You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Support persistent FPGA configuration (bitstream reuse) across OpenCL runs on AMD Alveo V80 #443
I’m currently working with the AMD Alveo V80 board and wanted to ask for a feature regarding FPGA configuration overhead.
Feature Description
Currently, when executing an OpenCL application targeting Vortex on AMD Alveo V80, the corresponding partial bitstream reconfigures the FPGA each time. I would like to request a feature that allows the user to program the Alveo V80 once and then run multiple OpenCL applications, without requiring the FPGA to be reconfigured every time.
changed the title [-]Support Πersistent FPGA configuration (bitstream reuse) across OpenCL runs on AMD Alveo V80[/-][+]Support Persistent FPGA configuration (bitstream reuse) across OpenCL runs on AMD Alveo V80[/+]on Nov 4, 2026
changed the title [-]Support Persistent FPGA configuration (bitstream reuse) across OpenCL runs on AMD Alveo V80[/-][+]Support persistent FPGA configuration (bitstream reuse) across OpenCL runs on AMD Alveo V80[/+]on Nov 4, 2026
Fixed by #490 together with SLASH vortex_3.x (720754bb): vrtd now skips writing a design the card already holds, as XRT does with an xclbin's UUID. Tested on the V80: after one design write, OpenCL vecadd, OpenCL sgemm and native sgemm ran repeatedly without reconfiguring the card, all passing. VRT_FORCE_PROGRAM=1 forces a write, and v80-smi forget-design resets vrtd's record after a JTAG load.
Activity
[-]Support Πersistent FPGA configuration (bitstream reuse) across OpenCL runs on AMD Alveo V80[/-][+]Support Persistent FPGA configuration (bitstream reuse) across OpenCL runs on AMD Alveo V80[/+][-]Support Persistent FPGA configuration (bitstream reuse) across OpenCL runs on AMD Alveo V80[/-][+]Support persistent FPGA configuration (bitstream reuse) across OpenCL runs on AMD Alveo V80[/+]slash: V80 userspace 2.0.0 with design UUIDs, as an optional manual i…
tinebp commentedon Nov 19, 2026
Fixed by #490 together with SLASH vortex_3.x (720754bb): vrtd now skips writing a design the card already holds, as XRT does with an xclbin's UUID. Tested on the V80: after one design write, OpenCL vecadd, OpenCL sgemm and native sgemm ran repeatedly without reconfiguring the card, all passing. VRT_FORCE_PROGRAM=1 forces a write, and v80-smi forget-design resets vrtd's record after a JTAG load.