Effects of Handbrake presets and RF quality settings across AV1, H.265, and H.264

 

"Fantastic!" 

Every once in a great while, I dive deep down the rabbit hole of media formats and the codecs1 that encode and decode them. Sometimes it's photos, sometimes it's audio, and this time it's video. I'm no expert in these things, but rather an enthusiast who (a) likes to create and organize digital media and (b) likes knowing that the formats I'm choosing are going to meet my present and future needs. In the past few months I've been adding Blu-ray movies to my media server. I've relied on H.264/AVC as my video format for a long time (more than a decade, maybe?), but with H.265/HEVC now pretty mainstream and AV1 emerging, I figured it was time to refresh my knowledge and get to know how these different tools might serve me.

This is going to be a long post, so I'm going to take a moment here to establish the four considerations I make whenever I'm encoding media, whether it's video, audio, or photos. The four considerations when choosing a codec and its settings are compatibility, compression, quality, and encoding/decoding speed:

  • Compatibility: Regardless of its technical merits, a codec isn't any good unless you can create, edit, and/or view the file it produces.
  • Compression: The whole point of codecs is to make files that take up less storage space and transmit with less bandwidth. If space and bandwidth weren't issues, we could just sit back with uncompressed, lossless formats and not worry about compression.
  • Quality: Depending on the settings you choose, modern codecs produce media that can look as good as the original or files that are compressed beyond recognition.
  • Encoding/Decoding Speed: I debated whether or not I should call this "computational intensity" or something similar. Our modern devices are powerful enough these days to play a lot of formats well, but that doesn't mean we like to wait hours and hours for media to encode.

With those definitions in place, here's how I see the H.264/AVC, H.265/HEVC, and AV1 formats comparing as of early 2023:

  • H.264/AVC is compatible with just about everything, often with hardware acceleration, especially for typical 1920x1080 or smaller videos. Its compression is good but dated compared to newer codecs. Because it's seen a lot of refinement since being introduced in 2004, you can get high quality at relatively good encoding/decoding speeds.
  • H.265/HEVC was released in 2013 and it plays on most modern devices that play H.264/AVC. Its greatest compatibility problem lies in the licensing and patent issues that prevent commercial creators from encoding videos in the format without paying fees, but that's not really a factor for home users. Its compression is better than H.264/AVC for the same quality output but you should expect slower encoding/decoding speeds than H.264.
  • AV1 is the new kid on the block, having only been released in 2018. It's not compatible with a lot of devices, including many media devices being sold in stores right now. This is expected to change because AV1 has broad industry support (including Amazon, Google, and Netflix) and it's designed to be free from patent and licensing concerns. AV1 was added to Handbrake 1.6.0 on December 28, 2022, and to Plex Media Server 1.30.2.6563 on January 10, 2023, which are two huge steps towards widening its use. AV1's compression should be better than H.265/HEVC for the same quality output. Because there are multiple AV1 encoders in development and hardware support is still early, there is a lot to learn about its encoding/decoding speed. AV1 is going to be most important for 8K and 4K video, but it works well at lower resolutions, too.

Let's get to the testing.

Testing H.264, H.265, and AV1

"No touching of the hair or face...and that's it."


There are many recommendations on the internet about how to get the most out of various codecs. Handbrake's documentation is a great start. Because AV1 is so new, there's less information about its usage on the internet and some of what has been posted might be outdated or unreliable. I find that when it comes to discussing codec performance, enthusiasts can sometimes let their opinions run on the wild side, claiming that they can see or perceive dramatic differences in quality that others cannot. Some of these claims might be attributed to misconfigured software or unusual or defective source material, but I suspect many are probably due to confirmation bias or pure fantasy. To avoid this, I'll be using VMAF, an objective video quality metric developed by Netflix, the University of Southern California, and the University of Texas at Austin. For reference, a VMAF score of 93 is the level where streaming engineers find that the video is either indistinguishable from the original, or, in cases where it's not, the differences are slight and not annoying. I used the the FFMetrics software, v1.3.1, in Windows to calculate the VMAF scores, comparing 5 minutes of each output against the source, skipping the first minute. It's not a perfect measure, but it's far better than me posting a ton of screenshots to squint at or having me ramble on about perceived sharpness, smoothness, and saturation2.

