x265 HEVC Encoder

[Z2697]

Originally Posted by Z2697

x265 will produce non-conformative bitstream when a very sudden change in chroma channel happens and triggered weighted prediction.

Some "synthetic" inupt to trigger that bug

import vapoursynth as vs
core=vs.core
i=core.imwri.Read('a.png').resize.Bicubic(format=vs.YUV420P8,matrix=1)
f1=core.std.BlankClip(i,color=[16,16,16])

f1=core.std.Merge(i,f1,0.9)
f2=i
f3=f2.std.FlipHorizontal()

clip=f1*30+f2*9+f3*30 #69
clip.set_output()

(This does not mean only synthetic input can trigger the bug. There are "legit" buggy videos in the wild)
Example encoding commandline:

vspipe aa.vpy -c y4m - | x265 --input - --y4m -b0 -o aa.265

I have to set bframes to 0, otherwise even with --weightb enabled, the weightd prediction won't catch up.
(Again this does not mean the bug will only be triggered when bframes=0)

I guess most images will work, but you can use this lovely nostalgic wallpaper:


A patch was submitted:
https://mailman.videolan.org/piperma...ay/014363.html
and you can see the process of figuring out the fix at https://github.com/Mr-Z-2697/x265-ex...e/E-2025-05-12

[excellentswordfight]

Originally Posted by benwaggoner

More to the point "the difference between AVX2 and AVX512 used to be marginal, BECAUSE of AVX512 thermal throttling."

The per clock throughput improvements were solid, but the throttling reductions in instructions per second nearly cancelled those out. An implementation that maintains IPS with the IPC gains would be quite impressive.

I dont think this is totally true, i've encoded on pretty muc every generation of xeon and epyc with avx512 support, still have not seen over 10% improvement. And isnt that also expected, I dont see why an encoder like x265 would benefit that much from 512bit wide registers. And in the case with downclocking, it wasnt a marginal difference for the early implementations, it was more outright an performance regression. I think the only systems were this didnt happen was were you could allow for the huge increase in powerdraw.

Originally Posted by _DLS_

In my tests with the 9950X3D, the speed bump with --asm avx512 can be between 16-36% depending on the other settings too.

on 2160p:

CTU 32, ref 4, subme 4, rd 4, rect, no-amp, aq-mode 2, tu-intra-depth 3, tu-inter-depth 3, max-merge 5, crf 20 => +15.5%
CTU 64, ref 5, subme 5, rd 4, rect, no-amp, aq-mode 2, tu-intra-depth 4, tu-inter-depth 4, max-merge 5, crf 18 => +36%

Temps are manageable on air cooling.

Can you share source and complete command line and build versions so this can be validated? I dont have a 9000-series, but I have plenty of systems with avx512 support and I have never seen anything close to this. Would be intresseting to see what the specific criteria is to get this increase.