CPU vs Graphics Card/GPU (Nvidia NVENC): x264 video encoding benchmark

July 30, 2019 posted by

Back in 2013, Nvidia introduced ShadowPlay – a video capture tool that was far lighter on performance than traditional
CPU recording, thanks to NVEENC Nvidia’s dedicated h.264 video encoder
built into all GeForce cards from the 600 series and up. The encoder is also capable
of rendering videos in traditional video editing programs and can be
significantly faster than the traditional encoding process that uses
only the CPU. In this video we’ll take a look at just how much faster GPU
accelerated encoding is compared to its CPU counterpart. The software used in
this comparison is Vegas Pro 15, which not only features built-in presets for
NVENC, but said presets are also identical in video parameters to
traditional presets that use only the CPU, making for an ideal apples-to-apples
comparison. Case in point, the two presets we’ll be using are the internet HD 1080p
59.94 fps and the internet HD 1080p 59.94 fps Nvidia NVENC,
with the only difference between them being that the latter makes use of the
dedicated hardware encoder on the graphics card. Let’s select the folder to save the
file and then start rendering. Another convenience that Vegas offers is keeping
track of the elapsed time, which lets us speed up our comparison. While the CPU is
hard at work, let’s open Hardware Info 64 to look at
the power consumption. The CPU is stable at around 68 Watts, while the 1080 Ti is
between 64 to 66 watts with the GPU voltage at 0.844 Volts. Together the CPU on a graphics card are consuming around 132 to 144
watts of electricity. Now that we’ve taken our measurements, let’s
fast-forward. Traditional CPU software encoding took 9 minutes and 43
seconds or exactly 583 seconds to render this video. Let’s see how much time
Geforce’s encoder can save. Go to file, then render as. Choose the ‘Internet HD
1080p 59.94 FPS Nvidia NVENC’ preset and render.
Let’s take another look at the hardware power consumption. CPU is consuming
around 58 to 62 watts, while the 1080 Ti is consuming between 82
to 86 watts with the GPU voltage stable at 1.050 Volts. Together that’s around
140 to 146 watts: about 10 watts greater than the CPU only encoding. The GPU
accelerated process completes at 2 minutes and 55 seconds or precisely 175
seconds – just 30% of the time conventional CPU encoding took to record
exact same video. ShadowPlay was used to record both encoding processes which
could have slowed NVENC’s rendering in the second process. However, after
recording the compression I rendered the video again using the NVENC preset
WITHOUT Shadowplay recording in the background and the difference was around
1 to 2 seconds – well within the margin of error. What can we conclude from this
analysis? If you have a GeForce graphics card, using NVENC can cut down the time
it takes to render your videos by 70 percent compared to traditional CPU only
encoding. Thank you for watching our analysis. There is more in-depth content
on the way, so please like, subscribe and hit the bell icon so you don’t miss out.


9 Replies to “CPU vs Graphics Card/GPU (Nvidia NVENC): x264 video encoding benchmark”

  1. SagnolTheGangster says:

    I think the best would be to record with a camera because software takes cpu bandwidth..

  2. Bass Junkie says:

    nice video man well done

  3. Th W. says:

    Are you a robot?

  4. Ajinkya Jumbad says:

    this is beautiful

  5. Arti csm says:

    Nice voice, robot 😉

  6. izw hechtsuppe says:

    You didn't actually use x264 at all in this comparison. You did a mainconcept avc vs. nvenc comparison. You should have matched bitrate and quality using a visual comparision of clips or psnr/ssim (disable psy tuning in encoders). Just cranking out valid h264 streams quickly is easy, doing it with good compression while preserving visual quality is much much harder and the default settings on software encoders tend to put more weight on the quality/compression side of the trade-off when compared to hardware-encoding solutions.

    Your comparison is not "apples-to-apples" AT ALL. x264 with veryfast preset might be comparable to nvenc but i haven't tested it.

  7. Linus PvP says:

    Only CPU (System: 144W): 9 minutes and 43 seconds
    CPU + GPU(System: 146W): 2 minutes and 55 seconds

  8. Alex Zubkov says:

    nonononono, i just can't rely upon someone who is using GUI for all those tests. Cmman man, get some linux distro and compile ffmpeg build with hardware acceleration feature and post some results. I think it's fair

  9. Zorrobeat says:

    Are the output files identical?
    Which one is encoded more efficient?

Leave a Comment

Your email address will not be published. Required fields are marked *