My testing approach is pretty simple: Run a source video through Handbrake over and over (200+ times!) using a variety of codecs, presets, and quality settings, and collect the results. As my source, I selected one of the greatest action sequences ever produced in modern cinema: The big fight scene from Anchorman: The Legend of Ron Burgundy. Here are the stats for that chapter of the Blu-ray, in its original H.264 state ripped directly from the disc: 1920x1080 resolution, 6 minutes and 9 seconds, 1573.74 MB, and 8847 frames of video. I used Handbrake 1.6.0 on my Linux workstation running a Ryzen 9 5900x, with the exception of the AV1 (QSV) tests, which I ran on a Windows computer running a Ryzen 7 5700x and an Intel Arc A380 GPU.

"Aaarrrrgggghhhhh!!!"

Handbrake has a lot of customizable settings but in my testing I focused on two. The first is the "RF" level, which is the primary setting that controls video quality. To put it simply, lower RF numbers means more bits in the output video, and higher RF numbers means fewer bits in the output video. All other things being equal, more bits equals more quality. I kept changing the RF level until I got VMAF scores as high as 98 (which should almost always be indistinguishable from the original) or as low as 90 (which is still quite good). The other setting is the preset level. This one seems to get widely misunderstood, as some people think it has a significant effect on quality. While it can affect quality, what the preset mostly does is determine how many "efficiency features" the codec uses to compress the file, and to what degree. It makes more sense if you look at this table in the SVT-AV1 documentation or this table in the x265 (an H.265 encoder) documentation. Lower presets "try harder" to compress the file while maintaining quality, at the expense of (often dramatically) longer encoding times, while higher presets work faster but tend to yield bigger file sizes. How presets work can vary across codecs and be influenced by how you're specifying the bitrate, but don't worry about it too much. You'll see the effects below. I started with the "middle" presets in Handbrake and then went two steps up and two steps down. As for the rest of my Handbrake settings, I turned all the filters off, kept the framerate the same as the source, and left the Tune, Profile, and Level settings to their defaults, which for AV1 were None, Auto, and Auto.

In the tables below, you'll find the following information:


Preset Level
RF Level VMAF score
File size in MB (and as a percentage of the original)
Encoding time in minutes and seconds (and average frame rate)

AV1 (SVT)

With the AV1 (SVT) codec (8-bit; I didn't retest with 10-bit), I tried every combination between RF 23 and Preset 4 (which gave a VMAF score of 98.02) and RF 42 and Preset 8 (which gave a VMAF score of 89.69).


