v2026.07.00

This marks DAQIRI's first release. DAQIRI sends and receives Ethernet packets into CPU and GPU memory through a C++/Python API, with GPUDirect zero-copy paths between the NIC and GPU.

Stream types and engines

Raw Ethernet (raw): DPDK engine by default, or the optional ibverbs engine.
Socket (socket): UDP and TCP over Linux sockets, or RoCE over ibverbs.

A PCIe path for sensors on the PCIe bus (FPGAs, frame grabbers, and similar devices) is planned but not implemented, so this release covers raw Ethernet and sockets only.

Feature Highlights

GPUDirect zero-copy: receive straight into GPU memory, either whole packets to GPU or with header-data split (headers to CPU, payload to GPU).
Hardware flow steering: route packets on the NIC by UDP port or payload fields, set in YAML or added and removed at runtime.
Packet reorder: aggregate and reorder RX packets on the GPU or CPU into a single output buffer.
Container or source deployment: run from a container with all dependencies included, or build from source with CMake.

Initial PRs

#7 - Add Socket API by @cliffburdick in #20
#28 - Handle N/A memory clocks in check_nvidia_gpu_clocks() by @RamyaGuru in #30
#29 - Fix MPS check and topology ANSI parsing in tune_system.py by @RamyaGuru in #31
#5 - Merge tutorial site into main by @RamyaGuru in #36
#5 - Replace placeholder bench output and stale names in benchmarking tutorial by @dleshchev in #38
#37 - Fix stale doc references and rename tutorials/configuration.md by @RamyaGuru in #39
#37 - Add CI gate for documentation drift by @RamyaGuru in #40
#37 - Add documentation sync rule and drift guidance by @RamyaGuru in #42
#43 - Improve errors on peermem/dmabuf by @cliffburdick in #41
#46 update readme backends defaults tutorials by @RamyaGuru in #50
#45 - Add DGX Spark (GB10) support to docs and examples by @dleshchev in #49
#56 Fixed failed init leaves hugepage files behind in /dev/hugepages, blocking subsequent runs by @chloecrozier in #58
#44 - Auto-bump num_bufs below 1.5x DPDK ring size with a warning by @dleshchev in #48
#52 - base torch image option on build by @chloecrozier in #59
#53 - Clean up legacy Holoscan/Holohub references in docs by @RamyaGuru in #64
#65 - Add use-case decision tree to configuration tutorial for picking an example YAML by @RamyaGuru in #66
#56, #44 - Restore DPDK bench paths broken on main by @dleshchev in #68
#70 - Pin cmake-data alongside cmake in Dockerfile base stage by @dleshchev in #71
#77 - Fix nmcli mtu property and drop -lmc from Spark system-config tutorial by @RamyaGuru in #78
#45 - Style system-config tutorial: enlarge platform tabs, scope TOC … by @RamyaGuru in #80
#81 - Address Greptile CMake packaging feedback by @cliffburdick in #82
#57 - Add Greptile PR review config by @chloecrozier in #74
#65 - Docs/configuration walkthrough revamp by @RamyaGuru in #84
#86 - Unify docs site visual style: reskin Material to match custom HTML pages by @RamyaGuru in #87
#89 - Make Manager::shutdown() idempotent to prevent post-finalize log crash by @RamyaGuru in #90
#91 - Point README doc links at the live docs site by @dleshchev in #92
#69 - Introduce Concepts page and restructure API Reference docs by @chloecrozier in #94
#15 - Fix RDMA TX metadata ownership by @cliffburdick in #105
#99 - Add multi-queue raw benchmark configs by @cliffburdick in #100
#101 - Avoid double-returning TX metadata on enqueue failure by @fallintoplace in #102
#15 #77 - Fix DGX Spark example to a true over-the-wire loopback (TX/… by @RamyaGuru in #108
#106 - Fix CUDA 12.6 raw benchmark build by @dleshchev in #107
#69 - Add Python bindings API reference by @dleshchev in #104
#112 - Add OpenTelemetry Metrics by @cliffburdick in #111
#69 - Apply docs review feedback by @chloecrozier in #109
#67 - Add bare-metal CMake build tutorial by @chloecrozier in #95
#15 - Add reusable C++ bench infrastructure for per-platform performance reports by @RamyaGuru in #96
#15 - Add DGX Spark sweep tooling and RDMA loopback prereqs by @dleshchev in #116
#15 - Reapply RDMA benchmark profiling updates by @dleshchev in #118
#113 - Fix RDMA cross-host SEND timeout and shutdown deadlock by @chloecrozier in #115
#57 - Tune Greptile review output by @chloecrozier in #122
#119 - Add uninstall script for the CMake install and container image by @chloecrozier in #120
#12 - Fix DPDK flow isolation ordering by @cliffburdick in #125
#127 - Fix docs nav: consistent hide:navigation and data-driven tab dropdowns by @RamyaGuru in #130
#54 - Add RTX PRO 6000 Blackwell Server Edition support to tune_system.py by @chloecrozier in #121
#135 - Harden DPDK split burst handling by @cliffburdick in #137
#141 - Port queue-specific DPDK RX flush by @cliffburdick in #143
#126 Add RDMA bench flow-control windows by @cliffburdick in #144
#93 Add light theme support to docs site by @chloecrozier in #146
#145 - Auto-build landing page from MkDocs snippets by @dleshchev in #154
#158 - Style updates to website images and removal of news by @awthomp in #158
#15 - Add DGX Spark v1 performance report by @RamyaGuru in #98
#55 - Add DAQIRI Holoscan integration tutorial by @dleshchev in #117
#151 - Rename manager/backend abstraction to engine (DAQIRI_MGR → DAQIRI_ENGINE) by @cliffburdick in #156
#170 - Guard socket engine RdmaEngine member behind DAQIRI_ENGINE_RDMA by @dleshchev in #171
#131 - Fix DPDK TX enqueue mbuf leak by @fallintoplace in #132
#133 - Validate MAC strings before formatting by @fallintoplace in #134
#152: Add pure-ibverbs MPRQ raw-Ethernet backend (engine: ibverbs) by @cliffburdick in #148
#129 - reject mixed standard and flex-item RX flows by @chloecrozier in #160
#128 - Harden DPDK RX flow creation error handling by @chloecrozier in #161
#176 Packet pacing by @cliffburdick in #179
#177: Drop DPDK as a dependency unless the dpdk engine is built by @cliffburdick in #178
#173 #174 #175 - fix Spark cross-host RDMA bench setup and build by @chloecrozier in #180
#165 #166 #167 - Complete DPDK flow rule lifecycle by @chloecrozier in #181
#183 - Reformat DAQIRI website examples and supply human readable descriptions of examples by @awthomp in #183
#186 - Improve website mobile experience by @awthomp in #186
#187 - Polish landing page links and footer CTA by @dleshchev in #188
#189 - Revamp DAQIRI Readme by @awthomp in #189
#190 - Remove blank row on README table by @awthomp in #190
#191 - Convert table to HTML to remove blank row by @awthomp in #191
#195 - Website - Eliminate column in example table and fix code expansion block by @awthomp in #195
#127 - Remove LHS navigation from Holoscan integration tutorial by @RamyaGuru in #201
#205 - dpdk: preserve kernel fallback for dynamic rx flows by @dleshchev in #204
#15 - Add reusable FFT/GEMM GPU workloads to receive-path benchmarks by @RamyaGuru in #203
#197 - Fix DPDK MTU sizing for asymmetric RX/TX queues by @MikesRND in #198
#192 - Add eCPRI matching by @cliffburdick in https://githu...