MKV -> MP4 CPU Choice
MKV -> MP4 CPU Choice
For HEVC it prioritizes core performance rather than raw throughput. With my partner now handling the work, I'm uploading smaller files since bandwidth is limited on ship internet. I can afford slower encodes now. Increasing from 3900X to 3950X only boosted encoding speed by about 15% in handbrake. Having more cores lets me run more jobs at once, which helps overall. Even with less than half the CPU usage, doing two jobs simultaneously makes sense.
I evaluated the quality and found it to be unusually poor. NVENC with Turing or higher performs noticeably better than x264, and HEVC is expected to offer only a modest improvement over H.264—around 25-50%.
In short, I'm looking at two HEVC encodings from identical input, one using NVENC and the other via software. We're not comparing H.264 versus HEVC here. At a fixed Constant Fate Factor, NVENC produces files roughly twice the size of software encoding when set to 'Slower'. NVENC is built as a dedicated ASIC, offering peak speed but not optimal for file size. Software encoders keep getting better, though they take longer on a regular CPU. For my needs—like compressing hours of anime into a compact file for a ship at sea—I prioritize efficiency over bitrate. It's about delivering content quickly and reliably, not squeezing every byte.
I realize I was only mentioning that since a direct match wasn't available. HEVC outperforms H264 by roughly 25-50% (likely in software). NVENC HEVC is expected to be similar to x264, not double the performance. Perhaps anime isn't the main issue? NVENC doesn't handle CRF support as I thought. CQ tends to be less efficient with that setting. The guide on slhck explains that keeping a constant QP at 18 maintains it across frames, adjusting it slightly for motion levels. Also, CRF and Rate Factor don't align between encoders.
HEVC is tailored for 1080p and above resolutions, delivering higher quality at lower noticeable levels when the resolution is above that threshold. Below 1080p, especially in standard formats like DVDs (720p or less), the improvement becomes minimal. Contemporary codecs often limit their thread count per frame to optimize performance; increasing threads beyond a certain point yields diminishing returns and can even hurt quality. It's usually wiser to use fewer threads but process multiple videos simultaneously, adjusting each encoder to around 12 threads for a combined total of 24.
Many people note that x264 suggests roughly one thread per 64 vertical pixels. Once you surpass 16 threads on high-resolution content, the benefit from additional threads diminishes significantly. You can still run two encoders at once, setting each to about 12 threads, achieving a total of 24 threads.
NvEnc offers less flexibility compared to software encoders and prioritizes speed and simplicity, aiming to reduce hardware requirements. This design choice means it can't perform as in-depth analysis as a dedicated software encoder would. Software encoders, when properly configured, can maintain better quality within the same storage footprint but demand more CPU resources.
Anandtech provides benchmark results showing x264’s speed on AMD Ryzen processors versus Intel, highlighting that while AVX extensions may boost Intel performance, they also increase power consumption.