Preset 4 Preset 5 Preset 6 Preset 7 Preset 8
RF 23 98.02
396.16 (25.2%)
47:41 (3.1 fps)
97.89
376.48 (23.9%)
29:02 (5.1 fps)
97.84
382.09 (24.3%)
11:39 (12.7 fps)
97.89
415.11 (26.4%)
7:19 (20.2 fps)
97.52
345.21 (21.9%)
2:13 (66.5 fps)
RF 24 97.83
356.68 (22.7%)
47:27 (3.1 fps)
97.68
337.94 (21.5%)
28:58 (5.1 fps)
97.61
341.79 (21.7%)
11:39 (12.7 fps)
97.67
374.37 (23.8%)
7:36 (19.4 fps)
97.36
324.68 (20.6%)
2:26 (60.6 fps)
RF 25 97.58
313.07 (19.9%)
45:26 (3.2 fps)
97.40
292.77 (18.6%)
27:55 (5.3 fps)
97.30
294.73 (18.7%)
10:49 (13.6 fps)
97.39
328.46 (20.9%)
6:59 (21.1 fps)
97.11
294.82 (18.7%)
2:22 (62.3 fps)
RF 26 97.28
270.11 (17.2%)
43:06 (3.4 fps)
97.08
250.30 (15.9%)
27:05 (5.4 fps)
96.96
251.72 (16.0%)
10:06 (14.6 fps)
97.06
283.99 (18.0%)
6:38 (22.2 fps)
96.82
264.50 (16.8%)
2:21 (62.7 fps)
RF 27 96.98
233.44 (14.8%)
40:57 (3.6 fps)
96.76
216.80 (13.8%)
25:42 (5.7 fps)
96.91
217.17 (13.8%)
9:49 (15.0 fps)
96.72
246.07 (15.6%)
6:27 (22.9 fps)
96.50
236.07 (15.0%)
2:21 (62.7 fps)
RF 28 96.69
204.93 (13.0%)
39:11 (3.8 fps)
96.45
188.50 (12.0%)
23:55 (6.2 fps)
96.25
188.09 (12.0%)
9:09 (16.1 fps)
96.39
214.97 (13.7%)
5:56 (24.9 fps)
96.18
210.58 (13.4%)
2:18 (64.1 fps)
RF 29 96.39
178.52 (11.3%)
37.12 (4.0 fps)
96.15
166.88 (10.6%)
22:53 (6.4 fps)
95.92
165.56 (10.5%)
8:23 (17.6 fps)
96.08
189.69 (12.1%)
5:39 (26.1 fps)
95.86
188.39 (12.0%)
2:13 (66.5 fps)
RF 30 96.11
160.38 (10.2%)
35:43 (4.1 fps)
95.86
148.42 (9.4%)
22:04 (6.7 fps)
95.59
147.07 (9.3%)
7:58 (18.5 fps)
95.75
168.34 (10.7%)
5:19 (27.7 fps)
95.53
168.56 (10.7%)
2:10 (68.1 fps)
RF 31 95.77
140.40 (8.9%)
33:54 (4.3 fps)
95.49
130.48 (8.3%)
20:47 (7.1 fps)
95.19
129.20 (8.2%)
7:29 (19.7 fps)
95.39
147.77 (9.4%)
5:02 (29.3 fps)
95.15
149.27 (9.5%)
2:07 (69.7 fps)
RF 32 95.38
122.87 (7.8%)
31:46 (4.6 fps)
95.10
114.74 (7.3%)
20:02 (7.4 fps)
94.74
113.43 (7.2%)
6:57 (21.2 fps)
94.96
