding different results. MI350-045A
Calculations by AMD Performance Labs in September 2025, based on the FP8/FP4 datatypes and the projected specifications for AMD Instinct™ MI455X 72xGPU “Helios” AI Rack vs. Publicly announced specs for the NVIDIA “Vera Rubin” 72xGPU “Oberon” AI Rack. Actual results based on production silicon may vary. Server manufacturers may vary configurations, yielding different results. MI350-046B - Jayant Tulsiani, DXC VP Cloud and Infrastructure Services: “Previously what required hundreds of virtual and physical firewalls to meet our segmentation and compliance requirements, is now being delivered natively inline on the platform, which has been transformation to the TCO model of our business, resulting in projected TCO savings of ~$65M” -- published on amd.com: https://www.amd.com/en/solutions/data-center/data-center-ai-premiere/industry-support.html
- Satinder Sethi, GM, IBM Cloud Infrastructure Services: “By collaborating with AMD and utilizing their programmable DPU, and turnkey SDN and security services software, IBM Cloud has been able to accelerate to market our award-winning secure high-performance “bare-metal servers for VPC” offering for VMware, Red Hat and SAP workloads, driving significant network throughput improvements for customers, and realizing 30% TCO savings from greater workload density and preserved CPU cores.” -- published on amd.com: https://www.amd.com/en/solutions/data-center/data-center-ai-premiere/industry-support.html
- Gerald De Grace, Lead, Microsoft Azure Accelerated Connections: “Our collaboration with AMD using the AMD Pensando DPU based smart appliance and full-stack AMD software in production for Accelerated Connections has resulted in 100X improvement in CPS performance over our existing solution and will extend to broader tiers within the Azure network in the future.”