FFmpeg’s Vulkan HEVC Encoder Hits H.264 Speed Parity

Software developers working with video pipelines have long accepted a tradeoff. Hardware-accelerated H.264 encoding flies. Its HEVC counterpart lags. Not anymore.

Today, a targeted set of changes landed in FFmpeg. The Vulkan-based HEVC encoder now matches H.264 encode performance on supported GPUs. The gap that frustrated transcoding farms, cloud encoders, and desktop users for years just closed.

From Second-Class Citizen to Peer

The optimizations center on the Vulkan Video encode path. A single commit adjusted parameter handling and feedback loops between the driver and the encoder. It introduced the ALLOW_ENCODE_PARAMETER_OPTIMIZATIONS flag. It enforced a minimum coding unit size of 16×16 when the coding tree block measured 32 or larger. And it synchronized GetEncoded SPS feedback for fields including extent, conf_win, minimum and differential CU and TB sizes, plus the transform hierarchy. One fix also corrected max_transform_hierarchy_depth_inter.

Benchmarks tell the story. Using a 1080p testsrc2 pattern at quality preset 1 for 1000 frames, the pre-optimization HEVC Vulkan encoder hovered between roughly 200 and 317 frames per second. Post-optimization numbers jumped to 289, 325, and 358 fps across test runs. H.264 Vulkan encode, for comparison, delivered 288, 317, and 359 fps. Parity achieved. The Phoronix report from August 17, 2026 highlighted the commit that made it happen: FFmpeg commit 426841da.

But this moment didn’t arrive in isolation. FFmpeg’s Vulkan Video support has built steadily. In September 2024 the project merged dedicated H.264 and H.265 encode paths. The H.264 implementation reached feature parity with VA-API. The HEVC side delivered full specification handling for I, P, and B frames. Those changes, covered by Phoronix in September 2024, set the stage. Today’s speed work completes the picture for real-time and high-throughput workloads.

Software encoders tell a parallel story. The libx265 encoder inside FFmpeg has always traded speed for compression efficiency. Guides from the FFmpeg wiki note that HEVC can deliver 25 to 50 percent bitrate savings over H.264 at similar quality. Yet encoding times stretch. A typical x264 preset might hit 40 frames per second on modest hardware. Drop to x265 and that often falls to 10 fps or lower, according to developer forums and Stack Overflow discussions that have persisted for years.

Recent x265 releases chip away at the problem. Version 4.2, released in April 2026, brought threaded motion estimation. The feature can deliver up to 1.5 times faster encoding on 1080p and lower resolutions when cores sit idle. An eight percent speed boost on ARM platforms arrived alongside new HEVC levels ready for 8K. The Free-Codecs coverage from April 2026 positioned the update as a shift from pure assembly tweaks to parallel precomputation. Later updates added SIMD improvements across loop filters, DCT, motion compensation, and more. Release notes for x265 document gains of nearly one percent from loopfilter optimizations alone on modern CPUs, with larger jumps on architectures such as RISC-V and AArch64.

These software gains matter. Many pipelines still rely on CPU-based encoding for quality control or when GPU resources stay locked behind other tasks. Cloud providers weighing HEVC licensing costs against AV1 adoption watch these numbers closely. Faster x265 shrinks the penalty for choosing better compression.

Hardware paths, though, drive the biggest shifts in production. AMD, Intel, and NVIDIA GPUs all expose video encode blocks. FFmpeg’s VA-API, AMF, and NVENC interfaces have matured. Yet Vulkan promised a cleaner, cross-vendor route. The API avoids some vendor-specific quirks. It scales across Linux desktops, embedded systems, and potentially Windows through the same code. Early Vulkan Video decode support reached FFmpeg 6.1. Encode followed. Now the performance gap has narrowed to zero on at least one key metric.

Real-world impact will vary. A 1080p synthetic test at low quality shows the ceiling. Content with motion, grain, or 4K resolution may expose different bottlenecks. Driver versions matter. GPU architecture matters. Still, the direction is clear. Engineers can plan HEVC pipelines without automatically defaulting to H.264 for speed. Storage savings and bandwidth reductions become practical without throughput penalties.

But don’t expect every workload to flip overnight. Legacy pipelines built around VA-API or NVENC won’t vanish. Licensing questions around HEVC patents linger in some markets. And AV1, despite its royalty-free appeal, still carries its own speed and compatibility tradeoffs. FFmpeg developers continue pushing patches for AV1 Vulkan encode as well.

The broader lesson sits in the details. Small adjustments to parameter flags and feedback synchronization delivered outsized gains. That’s typical in video codec work. Decades of accumulated micro-optimizations separate a usable encoder from one that dominates benchmarks. Today’s change adds another layer.

Teams integrating the latest FFmpeg master will see the difference immediately when targeting Vulkan HEVC. Others will wait for the next formal release. Either way, the gap that defined video encoding for half a decade just disappeared. Production encoders just got a quiet but meaningful upgrade.