Concrete Diagnosis and Steps
Use this sequence when the hardware encoder is selected but the reported speed is close to software encoding, or when CPU activity remains high.
- Record the complete job. Write down the source codec and resolution, selected video encoder and encoder preset, output resolution, audio tracks, subtitle handling, container, and every enabled video filter. Do not infer acceleration from a green or colored node in another application; use the actual type labels and HandBrake settings.
- Identify the hardware path. Check whether the job uses NVENC, VCN, QSV, or VideoToolbox, then consult the matching documentation. For NVENC, a video filter can disable hardware decoding and return decoding to the CPU. For VCN, decoding and all filters are CPU work. For QSV, decoding may be hardware-assisted only when supported and enabled, while most filters remain on the CPU. VideoToolbox has a documented GPU-capable filter list, but only with both hardware decoders and encoders selected.
- Separate required work from optional work. A crop that removes unwanted picture content, deinterlacing that fixes interlaced footage, subtitle rendering, or a required resize may be part of delivery. A filter added by habit is a candidate for removal, not an automatic error. Preserve any change that the brief, source condition, or destination requires.
- Create a controlled comparison. Keep the same source, encoder, quality or bitrate mode, resolution, audio, subtitles, container, and encoder preset. Run the original configuration and a second configuration with only confirmed-unnecessary filters disabled. Record elapsed time and reported frames per second, then inspect the output at the intended viewing size. This is a proposed comparison method, not a claim that a particular setting will be faster.
- Decide from the whole delivery. If the filter-free version is faster but changes framing, interlacing, subtitles, scale, or acceptable quality, retain the required filter. If it is materially faster and the visual and delivery requirements are unchanged, the filter was a plausible pipeline cost. If both versions are similar, investigate decoding support, audio, synchronization, muxing, storage, thermal limits, or the encoder preset instead of assuming the GPU is idle.
Acceptance Worksheet
The following is a hypothetical record for one 4K source. Its values illustrate a decision format; they are not HandBrake expectations or measured results.
| Case | Encoder and filters | Decode availability | Observed CPU / speed | Quality and decision |
|---|---|---|---|---|
| A | NVENC; crop/scale plus denoise | Software decode likely under the documented NVENC rule | 92% CPU, 41 fps | Framing and denoise pass; keep filters if required |
| B | NVENC; only confirmed-unnecessary denoise removed | Software decode still possible because crop/scale remains | 88% CPU, 44 fps | Small gain; crop/scale remains a possible CPU cost |
| C | NVENC; all optional filters removed, same output size | Hardware-decode eligibility requires checking the actual path | 54% CPU, 63 fps | Hypothetically acceptable picture; use only if delivery permits |
| D | QSV; resize plus subtitle rendering | Hardware decode may be available, most filters remain CPU work | 79% CPU, 52 fps | Compare subtitle and resize requirements before changing them |
The useful observation is not CPU percentage alone. Match it with the filter list, decode status, fps, elapsed time, output dimensions, interlacing result, subtitle presence, and a visual review. HandBrake’s performance documentation also warns that presets change several variables at once, so comparing unlike presets does not isolate a filter effect. See NVIDIA NVENC documentation, Intel QSV documentation, and performance guidance.
Limits and the Next Step After Failure
A hardware encoder can be fast while the job remains CPU-limited. High CPU usage is therefore compatible with successful hardware encoding, not proof that encoding has fallen back to software. Likewise, low GPU utilization is not by itself a complete diagnosis: audio, subtitles, synchronization, muxing, input decoding, and storage can limit throughput.
If removing an optional filter does not help, restore the required visual processing and compare the decoder and encoder selections for the specific path. Confirm that the source format is supported by the available hardware decoder where the documentation says this matters. Then compare audio and subtitle complexity, output resolution, and encoder preset one variable at a time. Do not change quality, codec, filters, and audio together and attribute the result to one cause.
For Apple systems, enabling hardware decoders may move the documented filter set to the GPU, but the VideoToolbox scaler uses a different algorithm and may produce a less sharp result than HandBrake’s software scaler. That is a quality tradeoff to inspect, not a promised improvement. AMD VCN’s documented lack of hardware decoding means removing a filter cannot create a hardware-decoding path in HandBrake. See AMD VCN documentation and Apple VideoToolbox documentation.
Three Short Questions
Does 100% CPU mean NVENC or QSV failed? No. The documentation assigns multiple parallel stages to the CPU. Treat it as a clue to inspect filters, decoding, audio, subtitles, and muxing, not as a pass/fail signal.
Should I disable crop and scale first? Only when the delivered framing and dimensions do not require them. For NVENC, crop and scale are specifically included among filters that can trigger software decoding, but removing them can change the result.
Why did a faster file look different? Hardware encoder speed and output appearance are separate acceptance criteria. Compare the same frame regions, text, edges, gradients, motion, dimensions, and audio before accepting the faster configuration.
Sources and next step
- HandBrake Documentation — NVIDIA NVENC
- HandBrake Documentation — Apple VideoToolbox
- HandBrake Documentation — AMD VCN
- HandBrake Documentation — Intel Quick Sync Video
- HandBrake Documentation — Performance
This guide interprets public documentation and uses explicitly hypothetical acceptance examples, not application tests. Controls and availability can change; verify the cited manual against your installed version.
For work created in Imagild, carry the input requirements and acceptance checklist into the film workspace. The external software controls discussed above belong to their respective products.