Xilinx Alveo adaptable accelerator cards
As technology evolves at a tremendous pace, so does the demand for high-performance computing. In the everlasting pursuit for higher processing speed and faster throughput, developers shifted their focus from using a single, ultra-fast computing device such as a CPU, to multi-parallel computing platforms capable of distributing the workload across an array of application-specific processing cores. Indeed, these platforms enabled significant processing acceleration and allowed many new technologies to emerge.
However, fixed-function, instruction-based hardware is built for specific tasks and is not optimized for unstructured algorithms and data processing. With the introduction of emerging technologies, such as Artificial Intelligence (AI) and its subsets, it became apparent that the application-specific hardware is unable to adapt to rapidly evolving software algorithms. Also, with the demise of Moore's law, both power consumption and the chip size became critical factors for a wide range of applications, from cloud to edge. To maintain an optimal level of synchronization between hardware capabilities and software requirements, a significantly different approach is required.
Based on the advanced 16 nm UltraScale+™ field-programmable architecture, Xilinx Alveo accelerator cards deliver unparalleled flexibility and processing power through the deployment of domain-specific architectures, enabling tremendous acceleration for memory-bound, compute-intensive applications. The Xilinx Alveo accelerator card series can be customized for different workloads at a hardware level, thereby ensuring optimal performance and power consumption, while consequently avoiding unnecessary delays associated with the traditional fixed-function hardware implementation. Xilinx Alveo accelerator cards are the most efficient processing accelerators available to date, delivering the highest performance per watt, which makes them a perfect choice for deployment both in the cloud and in on-premises data centers.
The Xilinx Alveo accelerator card series offers an adaptive and scalable solution for a broad range of different workloads through reconfigurable logic, including machine learning inference, quantitative finance, video transcoding, database search and analytics, and more. Starting with the power-efficient, passively cooled Alveo U50 model, packaged in an efficient 75-watt, small form factor, and armed with 100 GbE networking, PCIe Gen4, and HBM2, Xilinx offers solutions for many different use cases. With the capability to quickly adapt to continuously evolving workload algorithms, Xilinx Alveo accelerator cards offer a future-proof concept, reducing the total ownership cost.
Supported by Vitis™, a unified software platform that features a comprehensive core development kit, a rich set of hardware-accelerated open-source libraries, industry-leading domain-specific development environments, and a growing ecosystem of third-party libraries and pre-built applications, Xilinx Alveo Acceleration Cards allow the widest range of developers to build their own custom hardware-accelerated applications, using familiar high-level frameworks.
- Alveo U50
- Alveo U200/U250
- Alveo U280
Xilinx Alveo U50 is an adaptable accelerator card designed for a broad range of different workloads. This accelerator card can be used for optimized workload acceleration of many demanding applications, including financial computing, machine learning, computational storage, data search and analysis, and more. Packaged in an efficient 75-watt, featuring Ultrascale+™ programmable architecture, and armed with features such as PCI Express 4.0, 8GB of HBM2, 100 GbE networking, Alveo U50 accelerator card brings power-efficient and highly scalable adaptive hardware acceleration for both in the cloud and in on-premises data centers.
Due to its low power consumption, Alveo U50 is ideally suited for deployment in any server. Alveo U50 accelerator cards are supported by the Vitis™ Unified Software Platform, allowing a broad range of engineers and developers to unlock the full potential of adaptable hardware acceleration in their applications.
For ordering information and more details on Xilinx Alveo U50 accelerator cards with passive cooling, please visit the links below:
| XILINX ALVEO U50 FEATURES | |||
|---|---|---|---|
| Dimensions | Form factor |
|
|
| Width | Single Slot | ||
| Logic resources | Look-up Tables | 872,000 | |
| Registers | 1,743,000 | ||
| DSP Slices | 5,952 | ||
| Compute Performance | INT8 TOPs | 16.2 | |
| DRAM Memory | HBM2 Memory | 8 GB | |
| HBM2 Bandwidth |
|
||
| Internal SRAM | Total Capacity | 28 MB | |
| Total Bandwidth | 24 TB/s | ||
| Interfaces | PCI Express® |
|
|
| Network Interface | 1 x QSFP28 (100 GbE) | ||
| Timestamp | Clock Precision | IEEE 1588 | |
| Power and Thermal | Thermal cooling | Passive | |
| Typical power | 50 W | ||
| Maximum Power | 75 W | ||
| Links | Machine Learning Solution Brief | ||
| Alveo U50 Product Brief | |||
Xilinx Alveo U200/U250 are adaptable accelerator cards designed for more demanding workloads. These cards can be used for optimized workload acceleration of many different applications that require high data throughput and low processing latency, including machine learning inference, video transcoding, database search and analytics, genomics, and similar highly demanding workloads. Based on the programmable 16nm Ultrascale+™, Alveo U200/Alveo U250 accelerator cards bring adaptable and customizable hardware acceleration with up to 90x performance increase over the fastest fixed-function solutions available up to date.
