What Is H.264 (AVC)? The Codec Behind Most Online Video

H.264, also called Advanced Video Coding (AVC) or MPEG-4 Part 10, is a block-based video compression standard published jointly by ITU-T and ISO/IEC in 2003. It compresses video by predicting each 16x16 macroblock from nearby pixels or from earlier frames and then encoding only the leftover difference. Practically every browser, phone, smart TV and set-top box can decode it in hardware.

For a streaming team, what is H264 in practice? It is the fallback rendition in almost every HLS and DASH bitrate ladder in 2026. A 1080p30 H.264 stream at roughly 5 to 6 Mbps decodes on hardware sold over the last 15 years. HEVC and AV1 save an estimated 30 to 50 percent of bitrate at the same quality, but neither decodes on every device.

How the H.264 codec compresses a frame

The H.264 codec is a hybrid encoder. It combines prediction, a transform and entropy coding, and it uses a separate set of tools for each stage.

• Partitioning. Each frame is split into 16x16 macroblocks. A macroblock can be subdivided down to 4x4 blocks for motion prediction.
• Intra prediction. I-frames and intra blocks are predicted from already-decoded neighboring pixels, in 4x4, 8x8 (High profile only) or 16x16 modes.
• Inter prediction. P and B slices use motion vectors at quarter-pixel precision and can reference up to 16 earlier frames. B slices can also predict from both directions.
• Transform and quantization. The residual goes through an integer approximation of the DCT (4x4, plus 8x8 in High). The quantization parameter (QP) then sets how much detail is discarded.
• Entropy coding. The encoder uses either CAVLC or CABAC. CABAC costs more CPU and typically saves about 10 to 15 percent of bitrate (estimate).
• In-loop deblocking. A filter smooths block edges before the frame becomes a reference. This is what keeps low-bitrate AVC from looking like a mosaic.

The output is a stream of NAL (Network Abstraction Layer) units. The SPS and PPS parameter sets carry the profile, level and resolution. IDR frames are the only safe entry points: a decoder can start at an IDR without any earlier data.

Where AVC H264 sits in a streaming stack

AVC H264 is the payload, not the transport. An encoder such as x264 or a hardware encoder produces the elementary stream. A packager then wraps it in fragmented MP4 (CMAF) or MPEG-TS segments and writes HLS or DASH manifests. The CDN caches and serves those segments as ordinary files.

Players read the codec from a string in the manifest. For example, avc1.640028 means High profile, Level 4.0. The middle hex bytes are profile_idc (66 Baseline, 77 Main, 100 High), followed by the constraint flags and the level times ten. An incorrect codec string is a common reason a device rejects a rendition it could actually decode.

H264 profiles and levels explained

H264 profiles: which tools the decoder must support

H264 profiles define which coding tools a stream may use. Profiles do not set quality. They tell the decoder which features it must implement.

• Constrained Baseline: no B-frames and no CABAC. It was built for early mobile devices and video conferencing, and it is still the profile WebRTC requires.
• Main: adds B-frames and CABAC.
• High: adds the 8x8 transform and custom quantization matrices. It is the default for VOD and live streaming today.
• High 10, High 4:2:2, High 4:4:4 Predictive: add higher bit depth and chroma resolution. These are used for contribution feeds and production, and many consumer devices cannot decode them in hardware.

Levels: worked macroblock math

Levels cap decoder workload. The main limits are macroblocks per second, frame size in macroblocks and maximum bitrate. To find the minimum level for a stream, count macroblocks.

For 1080p30, the frame is coded as 1920x1088, which is 120 x 68 = 8,160 macroblocks. At 30 frames per second that is 244,800 macroblocks per second, which fits Level 4.0's limit of 245,760. At 60 frames per second it becomes 489,600 and needs Level 4.2.

Signal the lowest level your macroblock rate actually needs, because over-signalling a level is enough for some TVs and set-top boxes to refuse a rendition they could have decoded.

Encoder settings that matter when H.264 goes through a CDN

The CDN never sees motion vectors. It sees segment sizes, request rates and cache keys. Four encoder decisions shape all three.

1. Fixed, closed GOPs aligned to segments. A common pattern is a 2-second GOP (60 frames at 30 fps) with 4- or 6-second segments, so every segment starts on an IDR frame. In x264, set keyint and min-keyint to the same value and set scenecut to 0, or force keyframes at fixed timestamps. If scene-cut IDRs drift, segments across renditions stop aligning and ABR switches stall.
2. Capped VBR, not open VBR. Set a VBV maxrate somewhere between 10 and 50 percent above the target, and a bufsize of one to two seconds of maxrate. The HLS BANDWIDTH attribute must reflect peak segment bitrate. Uncapped spikes make players downshift and inflate edge throughput at the 95th percentile.
3. Deterministic output. Re-encoding the same title should produce stable segment names and URLs. Every new URL is a cold cache miss that goes back to origin.
4. Progressive MP4 with the moov atom first. For progressive download, move the index to the front of the file (ffmpeg's faststart option). The first byte-range request can then start playback instead of fetching the tail of the file.

