Programming the V80 FPGA

This section assumes that the Vollo SDK is already installed and set up on the machine that you are using. If you haven't done so already, instructions for how to do that may be found at: Vollo SDK Installation

Make sure the VOLLO_SDK environment variable is set by sourcing setup.sh from the Vollo SDK.

source <path-to-VOLLO_SDK>/setup.sh

Download the bitstream for your FPGA

The bitstream is available on the Github Release page alongside the Vollo SDK. For example to download the bitstream for the AMD V80 board with the c6b32 configuration of Vollo:

curl -LO https://github.com/MyrtleSoftware/vollo-sdk/releases/download/v30.1.0/vollo-amd-v80-c6b32-30.1.tar.gz
mkdir -p $VOLLO_SDK/bitstream
tar -xzf vollo-amd-v80-c6b32-30.1.tar.gz -C $VOLLO_SDK/bitstream

Alternatively, for the AMD V80LL, use:

curl -LO https://github.com/MyrtleSoftware/vollo-sdk/releases/download/v30.1.0/vollo-amd-v80ll-c6b32-30.1.tar.gz
mkdir -p $VOLLO_SDK/bitstream
tar -xzf vollo-amd-v80ll-c6b32-30.1.tar.gz -C $VOLLO_SDK/bitstream

Programming the FPGA via JTAG

Programming a V80 board over JTAG is necessary if the board does not yet have a Vollo image loaded on it or if the device does not enumerate correctly. Programming over PCIe is preferred. If the board does not enumerate or there is some other issue with PCIe programming then JTAG programming is the only option.

This requires a USB cable to be connected to the accelerator card and Vivado to be installed on the system so that the device can be programmed over JTAG.

  1. Download and install Vivado Lab Edition:

    • Navigate to the Vivado Design Tools download page.

    • Under "Vivado Lab Solutions" find "Vivado 2026.1: Lab Edition - Linux (TAR/GZIP - 1.99 GB)" (later versions may be available).

    • Download the file and extract it to a directory of your choice. You will need an AMD account to download the file. You can create an account for free.

    • Pick a location to install Vivado_Lab, e.g. /opt/Xilinx, a user directory like ~/Xilinx is also fine:

      VIVADO_DIR=~/Xilinx
      mkdir -p $VIVADO_DIR
      
    • Extract the tarball:

      tar xf Vivado_Lab_Lin_2025.2_1114_2157.tar
      cd Vivado_Lab_Lin_2025.2_1114_2157
      
    • Run the installer:

      ./xsetup --agree 3rdPartyEULA,XilinxEULA --batch Install --edition "Vivado Lab Edition (Standalone)" --location $VIVADO_DIR
      
    • Check that installation was successful:

      $ $VIVADO_DIR/2026.1/Vivado_Lab//bin/vivado_lab -version
      Vivado Lab Edition v2026.1 (64-bit)
      SW Build 6299465 on Fri Nov 14 21:19:43 MST 2025
      Tool Version Limit: 2025.11
      Copyright 1986-2022 Xilinx, Inc. All Rights Reserved.
      Copyright 2022-2025 Advanced Micro Devices, Inc. All Rights Reserved.
      
  2. Run the flash_vollo-amd-v80-c6b32.tcl script to program the V80 board:

    sudo $VIVADO_DIR/2026.1/Vivado_Lab/bin/vivado_lab -mode batch -source ./flash_vollo-amd-v80-c6b32.tcl
    

    This prints out a lot of lines while programming and takes about 10 minutes.

    If you get an error like this:

    ERROR: [Labtoolstcl 44-469] There is no current hw_target.
    

    Make sure that you ran vivado_lab with sudo and that the USB cable is plugged in.

    After programming you must power cycle the host for the new bitstream to be loaded.

    Sometimes a V80 host machine will hang on boot. You may need to force another power cycle of the host to bring it back. Occasionally a power cycle isn't enough and you may need to turn the power off for several minutes before turning it back on.
  3. If successful the device should now enumerate as a Myrtle.ai Vollo device:

    $ lspci -d 1ed9:
    01:00.0 Processing accelerators: Myrtle.ai Device 000a
    01:00.1 Processing accelerators: Myrtle.ai Device 100a
    

Programming the FPGA over PCIe

If your FPGA is already programmed with the Vollo accelerator then you can update the bitstream over PCIe. You can check if the device is programmed with the Myrtle.ai Vollo bitstream by running:

$ lspci -d 1ed9:
01:00.0 Processing accelerators: Myrtle.ai Device 000a
01:00.1 Processing accelerators: Myrtle.ai Device 100a

If the device has not been programmed with the Vollo bitstream then you will need to program the board over JTAG. See Programming the FPGA via JTAG.

Programming over PCIe is the preferred method of programming the board as it is faster than programming over JTAG, and does not require a USB programming cable or for Vivado to be installed.

  1. Bind the card to the Vollo driver. This takes all of the card's functions, including the management function (100a) that programs the flash, so no other driver is needed.

    sudo $VOLLO_SDK/load-kernel-driver.sh vfio
    

    Earlier releases used AMD's ami kernel driver for this step. If it is still loaded from one of them, unload it first with sudo rmmod ami.

  2. Program the flash with vollo-tool. If you only have one board, device_index is 0.

    sudo $VOLLO_SDK/bin/vollo-tool fpga-config overwrite-partition ${device_index:?} $VOLLO_SDK/bitstream/vollo-amd-v80-c6b32.pdi USER_IMAGE
    

    Use the vollo-amd-v80ll-c6b32.pdi file for a V80LL. Before erasing or writing, vollo-tool compares the file's target FPGA with the selected card's manufacturing EEPROM, read over PCIe. This requires no JTAG connection and works even when an older V80LL image reports the V80 subsystem ID. A mismatch, unknown part number, or unreadable EEPROM stops the update. The --ignore-device-mismatch option bypasses this identity check for recovery; an image for the other FPGA cannot boot.

    There will be a progress bar and it should take around 5 minutes to program the flash. You will need to power cycle the host for the new bitstream to be loaded.

    Sometimes a V80 host machine will hang on boot. You may need to force another power cycle of the host to bring it back. Occasionally a power cycle isn't enough and you may need to turn the power off for several minutes before turning it back on.
  3. If successful the device should now enumerate as a Myrtle.ai Vollo device:

    $ lspci -d 1ed9:
    01:00.0 Processing accelerators: Myrtle.ai Device 000a
    01:00.1 Processing accelerators: Myrtle.ai Device 100a
    

Troubleshooting

Some troubleshooting recommendations may be found in Troubleshooting the V80