Alveo U200/U250 cards feature a set of high-speed options, enabling full-scale acceleration across the entire application, reducing any performance bottlenecks. These accelerator cards are available with both active and passive cooling, depending on specific deployment needs or constraints. Alveo U200/U250 accelerator cards are supported by the Vitis™ Unified Software Platform, allowing a broad range of engineers and developers to unlock the full potential of adaptable hardware acceleration in their applications.
For ordering information and more details on Xilinx Alveo U200/U250 accelerator cards with active and passive cooling, please visit the links below:
- Xilinx Alveo U200A (active cooling)
- Xilinx Alveo U200P (passive cooling)
- Xilinx Alveo U250A (active cooling)
- Xilinx Alveo U250P (passive cooling)
| XILINX ALVEO FEATURES | ALVEO U200 | ALVEO U250 | |
|---|---|---|---|
| Dimensions | Form factor (passive) Form factor (active) |
|
|
| Width | Dual Slot | Dual Slot | |
| Logic resources | Look-up Tables | 1,182,000 | 1,728,000 |
| Registers | 2,364,000 | 3,456,000 | |
| DSP Slices | 6,840 | 12,288 | |
| Compute Performance | INT8 TOPs | 18.6 | 33.3 |
| DRAM Memory | DDR Format | 4 x 16 GB 72b DIMM DDR4 | 4 x 16 GB 72b DIMM DDR4 |
| DDR Total Capacity | 64 GB |
|
|
| DDR Total Bandwidth | 77 GB/s | 77 GB/s | |
| Internal SRAM | Total Capacity | 43 MB | 57 MB |
| Total Bandwidth | 37 TB/s | 47 TB/s | |
| Interfaces | PCI Express® | Gen3 x16 |
|
| Network Interface | 2 x QSFP28 (100 GbE) | 2 x QSFP28 (100 GbE) | |
| Power and Thermal | Thermal cooling |
|
|
| Typical power | 100 W | 110 W | |
| Maximum Power | 225 W | 225 W | |
| Links | Machine Learning Solution Brief | ||
| Alveo U200/U250 Product Brief | |||
Xilinx Alveo U280 acceleration card is designed to meet the continuously changing needs of a modern data center. Based on the proven Ultrascale+™ programmable architecture, and armed with 8 GB of HBM2 up to 460 GB/s bandwidth, it provides high-performance, adaptable acceleration for memory-bound, compute-intensive applications such as database, analytics, and real-time machine learning inference.
Besides 8 GB of HBM2 memory, the Alveo U280 accelerator card also includes two DDR 4 slots, each supporting up to 16 GB, reaching total bandwidth of 38 GB/s. Also, it has an internal Synchronous RAM capacity of 41 MB, reaching up to 30 TB/s.
Alveo U280 is Xilinx's top tier accelerator card. It offers many additional features, including PCI Express 4.0 interface with CCIX support, allowing it to leverage the latest server interconnect infrastructure for high-bandwidth, low latency, cache-coherent shared memory access with CCIX host processors. 100 GbE is also available, enabling utilization of the hardware acceleration for network performance optimization on a large scale. Xilinx Alveo U280 can be volume-deployed in data centers and used for application development and prototyping. Alveo U280 accelerator cards are supported by the Vitis™ Unified Software Platform, allowing a broad range of engineers and developers to unlock the full potential of adaptable hardware acceleration in their applications.
For ordering information and more details on Xilinx Alveo U280 accelerator cards, please visit the links below:
| XILINX ALVEO U280 FEATURES | ||
|---|---|---|
| Compute | INT8 TOPs (peak) | 24.5 |
| Dimensions | Width | Dual Slot |
| Form factor | Full Height, Full Length | |
| DRAM Memory | HBM2 Total Capacity | 8 GB |
| HBM2 Total Bandwidth | 460 GB/s | |
| DDR Format | 2x 16 GB 72b DIMM DDR4 | |
| DDR Memory Capacity | 32 GB | |
| DDR Total Bandwidth | 38 GB/s | |
| SRAM Memory |
Internal SRAM Capacity | 41 MB |
| Internal SRAM Total Bandwidth | 30 TB/s | |
| Interfaces | PCI Express® | Gen4x8 with CCIX |
| Network Interfaces | 2x QSFP28 (100GbE) | |
| Logic resources | Look-up Tables (LUTs) | 1,079,000 |
| Power | Maximum Total Power | 225 W |
| Links | Alveo U280 Product Brief | |
Do you have a Question?
Contact EBV
If you need any assistance, please click below to find your closest EBV sales office.