129.37 (8.2%)
4:48 (30.7 fps)
94.71
131.40 (8.3%)
2:04 (71.3 fps)
RF 33 95.01
109.47 (7.0%)
30:26 (4.8 fps)
94.72
102.95 (6.5%)
19:00 (7.8 fps)
94.31
101.80 (6.5%)
6:38 (22.2 fps)
94.55
115.77 (7.4%)
4:30 (32.8 fps)
94.28
117.55 (7.5%)
1:59 (74.3 fps)
RF 34 94.64
98.51 (6.3%)
28:57 (5.1 fps)
94.34
93.54 (5.9%)
17:53 (8.2 fps)
93.89
92.42 (5.9%)
6:17 (23.5 fps)
94.16
104.73 (6.7%)
4:18 (34.3 fps)
93.86
106.49 (6.8%)
1:56 (76.3 fps)
RF 35 94.23
88.54 (5.6%)
27:52 (5.3 fps)
93.13
84.93 (5.4%)
17:12 (8.6 fps)
93.44
84.04 (5.3%)
5:57 (24.8 fps)
93.73
95.28 (6.1%)
4:07 (35.8 fps)
93.41
96.69 (6.1%)
1:54 (77.6 fps)
RF 36 93.76
79.29 (5.0%)
26:08 (5.6 fps)
93.49
77.34 (4.9%)
16:52 (8.7 fps)
92.96
76.65 (4.9%)
5:41 (25.9 fps)
93.26
86.54 (5.5%)
3:57 (37.3 fps)
92.91
87.82 (5.6%)
1:51 (79.7 fps)
RF 37 93.31
72.47 (4.6%)
25:06 (5.9 fps)
93.07
71.50 (4.5%)
16:00 (9.2 fps)
92.48
70.85 (4.5%)
5:30 (26.8 fps)
92.81
79.75 (5.1%)
3:45 (39.3 fps)
92.43
80.80 (5.1%)
1:48 (81.9 fps)
RF 38 92.79
65.23 (4.1%)
23:42 (6.2 fps)
92.58
65.57 (4.2%)
14:54 (9.9 fps)
91.95
65.09 (4.1%)
5:12 (28.4 fps)
92.30
72.37 (4.6%)
3:35 (41.1 fps)
91.92
74.04 (4.7%)
1:46 (83.5 fps)
RF 39 92.30
60.39 (3.8%)
22:37 (6.5 fps)
92.10
61.00 (3.9%)
14:16 (10.3 fps)
91.41
60.57 (3.8%)
5:03 (29.2 fps)
91.79
67.21 (4.3%)
3:27 (42.7 fps)
91.40
68.66 (4.4%)
1:44 (85.1 fps)
RF 40 91.76
55.76 (3.5%)
21:43 (6.8 fps)
91.61
56.97 (3.6%)
13:48 (10.7 fps)
90.89
56.71 (3.6%)
4:54 (30.1 fps)
91.27
62.68 (4.0%)
3:21 (44.0 fps)
90.87
64.10 (4.1%)
1:42 (86.7 fps)
RF 41 91.17
51.52 (3.3%)
21:09 (7.0 fps)
91.08
53.27 (3.4%)
13:20 (11.1 fps)
90.30
53.05 (3.4%)
4:45 (31.0 fps)
90.71
58.34 (3.7%)
3:21 (44.0 fps)
90.27
59.70 (3.8%)
1:40 (88.5 fps)
RF 42 90.60
48.02 (3.1%)
20:12 (7.3 fps)
90.55
50.10 (3.2%)
12:43 (11.6 fps)
89.71
49.86 (3.2%)
4:44 (31.2 fps)
90.14
54.66 (3.5%)
3:10 (46.6 fps)
89.69
55.94 (3.6%)
1:39 (89.4 fps)