The bitrate decision is also an egress decision. A 6 Mbps 1080p rendition works out to 6,000,000 bits per second times 3,600 seconds, divided by 8, which is 2.7 GB per viewer-hour. At 100,000 viewer-hours a month that is 270 TB, so trimming 10 percent of the top rung's bitrate removes about 27 TB of monthly delivery.

BlazingCDN's HLS streaming CDN for pre-encoded H.264 ladders delivers HLS, LL-HLS and DASH segments you have already encoded. It does not transcode, so these encoder settings stay in your own pipeline.

H.264 vs. HEVC, MP4 and x264

H.264 vs. HEVC (H.265): HEVC replaces macroblocks with coding tree units of up to 64x64 and saves an estimated 40 to 50 percent of bitrate at equal quality. It costs more encoder CPU, and its patent licensing is more fragmented. The usual 2026 ladder ships HEVC or AV1 to capable devices and keeps H.264 as the rendition every device can play.

H.264 vs. MP4: MP4 is a container format and H.264 is a codec. An MP4 file can hold H.264, HEVC or AV1 video. An H.264 stream can also live in MPEG-TS, Matroska or fragmented MP4.

H.264 vs. x264: x264 is an open-source encoder that implements the H.264 standard. Other implementations include hardware encoders such as NVENC and Quick Sync. Two encoders can both output valid H.264 at the same bitrate and still differ visibly in quality.

Common misconceptions about H.264

"High profile means higher quality." A profile only enables coding tools. High profile tends to reach a given quality at roughly 10 percent less bitrate than Main (estimate), but the bitrate and the preset decide the actual picture.

"Baseline is safest for compatibility." This was true around 2010. In 2026, High profile at Level 4.0 or 4.1 decodes in hardware on mainstream devices. Baseline now mostly belongs in WebRTC and in very old embedded players.

"H.264 is obsolete." Its compression is a generation behind. Its reach is not: H.264 is still the one rendition guaranteed to play everywhere.

FAQ: what is H264 and how AVC H264 behaves in streaming

Is H.264 the same as AVC?

Yes, H.264 and AVC are the same standard under two names. ITU-T publishes it as Recommendation H.264, and ISO/IEC publishes it as MPEG-4 Part 10, Advanced Video Coding. The two texts are technically aligned. Encoders, players and codec strings such as avc1 refer to the same bitstream format, so the names are interchangeable in streaming documentation and manifests.

Which H264 profile should I use for HLS streaming?

Use High profile for nearly all HLS and DASH renditions in 2026. Hardware decoders on phones, browsers and TVs support it, and it compresses better than Main or Baseline. Pick the level from the macroblock rate: 4.0 for 1080p30, 4.2 for 1080p60. Keep Constrained Baseline only for WebRTC or for confirmed legacy devices.

What is a good H.264 bitrate for 1080p?

Most streaming ladders put 1080p30 H.264 between about 4.5 and 8 Mbps. Simple content such as talking heads or slides sits at the low end. Sports and high-motion games need the top of the range or more. Use capped VBR with a VBV maxrate so peak segment bitrate stays predictable for player switching and for CDN throughput planning.

Why do H.264 keyframes matter for CDN delivery?

H.264 keyframes decide where a player can start or switch renditions. Every HLS or DASH segment should begin with an IDR frame at the same timestamp across the ladder. Fixed, closed GOPs that divide evenly into the segment duration keep renditions aligned. Stable segment URLs then avoid cold cache misses at the edge and repeated requests to origin.

Audit your H.264 ladder before you scale delivery

This week, run ffprobe across every rendition in your ladder. Record four things for each one: profile, level, keyframe spacing and peak segment bitrate. Flag any rendition that signals a higher level than its macroblock rate needs, any rendition whose IDR frames drift from segment boundaries, and any rendition whose peak segment bitrate exceeds its manifest BANDWIDTH value. Then multiply your top rung's bitrate by your monthly viewer-hours to estimate what a 10 percent bitrate cut would save in egress.

Once the ladder is clean, the BlazingCDN Video CDN for H.264 libraries replicates your encoded library inside the CDN and keeps the origin out of the delivery path. Its 14-day testing period runs on real production traffic, so you can check the result against your own segments.