The HPC-A10-K1GPU is the most versatile and highest-performing FPGA accelerator card on the market, from video analytics to model and simulation, from radar processing to computational finance, from cyber security to broadcast video processing, to image processing with OpenCV. With the ability to now run standard OpenCL on the Altera Arria 10, the HPC-A10-K1GPU card can be easily programmed without the need to dive down into complicated Verilog or VHDL code.



With dual 40 GbE ports, this card is agile enough to handle the most demanding networking applications. Unlike a traditional Host Bus Adapter (HBA), the HPC-A10-K1GPU can smartly pre/post data in the highly capable Altera Arria 10 and then feed external Tesla K40/K80 GPGPUs or Quadro GPU cards via GPUDirect.

The Dual Tegra K1 GPU SoCs offer the integrator many new options. NVIDIA CUDA code can run on this card and interact with Altera FPGAs running OpenCL, Verilog, or VHDL. The designer can preserve previous software and firmware development and move to this flexible platform.

For image processing, this card gives the developer many options. The Altera FPGA has extensive image processing IP available. USB3, DisplayPort, 1 GigE, 10 GbE and 40 GbE video streams can be handled by the FPGA, while the NVIDIA GPU SoC handles USB and GigE streaming video. Whether you need to process millions to billions of minutes of video, upscale, downscale, stitch, H.264 encode/decode, H.265 encode/decode, or work with 4K video, this card offers you the flexibility to hit your objective.

For Radar/Sonar processing, this card offers a scalable backend processor that meets the objective of Modular Open Systems Architecture (MOSA). A highly capable two-channel Digital Receiver/Exciter (DREX) can be implemented using 4DSP FMCs and IP available from Altera. CEI provides extensive systems engineering consulting and is standing by to discuss your needs.


  • Altera Arria 10 with 144 Mb QDR-II+ with 2/4/6/8 GB HMC for 2018 GLFOPS
  • Native OpenCL & CUDA Support
  • NVIDIA GPUDirect Support
  • Dual NVIDIA Tegra K1 GPU
  • Dual 10/40 GbE Ports
  • High Pin Count VITA 57 FMC
  • Image Processing, Radar, Sonar, Analytics, Cyber, Broadcast, Finance


Arria 10 FPGA

  • GX1150—10AX115 (Std)
  • GX900—10AX090 (Opt)
  • Native OpenCL Support
  • GPUDirect Support
  • 288 Mb QDR-II+
  • 1150K LEs
  • 427,200 ALMs
  • 1,708,800 Registers
  • 96 Transceivers
  • 4 IP Gen3 PCIe Blocks
  • 2,713 M20k Memory Blocks
  • 53 M20K Memory
  • 1,518 Variable DSP Blocks
  • 3,306 GMACS
  • 1,366 GFLOPs


  • 326 GLOPs
  • CUDA 6.5+
  • OpenCL
  • 192 CUDA Cores
  • 4 ARM Cores
  • 16 GB eMMC Flash
  • 2 GB DDR3 RAM
  • Standard Linux Support

Hybrid Memory Cube RAM

  • Two HMCs with either 2GB or 4GB Capacities
  • 2/4/6/8 GB
  • Up to 17x Faster than DDR3
  • Up to 70% Less Power than DDR3
  • Up to 80 GB/s Raw Memory Bandwidth per Hybrid Memory Cube

FMC Site (VITA 57)

  • High Pin Count (HPC)
  • Rear I/O
  • Pre-integrated with 4DSP FMCs: ADC,  DAC and Mixed A/D & D/A Options

MCU – Health Monitor

  • Freescale Kinetis K61 (K61P144M150SF3)
  • Up to 150 MHz ARM Cortex-M4 Core
  • System Configuration
  • Power Sequencing
  • Power Monitoring
  • User-defined Functions