Based on these tests, I'm encoding most things with RF 30, Preset 8. That gives me a VMAF score above 95, encoding speeds that are more than twice the framerate as the source material, and compression that I'm happy with. Also, I was surprised that changing presets to make the codec "try harder" to compress didn't always mean it succeeded. Look at the RF 30 row — who would expect encoding a file for 2 minutes on Preset 8 would result in a file only a few megabytes larger than one encoded for 35 minutes on Preset 4? And at lower RF values, the Preset 4 files are often larger than the Preset 8 files!

H.265/HEVC

With H.265/HEVC, I again used the 8-bit encoder and tried every combination between RF 19 with the slower preset (which gave a VMAF score of 98.32) and RF 28 with the faster preset (which gave a VMAF score of 89.14).


Preset Slower Preset Slow Preset Medium Preset Fast Preset Faster
RF 19 98.32
662.52 (42.1%)
53:43 (2.7 fps)
98.36
657.96 (41.8%)
12:34 (11.7 fps)
97.62
562.24 (35.7%)
5:02 (29.3 fps)
97.39
498.41 (31.7%)
2:41 (55.0 fps)
97.37
498.9 (31.7%)
2:31 (58.6 fps)
RF 20 97.92
535.44 (34.0%)
49:03 (3.0 fps)
97.98
531.53 (33.8%)
11:52 (12.4 fps)
97.13
451.23 (28.7%)
4:39 (31.7 fps)
96.88
405.29 (25.8%)
2:35 (57.1 fps)
96.85
405.62 (25.8%)
2:24 (61.4 fps)
RF 21 97.46
426.23 (27.1%)
44:46 (3.3 fps)
97.53
421.51 (26.8%)
11:04 (13.3 fps)
96.58
358.65 (22.8%)
4:20 (34.0 fps)
96.30
324.59 (20.6%)
2:29 (59.4 fps)
96.25
324.73 (20.6%)
2:18 (64.1 fps)
RF 22 96.93
335.12 (21.3%)
40:16 (3.7 fps)
97.02
329.03 (20.9%)
10:15 (14.4 fps)
95.96
283.89 (18.0%)
3:59 (37.0 fps)
95.65
258.74 (16.4%)
2:23 (61.9 fps)
95.57
258.64 (16.4%)
2:12 (67.0 fps)
RF 23 96.33
261.11 (16.6%)
36:14 (4.1 fps)
96.45
253.98 (16.1%)
9:23 (15.7 fps)
95.27
223.96 (14.2%)
3:40 (40.2 fps)
94.92
205.73 (13.1%)
2:17 (64.6 fps)
94.81
205.29 (13.0%)
2:06 (70.2 fps)
RF 24 95.70
202.96 (12.9%)
32:58 (4.5 fps)
95.84
196.05 (12.5%)
8:33 (17.2 fps)
94.47
176.65 (11.2%)
3:23 (43.6 fps)
94.10
164.15 (10.4%)
2:11 (67.5 fps)
93.94
163.45 (10.4%)
1:58 (75.0 fps)
RF 25 95.05
158.93 (10.1%)
30:16 (4.9 fps)
95.20
153.36 (9.7%)
7:52 (18.7 fps)
93.56
140.51 (8.9%)
3:06 (47.6 fps)
93.19
132.54 (8.4%)
2:06 (70.2 fps)
92.95
131.62 (8.4%)
1:52 (79.0 fps)
RF 26 94.36
126.17 (8.0%)
28:01 (5.3 fps)
94.51
122.18 (7.8%)
7:16 (20.3 fps)
92.52
112.67 (7.2%)
2:54 (50.8 fps)
92.17
107.82 (6.9%)
2:01 (73.1 fps)
91.82
106.81 (6.8%)
1:46 (83.5 fps)
RF 27 93.61
102.18 (6.5%)
26:02 (5.7 fps)
93.72
99.18 (6.3%)
6:48 (21.7 fps)
91.34
91.61 (5.8%)
2:42 (54.6 fps)
91.02
88.91 (5.6%)
1:57 (75.6 fps)
90.55
87.80 (5.6%)
1:42 (86.7 fps)
RF 28 92.75
84.36 (5.4%)
24:24 (6.0 fps)
92.79
82.00 (5.2%)
6:23 (23.1 fps)
90.01
75.90 (4.8%)
2:34 (57.4 fps)
89.77
74.69 (4.7%)
1:53 (78.3 fps)
89.14
73.48 (4.7%)
1:37 (91.2 fps)

