FPGA Acceleration of Convolutional Neural Networks (CNNs)
White Paper FPGA Acceleration of Convolutional Neural Networks Overview Convolutional Neural Networks (CNNs) have been shown to be extremely effective at complex image recognition problems.
The Alveo U25N, U200, U250, and U280 FPGA accelerator cards from AMD (formerly Xilinx) have reached or will reach end of life in 2024. These devices use Virtex UltraScale+ FPGAs exclusive to these designs. However, BittWare has similar UltraScale+ FPGAs on cards that are currently in production.
We would recommend these BittWare cards as alternatives if you were using the listed Alveo FPGA boards for past programs or continue to need solutions with similar devices. Get in touch with us to discuss your specific needs.
These cards use FPGAs exclusive to the cards, two QSFP28 ports and 64GB DDR4 memory. Host I/O is provided by a PCIe Gen3 x16 interface.
BittWare recommends as equivalents the XUP-P3R or XUP-VV8. Both of these cards feature similar large AMD Virtex UltraScale+ FPGAs, with the XUP-VV8 offering the larger VU13P in addition to VU9P. Prefer more than 64GB? We offer up to 128GB DDR4 or QDR-II+.
I/O options match the PCIe Gen3 x16 but you get additional interfaces: SEP connector on the XUP-P3R and two OCuLink on the XUP-VV8. You get 2x-4x more 100G network links as well.
These cards also feature timing headers, BittWare’s BMC, and the BittWorks Toolkit II.
Check the product pages (XUP-P3R or XUP-VV8) for in-stock status or connect with us to get a quote!
Card | FPGA | LUTs | Memory | I/O |
---|---|---|---|---|
Alveo U200/U250 | XCU200 or XCU250 | 1,182K/1,728K | 64GB DDR4 | 2x QSFP28, PCIe Gen3 x16 |
BittWare XUP-P3R | VU9P Speed grade -2 | 1,182K | 4x DIMM sites, up to 128GB DDR4 or QDR-II + options | 4x QSFP28 (100G each), expansion port with 20x GTY transceivers, PCIe Gen3 x16 |
BittWare XUP-VV8 | VU9P or VU13P, Speed grade -2 or -3 | 1,182K (VU9P)/1,728K (VU13P) | 4x DIMM sites, up to 128GB DDR4 or QDR-II+ options | 4x QSFP-DD (8x25G each), 2x OCuLink (4x 25G + GPIO each), PCIe Gen3 x16 |
These cards use FPGAs featuring 8GB of HBM2. Host I/O is provided by a PCIe Gen4 x8 interface.
BittWare recommends as equivalent the AV-860h. This card features 32GB HBM2e, high-speed I/O via ARC6-16 connectors, and PCIe Gen5 x8.
The AV-860h gives you more I/O, HBM2e, and PCIe Gen5 instead of Gen4.
Check the product page for in-stock status or connect with us to get a quote!
Card | FPGA | LUTs | Memory | I/O |
---|---|---|---|---|
Alveo U280 | UltraScale+ XCU280 | 1,304K | 8GB HBM2 16 GB DDR4 RDIMMs | 2x QSFP28 PCIe Gen3 x16 or Gen4 x8 |
BittWare AV-860h | Versal XCVH1582 | 3,837K | 32GB HBM2e up to 64GB LPDDR4 | 8x ARC6-16, PCIe 4.0 x16 or 2x PCIe 5.0 x8 |
These cards use FPGAs exclusive to the cards, two SFP28 ports and 6GB DDR4 memory. Host I/O is provided by a dual PCIe Gen3 x8 interface.
BittWare recommends as equivalent the 250-SoC. This card features a similar large AMD Zynq UltraScale+ FPGAs, with 8GB DDR4. I/O options match the dual PCIe Gen3 x8 but you get QSFP28 ports instead of SFP28s.
These cards also feature BittWare’s built-in self test (BIST).
Check the product page for stock status or connect with us to get a quote!
Card | FPGA | LUTs | Memory | I/O |
---|---|---|---|---|
Alveo U25N | Zynq® UltraScale+™ XCU25 | 522k | 1x 2GB x 40 DDR4-2400 1x 4GB x 72 DDR4-2400 | 2x SFP28 2x PCIe Gen3 x8 |
BittWare 250-SoC | Zynq® UltraScale+™ ZU19EG | 1,143K | 2x 4GB x72 DDR4 | 2x QSFP28 PCIe Gen3 x16 |
Our technical sales team is ready to provide availability and configuration information, or answer your technical questions.
"*" indicates required fields
White Paper FPGA Acceleration of Convolutional Neural Networks Overview Convolutional Neural Networks (CNNs) have been shown to be extremely effective at complex image recognition problems.
PCIe FPGA Card XUP-VV4 UltraScale+ FPGA PCIe Board with VU13P 4x 100GbE and up to 512GB DDR4 Need a Price Quote? Jump to Pricing Form
BittWare Webinar Arkville PCIe Gen4 Data Mover Using Intel® Agilex™ FPGAs Webinar The Arkville IP from Atomic Rules was recently updated to support Intel Agilex
Panel Discussion How Today’s FPGAs are Taming the Data Deluge Problem From Gen5 to AI, NOCs to RF at the Edge Watch the recording for