MPN: A-U250-P64G-PQ-G·By: AMD
Ships from the US.
Order by 1:00 PM CST for overnight shipping
Buying in volume or need a PO?
Have a target price for A-U250-P64G-PQ-G? Call +1 (551) 421-5162 or request a bulk quote and our sales team will reply within one business day.
The AMD Alveo U250 Data Center accelerator card (A-U250-P64G-PQ-G) is a passive-cooled FPGA solution built on 16nm UltraScale+ architecture, designed for hyperscale datacenter environments requiring adaptable acceleration. It delivers up to 90X higher performance than CPUs and 4X inference throughput over GPUs for machine learning inference, video transcoding, database search, and analytics workloads. Key specifications include 33.3 peak INT8 TOPs, 1,728K LUTs, 64GB DDR4 memory at 77 GB/s bandwidth, 54MB internal SRAM at 38 TB/s, PCIe Gen3 x16 host interface, and dual QSFP28 100GbE network interfaces for high-speed data transfer. The full-height, 3/4-length, dual-slot form factor with 225W maximum power supports dense rack deployments. Reconfigurable FPGA fabric enables algorithm updates without hardware changes, reducing TCO through cloud/on-premises mobility and ecosystem support via Vitis development tools. Ideal for financial services risk analysis, genomics sequencing, 8K video processing, and custom datacenter acceleration where fixed-function ASICs fall short in adaptability.
| Manufacturer part # | A-U250-P64G-PQ-G |
|---|---|
| Brand | AMD |
| Condition | New |
| UltraRAM | 1,280 blocks - High bandwidth memory |
| DSP Slices | 12,288 - High precision compute blocks |
| Device Type | FPGA Data Center Accelerator Card - PCIe Dual Slot |
| Compute Performance | INT8 TOPs (peak): 33.3 ¦ LUTs: 1,728K ¦ Registers: 3,456K |
| Memory Capacity | Off-chip DDR4: 64 GB ¦ Internal SRAM: 54 MB |
|---|---|
| Memory Bandwidth | DDR4: 77 GB/s ¦ SRAM: 38 TB/s |
PCIe Gen3 x16 compatible with major OEM datacenter servers (Dell PowerEdge, HPE ProLiant, Lenovo ThinkSystem). Linux driver support via Vitis platform.
PCIe Gen4 slots backward compatible to Gen3 x16
iLO5 management with PCIe riser support
XClarity controller PCIe Gen3/4 validation
High-throughput inference for TensorFlow/PyTorch models in production serving with 4X GPU throughput and 3X lower latency using Vitis AI optimization stack.
Real-time 8K/4K video processing, live streaming transcoding, and broadcast workflows with FPGA-optimized OpenCV and FFmpeg pipelines.
In-memory database acceleration for SQL/NoSQL queries, graph analytics, and pattern matching in financial trading and cybersecurity applications.
Details
Shipping, warranty & returns
No reviews yet — be the first to review this part.
Secure SSL checkout. Visa, Mastercard, Amex and Discover accepted. ACH / wire available for businesses.
New, certified-refreshed and OEM stock with guaranteed compatibility, backed by warranty on every part.
We process quotes, POs, stock documentation and compliance invoicing.
[email protected]| Network Interfaces | 2x QSFP28 (100GbE) - Dual 100G Ethernet ports |
|---|
| Power Consumption | Maximum Total Power: 225W - PCIe + AUX power |
|---|
| Weight | 1066g |
|---|---|
| Form Factor | Full Height ¦ 3/4 Length ¦ Dual Slot Width |
| Thermal Solution | Passive Cooling - Rackmount optimized |
|---|
| PCIe Interface | Gen3 x16 - Full bandwidth host connectivity |
|---|
Real-time risk analysis, algorithmic trading, and Monte Carlo simulations leveraging 1,728K LUTs and 38 TB/s SRAM bandwidth.
DNA sequencing alignment, variant calling, and protein folding simulations with custom FPGA kernels via Vitis HLS.
5G vRAN, packet processing, and encryption acceleration with 100GbE QSFP28 interfaces for telco edge deployments.
Machine learning inference, video transcoding, database search/analytics, financial risk modeling, genomics sequencing. Delivers 90X CPU performance and 4X GPU inference throughput using reconfigurable FPGA fabric with Vitis AI optimization for TensorFlow/PyTorch models.
Dual QSFP28 ports support 100GbE with RoCEv2, InfiniBand over fiber. Enables direct rack-scale connectivity for distributed training and NFV without external switches, providing 200Gbps aggregate bandwidth.
225W maximum TDP via PCIe slot (75W) + 6-pin AUX (150W). Passive cooling solution optimized for high-CFM rack environments (45°C max operating temp) with 1066g weight in 3/4-length dual-slot form factor.
AMD Vitis unified software platform with HLS, OpenCL, and Vitis AI libraries. Ecosystem includes pre-built applications for common datacenter workloads plus custom kernel development for proprietary acceleration algorithms.
Compatible with datacenter servers featuring PCIe Gen3/4 x16 slots including AMD EPYC and Intel Xeon Scalable platforms. Full-height 3/4-length dual-slot requires appropriate riser cards and chassis airflow.
A-U250-P64G-PQ-G uses passive heatsink relying on server chassis airflow (high-CFM racks). Active version (A-U250-A64G-PQ-G) includes fans for standalone/open-air cooling. Both rated 225W with identical performance.
You might also want to check out these products.

5520-VIM-4X - Extreme Networks 4-Ports SFP+ Interface Expansion Module

700469968 - Avaya DBM32 Expansion Module

HPE P12392-B21 2-port 1GbE RJ45 network module

HPE Hewlett Packard Enterprise JD557A 1-port Serial SIC Module

HPE JC107A 48-port 10/100/1000Base-T expansion module

HPE JC756A 48-port 10GbE SFP+ SF module line card

J9534A#ABA - HPE 24x Ports 10/100/1000 RJ-45 PoE+ Expansion Module

MPC2E-3D-NG - Juniper Modular Interface Card for MX2020