Compared to AV1, you can see that H.265 has bigger file sizes for the same quality. Somewhat surprisingly, the encoding times aren't all that different. The best comparison I can find for AV1's RF 30, Preset 8 that I prefer (95.53 VMAF, 168.56 MB, 68.1 fps) is H.265's RF 22, faster preset (95.57 VMAF, 258.64 MB, 67.0 fps). The quality and speed are almost exactly the same but the AV1 file is only 65% the size of the H.265 file.

H.264/AVC

With H.264/AVC, I tried every combination between RF 18 with the slower preset (which gave a VMAF score of 98.25) and RF 28 with the faster preset (which gave a VMAF score of 88.34).


Preset Slower Preset Slow Preset Medium Preset Fast Preset Faster
RF 18 98.25
1030.62 (65.5%)
5:52 (25.1 fps)
98.18
1032.72 (65.6%)
3:08 (47.1 fps)
98.10
1033.96 (65.7%)
1:57 (75.6 fps)
98.10
1050.35 (66.7%)
1:33 (95.1 fps)
98.07
1018.4 (64.7%)
1:08 (130.1 fps)
RF 19 97.86
859.02 (54.6%)
5:28 (27.0 fps)
97.77
862.22 (54.8%)
2:57 (50.0 fps)
97.70
872.67 (55.5%)
1:52 (79.0 fps)
97.67
903.40 (57.4%)
1:30 (98.3 fps)
97.61
867.83 (55.1%)
1:07 (132.0 fps)
RF 20 97.41
695.99 (44.2%)
5:05 (29.0 fps)
97.32
696.92 (44.3%)
2:45 (53.6 fps)
97.24
707.96 (45.0%)
1:45 (84.3 fps)
97.17
750.56 (47.7%)
1:25 (104.1 fps)
97.08
717.09 (45.6%)
1:04 (138.2 fps)
RF 21 96.91
546.77 (34.7%)
4:42 (31.4 fps)
96.81
547.53 (34.8%)
2:33 (57.8 fps)
96.73
549.90 (34.9%)
1:39 (89.4 fps)
96.60
591.62 (37.6%)
1:21 (109.2 fps)
96.47
563.99 (35.8%)
1:02 (142.7 fps)
RF 22 96.34
419.54 (26.7%)
4:18 (34.3 fps)
96.23
423.12 (26.9%)
2:22 (62.3 fps)
96.13
416.78 (26.5%)
1:33 (95.1 fps)
95.95
450.09 (28.6%)
1:18 (113.4 fps)
95.76
427.25 (27.1%)
0:59 (149.9 fps)
RF 23 95.66
319.71 (20.3%)
3:58 (37.2 fps)
95.54
324.60 (20.6%)
2:08 (69.1 fps)
95.42
316.10 (20.1%)
1:27 (101.7 fps)
95.18
343.74 (21.8%)
1:13 (121.2 fps)
94.92
325.74 (20.7%)
0:57 (155.2 fps)
RF 24 94.83
243.40 (15.5%)
3:39 (40.4 fps)
94.70
248.43 (15.8%)
1:57 (75.6 fps)
94.58
243.86 (15.5%)
1:22 (107.9 fps)
94.30
265.10 (16.8%)
1:09 (128.2 fps)
93.97
252.31 (16.0%)
0:55 (160.9 fps)
RF 25 93.82
188.07 (12.0%)
3:24 (43.4 fps)
93.68
191.81 (12.2%)
1:46 (83.5 fps)
93.59
191.37 (12.2%)
1:16 (116.4 fps)
93.27
204.86 (13.0%)
1:04 (138.2 fps)
92.86
195.40 (12.4%)
0:52 (170.1 fps)
RF 26 92.60
147.89 (9.4%)
3:13 (45.8 fps)
92.43
150.09 (9.5%)
1:37 (91.2 fps)
92.38
152.55 (9.7%)
1:14 (119.6 fps)
92.05
160.81 (10.2%)
1:03 (140.4 fps)
91.54
153.45 (9.8%)
0:50 (176.9 fps)
RF 27 91.15
118.96 (7.6%)
3:06 (47.6 fps)
90.95
120.02 (7.6%)
1:31 (97.2 fps)
90.96
124.38 (7.9%)
1:11 (124.6 fps)
90.63
129.43 (8.2%)
1:00 (147.5 fps)
90.04
123.92 (7.9%)
0:49 (180.6 fps)
RF 28 89.47
98.45 (6.3%)
3:00 (49.2 fps)
89.24
99.06 (6.3%)
1:26 (102.9 fps)
89.34
103.91 (6.6%)
1:08 (130.1 fps)
89.02
107.06 (6.8%)
0:59 (149.9 fps)
88.34
103.27 (6.6%)
0:48 (184.3 fps)

H.264/AVC is old, predictable, and fast, but file sizes are so much bigger than the newer formats. Here, the best comparison I can find for AV1's RF 30, Preset 8 (95.53 VMAF, 168.56 MB, 68.1 fps) is H.264's RF 23, slow preset (95.54 VMAF, 324.60 MB, 69.1 fps). The quality and speed are almost exactly the same but the AV1 file is only 52% the size of the H.264 file.

AV1 (Intel QSV)

I picked up an ASRock Intel A380 graphics card which features AV1 hardware encoding and decoding. The knock on hardware encoders is that they're very fast but the speed comes at the cost of quality. But by how much? For these tests, I tried every ICQ (constant quality, like RF) setting between 20 and 30 on all three preset settings: quality, balanced, and speed. Like the other codecs, this gave me VMAF scores from almost 98 to 90.


