Use H.264 (AVC) as a compatibility baseline, especially when you cannot control every playback device. Consider H.265 (HEVC) for fleets that already support the profiles you need. Test AV1 where your encoder and clients support it and measured bitrate savings justify the change. The best choice depends on both ends of the pipeline: a device that decodes AV1 does not necessarily encode it.
At-a-glance comparison
| Property | H.264 (AVC) | H.265 (HEVC) | AV1 |
|---|---|---|---|
| Specification | ITU-T H.264 | ITU-T H.265 | AOMedia AV1 |
| Bitrate at target quality | Useful baseline to measure | Potential savings over AVC; measure on your clips | Designed for efficient compression; measure on your clips |
| Software encoder examples | x264 | x265 | libaom, SVT-AV1, rav1e |
| Hardware encode examples | NVENC on supported NVIDIA GPUs | NVENC on supported NVIDIA GPUs; check profile and bit depth | GeForce RTX 4090 in NVIDIA's matrix; Apple's M3 specification lists AV1 decode |
| Hardware decode examples | Supported NVIDIA GPUs and the Apple M3 media engine | Supported NVIDIA GPUs and the Apple M3 media engine | GeForce RTX 3090 and 4090, Apple M3 |
| Playback verification | Test the exact container and AVC profile | Test the HEVC profile, bit depth and OS/browser path | Test the AV1 profile, container and actual decoder path |
| Licensing | Check applicable AVC terms | Check applicable HEVC terms | AOMedia offers a royalty-free patent license under its terms |
The NVIDIA encode/decode support matrix and Apple's M3 MacBook Pro specifications are the sources for these hardware examples, checked November 7, 2026. Resolution, chroma sampling, bit depth and profiles are capability limits of a specific codec implementation; a single maximum-resolution row would hide those conditions.
Compression efficiency: what "50% better" actually means
A bitrate saving needs a baseline and a quality target. If one encoder needs 16 Mbps and another needs 8 Mbps at the same measured quality target, the second uses 50% less bitrate for that comparison. That arithmetic says nothing about the quality score improving by 50%. A BD-rate comparison summarizes a bitrate difference over the overlapping portion of measured rate-quality curves, rather than relying on one pair of points.
Here is an illustrative storage calculation, not a benchmark or a claim that these bitrates produce equal quality. Assume a 60-second clip, decimal megabits and megabytes, and video payload only. Audio and container overhead are excluded.
| Example encoder output | Assumed video bitrate | Payload for 60 seconds |
|---|---|---|
| AVC baseline | 16 Mbps | 120 MB |
| HEVC candidate | 10 Mbps | 75 MB |
| AV1 candidate | 8 Mbps | 60 MB |
To find your own tradeoff, encode the same set of clips at several rate points. Keep resolution, frame rate, color format and bit depth fixed. Record the encoder build, preset, rate-control settings and hardware. Measure video-only bitrate and a named quality metric, then inspect representative frames and motion as well. Report the comparison only over quality ranges both encoders actually reached.
Encoding cost: the practical asymmetry
A standard specifies the bitstream and decoding behavior; it leaves encoder search and performance to the implementation. Presets trade search effort for compression. Comparing a fast x264 preset to a slow libaom preset does not give an intrinsic speed ratio between H.264 and AV1, and a software encoder result does not predict an NVENC result.
For live work, record sustained frames per second, end-to-end delay, peak load and available parallel sessions on the actual device. For offline work, measure elapsed encode time and the resulting bytes at your quality target. Use the same workload and constraints when comparing candidates.
Hardware decode: where adoption actually happens
Encode and decode are separate capabilities. NVIDIA lists AV1 decoding on the RTX 3090, while the RTX 4090 also supports AV1 encoding. Apple's M3 media-engine specification lists AV1 decode; it does not document AV1 hardware encoding. A platform can still offer a software encoder, which has a different cost profile.
Test the complete playback path: codec, profile, level, bit depth, container, OS, browser, driver and player. For web playback, Media Capabilities lets the player query an exact decoding configuration and ask whether the browser expects it to be supported, smooth and power efficient. Those signals help choose a stream; confirm behavior on representative devices. Keep a tested fallback for clients that cannot play your preferred version.
Patent licensing: the AV1 differentiator
The AOMedia Patent License grants a royalty-free license for covered necessary claims under its terms. That is a defined grant, not a statement about every patent holder or every deployment cost. AVC and HEVC have their own licensing considerations. Evaluate the applicable terms separately from encode capacity, playback reach and bandwidth savings.
Use-case decision matrix
| Use case | Starting choice | What to verify |
|---|---|---|
| Mixed or older playback fleet | H.264 | The actual profile, container and minimum supported device |
| Managed HEVC-capable fleet | H.265 | Decode path, HDR/bit-depth needs, and expected bandwidth savings |
| High-volume on-demand video | Test AV1 alongside the current codec | Quality at target bitrate, encoding budget, and fallback reach |
| Interactive video calls | Start with the endpoint's supported real-time formats | Negotiation, latency, thermal limits and performance under load |
| Security cameras and NVRs | H.264 or H.265 where the full pipeline supports it | Camera encode, recording, remote playback and retention cost |
| Live sports or esports | Test a supported hardware encoder | Sustained real-time throughput and a compatible viewer path |
| Web adaptive bitrate ladder | Keep a working baseline, add measured alternatives | Player selection, segment compatibility and actual fallback behavior |
How the codecs work under the hood
These standards share block-based prediction, transform coding, quantization and entropy coding. The encoder tries predictions from neighboring samples or reference frames, then codes the residual. H.265 and AV1 add partitioning and prediction options and use different in-loop filters. Their specifications define the legal block sizes, modes, transforms and profiles; the encoder decides how much search it can afford. More choices can help compression, but they do not establish a fixed percentage saving or a fixed encode slowdown.
For a deeper look at the shared pipeline, see H.264 Codec Fundamentals. Its prediction, transform, quantization and entropy-coding concepts provide a useful starting point for reading the newer specifications.
Reading further
- H.264 Codec Fundamentals (Part 1 of 3) — the encoding pipeline, block-based processing, motion estimation.
- H.264 Transform and Quantization (Part 2 of 3) — DCT, quantization, entropy coding.
- H.264 Implementation and Applications (Part 3 of 3) — profiles, levels, deployment patterns.
