Designed for extreme AI performance, F37X is the world's first FPGA AI accelerator card with integrated on-chip HBM2. With 28.1 INT8 TOPS, 460GB/s bandwidth and a power consumption of less than 75W for typical AI applications, it can be easily applied to machine learning inference, video transcoding, image recognition, voice recognition, , genome sequencing analysis, NLP NFV, etc., enabling high-performance, high-bandwidth, low-latency, and low-power consumption AI acceleration.
• Unprecedented Computing Performance
With the Xilinx Virtex UltraScale+ Architecture, F37X offers 2.852M System Logic Cells and 9024 DSP slices. It delivers peak performance of 28.1 INT8 TOPS.
Supports three mainstream programming languages - C/C++, OpenCL and RTL; Enables users to develop and migrate various customized AI algorithms and applications in a highly flexible way.
•World's First FPGA AI Accelerator Card with Integrated On-chip HBM2
The 8GB integrated on-chip HBM2 offers 460GB/s bandwidth that is 20 times faster than a DDR4 DIMM. It is able to handle as many as 8600 images per second, 40 times that of a CPU.
•Unparalleled High Density
In a compact full-height half-length form factor, a single AI server can accommodate more accelerator FPGA cards thus achieving extreme computing and communication performance.
•Ideal Choice for AI Applications
Machine Learning Inference, Video Transcoding, Image Recognition, Voice Recognition, Natural Language Processing, Genome Sequencing Analysis, NFV, Big Data Analysis, Etc.
F37X FPGA Card Technical Specification
28.1 INT8 TOPS
Supports 3 channels 72bit-DDR4 24GB memory in total
PCIe 3.0 x16
High-speed Network Interface
Supports 100GE dual-port QSFP28+
USB debugging interface
12V @75W via PCIE, plus 12V @75W via external Aux
150W (Peak), 75W (typical AI applications)
Dual slot passive cooling
Full-Height Half-Length (167mm x 111mm)
Supports intelligent BMC and enables reading power-supply control and specific information (includes temperature, power consumption, memory, SN, etc.) of the board; Enables reading SN