Quality Balanced Speed
ICQ 20 97.68
521 (33.1%)
0:45 (196.6 fps)
97.65
519 (33.0%)
0:36 (245.8 fps)
97.64
522 (33.2%)
0:34 (260.2 fps)
ICQ 21 97.69
520 (33.0%)
0:45 (196.6 fps)
97.66
518 (32.9%)
0:36 (245.8 fps)
97.62
519 (33.0%)
0:34 (260.2 fps)
ICQ 22 97.62
506 (32.2%)
0:50 (176.9 fps)
97.58
502 (31.9%)
0:36 (245.8 fps)
97.53
503 (32.0%)
0:34 (260.2 fps)
ICQ 23 97.30
447 (28.4%)
0:50 (176.9 fps)
97.21
437 (27.8%)
0:36 (245.8 fps)
97.21
445 (28.3%)
0:34 (260.2 fps)
ICQ 24 96.65
355 (22.6%)
0:50 (176.9 fps)
96.55
350 (22.2%)
0:36 (245.8 fps)
96.54
358 (22.7%)
0:34 (260.2 fps)
ICQ 25 95.61
258 (16.4%)
0:46 (192.3 fps)
95.47
255 (16.2%)
0:36 (245.8 fps)
95.42
262 (16.6%)
0:34 (260.2 fps)
ICQ 26 94.41
191 (12.1%)
0:46 (192.3 fps)
94.28
190 (12.1%)
0:36 (245.8 fps)
94.15
194 (12.3%)
0:34 (260.2 fps)
ICQ 27 93.69
164 (10.4%)
0:50 (176.9 fps)
93.55
164 (10.4%)
0:36 (245.8 fps)
93.37
167 (10.6%)
0:34 (260.2 fps)
ICQ 28 92.61
134 (8.5%)
0:50 (176.9 fps)
92.49
135 (8.6%)
0:36 (245.8 fps)
92.25
137 (8.7%)
0:34 (260.2 fps)
ICQ 29 91.48
113 (7.2%)
0:50 (176.9 fps)
91.36
114 (7.2%)
0:36 (245.8 fps)
91.09
116 (7.4%)
0:33 (268.1 fps)
ICQ 30 90.25
95 (6.0%)
0:50 (176.9 fps)
90.09
96 (6.1%)
0:36 (245.8 fps)
89.78
98 (6.2%)
0:34 (260.2 fps)

Compared to the software codecs, here the preset doesn't matter as much and the speed is the same up and down the quality levels. Comparing again to my preferred AV1 software settings, RF 30 with Preset 8 (95.53 VMAF, 168.56 MB, 68.1 fps), the QSV hardware encoder gets similar quality at ICQ 25 with the balanced preset (95.47 VMAF, 255 MB, 245.8 fps). The file size is 51% bigger with the hardware encoder, but encoding happens almost four times faster. It's a tradeoff that some will choose, but I'm pleased with the results I'm getting with the software AV1 encoder on my Ryzen 9 5900x. This is also one of the rare cases where I can see the slightest differences between videos with similar VMAF scores and I prefer what I see with the software encoder.

Conclusion and why your mileage may vary

"I mean, that really got out of hand fast."

For the last two months I've been encoding everything with AV1 and Opus audio. It's a bit of a gamble because it's not compatible with everything yet, but it encodes fast enough for me with a significant space savings over H.265 and H.264. Right now, my Plex server has to transcode my AV1 files if I'm playing them on my Roku devices or my Google Chromecast with Google TV. It's mostly okay, although as I write this the Plex app for Android (which is on the Chromecast) thinks it supports playback of AV1 when it doesn't, so I have to manually tell it to transcode3. That's annoying but I assume it will be temporary. I also have the cheaper, newer, HD version of the Google Chromecast with Google TV, which supports AV1 decoding, and on it the direct play of AV1 files works great.

I have to be reminded that all of the testing above was with one clip from one movie, and I would have gotten different results with a different source. In fact, today I was encoding Ken Burns's The Civil War and was really struggling to get the file size low, which is a surprise for a film with so many still black and white shots. AV1 has some synthetic grain settings that I experimented with but ended up simply changing to RF 33 and Preset 6 to cut the file size in half compared to RF 30 and Preset 8. There are going to be exceptions like this on occasion, but for most Blu-ray discs I've been satisfied with AV1 (SVT) 8-bit at RF 30 and Preset 8. In casual viewing, I can't remember ever noticing it was different from the original and the file sizes are so, so small. With any luck, AV1 will be the format we stick with for the next 10+ years, much as we've been able to do with H.264/AVC.


1 Disclaimer: Yes, I know that there's a difference between the format, say H.264, and a codec, like x264. But we kinda all know these go hand-in-hand, right? I'll be specific where I need to be.

2 That said, you do you. If your encoding choices make you happy, then keep them and stay happy.

3 Fun fact: People will tell you all the time to avoid transcoding because it's harmful for image quality. I think Plex's default transcode settings are to use H.264 using the very fast preset and a bitrate of 8000 kbps. When I transcoded one of my AV1, RF 30, Preset 8 output files with a VMAF above 95 to H.264 with those same settings, the double-transcoded file still got a VMAF score of 93 compared to the source directly from the Blu-ray. In other words, your media can survive two transcodes and still look good enough that you don't have to worry about it.

Popular posts from this blog

My Podcast Predilection

Think before you shoot