Evidence

Verified, on camera

Not claims — encodes. The RTL reconstruction against the reference decoder, and the HEVC path choosing its own transform partitions.

Major milestone

Byte-exact full-trailer encode

H.264 High-10, the full 1,253-frame Sintel trailer at 256×256, 10-bit 4:2:0, QP20 — the RTL reconstruction is identical to the JM reference decode, frame for frame. Verified 2026-05-31.

Original
RTL recon
Diff ×24 (boosted)

0

Diffs vs JM ldecod_hi10

1,253

Frames — full trailer

40.1×

Compression · 246.3 → 6.14 MB

48.4dB

Mean luma PSNR @ QP20

H.265 multi-TU showcase

Recursive transform-unit splitting on the HEVC path — the encoder choosing TU partitions per CTU as the scene changes.



Request the data sheet

The power & area whitepaper and datasheet: per-resolution power tables, area breakdown by subsystem, DVFS operating points, coverage summary and the disclosed-gaps list. Distributed under NDA — see the boilerplate mutual NDA.