AMD Alveo V80
High memory bandwidth is just as important to optimal performance for large-scale data processing as sheer compute capability. The new AMD Alveo V80 compute accelerator optimises memory-bound applications with large data sets and FPGA hardware adaptability. The Alveo V80 accelerator card, which is currently being produced in large quantities, provides up to double the bandwidth and compute density of earlier generation cards. It also comes with an easier-to-use development process for FPGA designers that use the AMD Vivado Design Suite.
AMD Compute Accelerator Card, Alveo V80
Based on the 7nm Versal adaptable SoC architecture, the AMD Alveo V80 accelerator card is an HBM-enabled compute accelerator card intended for memory-intensive workloads such as data analytics, HPC, network security, sensor processing, computational storage, and fintech.
The new card has a full-height, ¾ length (FH¾L) form factor and is powered by an AMD Versal HBM adaptive SoC. To help overcome performance limitations, it has 2.6M LUTs of FPGA fabric, 10,848 DSP slices of compute, and 820 GB/ of memory bandwidth.
Featuring up to twice the logic density, twice the memory bandwidth, and four times the network bandwidth of its predecessor, the AMD Alveo U55C compute accelerator, the Alveo V80 maximises the number of cards, servers, and rack space while enabling robust compute clusters.
Dedicated, Network-Attached Accelerator for Large-Scale Data Sets and Memory-Heavy Tasks
The Alveo V80 card’s hardware adaptability enables wide use with a variety of unique applications. Since the card is a 4x200G network-attached accelerator, it can handle large amounts of incoming data in real-time, avoiding the PCIe communication issues that GPUs have.
The Alveo V80 accelerator is perfect for a variety of high-performance computing (HPC) applications, such as molecular dynamics, sensor processing, and genomic sequencing, since it can grow to hundreds of nodes across Ethernet for compute clusters. The FPGA hardware flexibility and integrated 400G cryptographic engines and 600G Ethernet hard blocks of the Alveo V80 accelerator make it suitable for AI-enabled anomaly detection and line-rate packet inspection in the context of network security.
Because it can combine query acceleration and compression on the same card, the accelerator is also perfect for computational storage and data analytics. This feature increases effective storage capacity while speeding up the time to insights. It is also a good fit for a number of FinTech applications, such as financial modelling and simulation, options pricing, and strategy backtesting.
Case Study: An Advancement in Astrophysics Computation
Australia’s national research organisation, the Commonwealth Scientific and Industrial Research Organisation (CSIRO), is building the largest radio astronomy antenna array in the world. It presently consists of 420 Alveo U55C accelerator cards, which are used to process radio waves in order to study the early universe and investigate galaxy evolutions.
With the Alveo V80 accelerator, CSIRO hopes to handle additional signal processing duties from the telescope’s 131,000 antennas while cutting footprint, cost, and the number of cards required by up to 66%. The increase in compute capacity per card can result in a TCO reduction of up to 20%, including the possible savings on cards, servers, rack space, and power.
“AMD initially embraced the Alveo product line because of its capacity to handle enormous volumes of sensor data instantly,” stated CSIRO Research Engineer Grant Hampson, who works in the Space and Astronomy Division.AMD next-generation beamformer and correlator need lower TCO. The Alveo V80 accelerator offers a small, power-efficient solution in an affordable footprint, representing a technological step-function advancement over the previous generation Alveo U55C cards.
Development Made Easy for FPGA Designers
With the Alveo Versal Example Design (AVED), which is already accessible on GitHub, traditional hardware engineers can fully utilise the Alveo V80 accelerator card. AVED is built on the well-known Vivado tool flow and streamlines hardware bring-up utilising conventional FPGA and RTL flows. Using a pre-built subsystem that is specifically targeted for the Alveo V80 accelerator card and implemented on the AMD Versal adaptive SoC, the sample design offers a productive starting point.
The Alveo V80 compute accelerator offers a quick route to production and streamlines system integration at the system level. Design teams can avoid PCB integration, inventory management, and product lifecycle management responsibilities by utilising a pre-validated deployment card.
AMD Alveo V80 availability
Alveo V80 is currently available from AMD and approved distributors and is being produced in large quantities.
- Predicted on specifications as of April 2024 that are available to the general public in the AMD Alveo Product Selection Guide. (ALV-13).
- Comparing an estimated implementation of 140 AMD Alveo V80 accelerator cards with a current implementation of 420 Alveo U55C accelerator cards, based on independent “Early Access” performance and cost analysis estimations by CSIRO in October 2023. The anticipated costs of power and cooling OPEX were factored into an estimated three-year Total Cost of Ownership. AMD has not independently verified any of the performance or cost-savings claims, which are all estimations from CSIRO. Numerous presumptions and variables affect performance and cost benefits, which might vary depending on system setup and other circumstances. The outcomes may not be normal and are unique to CSIRO.
