YakoGLWhat runs today
Three GPU families running on Zephyr. Here are the recorded results, the conditions behind them, and the work still in progress.
OpenGL ES 2.0: zero test failures on all three boards. These are our own runs of the Khronos dEQP-GLES2 test suite on real hardware. YakoGL has not been through the Khronos Adopters Program and is not a certified conformant implementation. Not-supported cases and warnings remain visible below.
Records checked at .
Each result names its measurement date and build. The latest runs use different builds. A pre-publication build belongs to the earlier development repository; its hash is an identifier, and the public record is linked directly. A missing build commit is marked unverified.
Toradex Verdin AM62P · PowerVR BXS-4-64
| Measurement | Result | Build, date and public record |
|---|
Toradex Verdin iMX8M Plus · Vivante GC7000UL
| Measurement | Result | Build, date and public record |
|---|
Raspberry Pi 4 · Broadcom V3D 4.2
| Measurement | Result | Build, date and public record |
|---|
Reading the results
- dEQP-GLES2: the ES 2.0 mustpass list contains 17,165 cases. The latest AM62P and i.MX8MP runs also include 206 cases outside that list. Not supported includes optional features and multisample cases a 1x surface cannot run. The AM62P's separate 4x result comes from an older build; it is not added to the new 1x row.
- gl_test: YakoGL's regression cases read their results back from the GPU. The set differs by board. A passing run does not erase intermittent failures noted in the record.
- glmark2-es2: 800×600, on-screen, all 33 scenes. Scores are benchmark results, not a promised application frame rate. The public Linux reference rows use different configurations; they are context rather than a controlled driver comparison.
- Boot: the recorded tests load Zephyr from the board's own storage. Permanent-default boot modes and measured power-on latency are separate questions; see the boot record.
In development
OpenGL ES 3.0
Product builds report ES 2.0 by default. The AM62P full-tree baseline is shown above; later targeted reruns improve individual groups, but do not constitute a new complete result. The Pi 4 has partial hardware results, with failures still present. i.MX8MP host capability work is recorded separately from hardware validation. Read the ES 3.0 record.
Mali-G610 / NanoPC-T6 WIP
The fourth GPU back end is actively being brought up on the FriendlyElec NanoPC-T6
(Rockchip RK3588, Arm Mali-G610). The gpu-mali back end already runs
gl_test on the board's Linux through the panthor kernel driver; the
2026-10-24 sweep records 98 of 256 cases PASS and 0 FAIL, with the remaining
cases refused, not present, or skipped. Zephyr board support, board glue and
GPU bring-up probes are already in the tree, while the complete Zephyr graphics
path and board-level validation are still in progress. This is therefore a WIP
target, not yet one of the three supported Zephyr boards.
Mali bring-up details ·
support criteria.
What to assess before integration
- Your UI's performance. Heavy benchmark scenes remain slow, and the i.MX8MP's panel synchronization limits many lighter scenes. Measure your own application and display configuration.
- Visual validation. Every benchmark scene has picture values recorded. Pi 4 and i.MX8MP results still need visual review; an i.MX8MP texture/mipmap picture change is unexplained. The records distinguish camera inspection from numerical comparisons.
- Input and reliability. The demos currently run without touch input. The i.MX8MP regression record includes intermittent failures, even though its latest listed run passes the cases executed.
The roadmap sets the project direction. The board record contains the measurement details and qualifications.