Cloud Computing Acceleration of Nvidia Bluefield-3 Dpu Smart Network Card

• Customization: Available
• Type: Wired
• Bandwidth: 400GB/S/Ndr
• Model NO.: BlueField-3 DPU
• Interface: PCI Express
• Transmission Rate: 10/100/1000Mbps
• Application: Server
• Connecting Medium: Optical Fibre
• Motherboard Integrated: Separate Network Card
• Frequency Range: -
• Type of Network: 10-Gigabit Ethernet
• Field: Server
• Virtualization Delay: <1μs (Rdma Scenario)
• Packet Processing Rate: 240m PPS (64-Byte Packets)
• Power Consumption: 75W (Full Load)
• Processor: 16-Core Cortex-A78 (Max Configuration)
• Transport Package: Carton
• Specification: Standard PCIe FHHL
• Trademark: NVIDIA
• Origin: China

Product Description

Product Description

NVIDIA BlueField-3 DPU: Powering the Line-Rate Intelligence Era

World's first 400Gb/s/NDR line-rate DPU redefines data center networking, storage & security:

Line-rate Processing: 400Gb/s smart NIC + multi-core ARM offloads traffic analysis, storage protocols & encryption without performance loss for AI/supercomputing.
Heterogeneous Platform:Dual-mode InfiniBand NDR/Ethernet 400Gb/s for seamless network compatibility.
DOCA SDK enables custom network functions (e.g., routing, traffic shaping) for cloud/edge.
NVMe-oF/GPUDirect Storage cuts latency <100μs, triples throughput vs. traditional storage.
Zero Trust Security: Hardware encryption + Secure Boot ensures end-to-end data protection, meeting finance/healthcare compliance.

network interface : 1 or 2 ports up to 400Gb/s Ethernet or NDR InfiniBand network connection.	PCI Express interface : 32-channel fifth generation PCIe;; PCIe switching splitting with up to 16 downstream ports	ARM CPU core : Up to 16 ARMv8.2+ A78 Hercules cores (64-bit)
Programmable data path accelerator ball : 16 cores and 256 threads; Realizing programmability through DOCA	Support DDR DIMM:  Dual DDR5 5600MT/s DRAM controller; 32GB onboard DDR5 memory	Storage:  Bluefieldsnap-elastic block storage; > NVMe-oF™ and NVMe/TCP™ acceleration
Network:  RoCE, zero contact roce; Speed up switching and data packet processing; VirtIO acceleration; Overlay network acceleration	HPC/AI acceleration:  HPC/AI All-to-All engine; NVIDIA GPUDirect:	Management: 1GbE out-of-band management port; NC-SI, MCTP over SMBus and MCTP over PCIe;; PLDM for monitoring and controlling DSP0248

processor	Multi-core Arm CPU	16-core Cortex-A78 (Max Configuration)
Acceleration engine	Network/Storage/Security Acceleration	Dedicated Hardware Acceleration Modules
network interface	Network Bandwidth	Dual-port 400GbE (2×200G or 1×400G)
PCIe interface	PCIe Version	PCIe Gen4 x16 (Backward Compatible with Gen3)
Memory support	DDR Memory Capacity	Up to 32GB LPDDR5
Safety characteristics	Zero-Trust Security Architecture	Root of Trust / Hardware Encryption, IPSec Acceleration, Root of Trust
Storage unloading	NVMe over Fabrics Support	Storage Virtualization & Offload
Virtualization capability	SR-IOV Virtualization	Supports 256 Virtual Functions
Network protocol support	Protocol Acceleration	RoCEv2, RDMA, OVS Offload
power consumption	Typical Power Consumption	75W (Full Load)
exploitation environment	Software Development Kit	NVIDIA DOCA 2.0 Framework
Application scenario	Target Applications	Cloud Data Centers, 5G Edge, AI Inference
Physical size	Form Factor	Standard PCIe FHHL (Full-Height Half-Length)
Authentication and compatibility	OS Compatibility	Linux/Windows Server/Hypervisor / Linux, Windows Server, Hypervisor
extended function	Programmable Data Path	P4 & FlexIO Programmability

Advantages of BlueField

End-to-End Security
Supports zero-trust security architecture, ensuring safety from data centers to edge computing.
Elastic Storage for Scalable Workloads
Leverages NVMe-oF, GPUDirect Storage, encryption, and data integrity features to deliver high-performance storage with ultra-low latency (comparable to direct-attached storage) for remote storage.
High-Performance Networking
Acts as a powerful data center accelerator, offering up to 400 Gb/s Ethernet/InfiniBand connectivity for traditional and GPU-accelerated apps while freeing host CPU cores from infrastructure tasks.
Software-Defined Infrastructure
Enables flexible, programmable infrastructure via open SDKs for custom network/storage functions (e.g., smart routing, traffic shaping).

Company Profile

Beijing Hengzheng TC Sci-Tech Co.,Ltd. is a  trusted supplier of rovider of data communication products and information security products solutions,headquartered in Beijing, China, with over 16 years of industry experience since our founding in 2009. Products including but not

 limited to:
Network cards, Optical modules, High-speed cables, Switches, and Industrial control boards.
 
Our dedicated team (50+ employees) specializes in networking, storage, and connectivity solutions, ensuring global customers receive high-performance hardware at competitive prices, and a prompt 7*24h service to keep your business guaranteed.

Packing & Delivery

FAQ

Q:Are you trading company or manufacturer?
A:We are manufacturer.
Q:How long is your delivery time?
A:Generally it is 2-5 days if the goods are in stock. or it is 10 days if the goods are not in stock, it is according to quantity.
Q:Can you offer a better price?Any discount?
A:Yes we can give you discount.But it is based on quantity.More quantity more attactive price.
Q:What is your warranty period?
A:1.Different products have different warranties.
2.comprehensive after-sales service, including quality service, guidance services,We focus on customer needs in every
detail.Please contact us if you need our detail warranty term.
Q:What's the package?
A:Original packing. For the outside, we will pack it in an empty box to protect from damaged during transportation, and ensure youwill receive the goods in good condition.