← Home ← Back to /g/

Thread 106403437

27 posts 20 images /g/
Anonymous No.106403437 >>106403466 >>106403474 >>106403857 >>106404209 >>106404653 >>106406258 >>106407103
Use case for AVX512?
Anonymous No.106403451 >>106403470 >>106403946 >>106406258 >>106406477
Use case for ugly tranime shit?
Anonymous No.106403466
>>106403437 (OP)
Big-ass floating point vector instructions = slightly faster x265. And a hotter room in the winter.
Anonymous No.106403470
>>106403451
satania is the /g/ mascot
Anonymous No.106403474
>>106403437 (OP)
Winning bit wars arguments
Anonymous No.106403857
>>106403437 (OP)
Zen 4-5 are video encoding monsters, especially Zen 5 with the full 512 pipeline so it's apparently good for those types of tasks. What I understand benefits the most is computational heavy workloads like machine learning and simulation/data-analysis.
termux-termite !!1GSw688pHqQ No.106403946 >>106408875
>>106403451
I wasn't even going to respond to this thread since it's a sorry copy-paste but now I will just to spite a little faggot frognigger like (You).

OP, AVX-512 is practically speaking the only way to encode AV1 video with high compression efficiency quickly while using low power. For example the 8500G uses about 30 watts of power under libsvtav1 yet achieves about 100 FPS encoding M5 SVT-AV1 1080p video which achieved about 2 GB file sizes for most 2 hour 1080p movies and that's passing quality validation too btw (ie exceeds 95 VMAF).

Roughly this means a 2 hour 1080p movie will finish encoding in about 30 minutes so the total power used for this task is like 15Wh which is practically nothing.
t. pogeet !!b2oSUmilA2N No.106404209 >>106404371
>>106403437 (OP)
Don't mind me. I'm just waiting for Daiz to pop up
Anonymous No.106404371 >>106404435 >>106404602
>>106404209
termux-termite is SS2 confirmed and maybe Daiz
t. pogeet !!b2oSUmilA2N No.106404435 >>106404602
>>106404371
>termux-termite is SS2
ah, that's Daiz then.
Thanks for letting me know, anon saar.
termux-termite !!1GSw688pHqQ No.106404602 >>106404626 >>106405003
>>106404371
>>106404435
No, I'm for Jpeg XL actually.
t. pogeet !!b2oSUmilA2N No.106404626 >>106404753
>>106404602
yes.
You are for Jpeg XL and everything related to it. Especially for all the alternatives that exist for it.
Anonymous No.106404653
>>106403437 (OP)
FRYING YOUR CPU
termux-termite !!1GSw688pHqQ No.106404753 >>106404785
>>106404626
So this definition of a "daiz" now applies to BOTH avif and jpeg xl shills alike? The term has completely lost all meaning?
t. pogeet !!b2oSUmilA2N No.106404785
>>106404753
>The term has completely lost all meaning?
The term "Daiz" is not as simple as it seems, it seems.
Anonymous No.106405003 >>106405328
>>106404602
You corrected some of your previous chan persona retarded beliefs and gave it a new name, that's alright by me.

>I'm for Jpeg XL
I'm honestly not calling you Daiz or whatever but so is Daiz, every AVIF/WebPiss user (even AV1 enc. developers) seem to end up agreeing on this sooner or later.
termux-termite !!1GSw688pHqQ No.106405328 >>106408777
>>106405003
Or maybe I'm a different person and we're here because we all like to argue about similar things? God, we have the most dogshit tripfag drama here circlejerking about a banned? tripfag who liked to shill H264 even after VP9 was created and now we think he's multiple tripfags? If this is meant to discourage tripfagging then it's not working because there's just more of them now.

Anyway back to the topic and as punishment for the frogniggers, AVX-512 has shown up in a deinterlacing ffmpeg algorithm. Is that useful? Because these are some pretty huge gains. Like CPUs-without-avx-512-look-obsolete kind of gains...

https://www.phoronix.com/news/FFmpeg-Bwdif-AVX-512
Anonymous No.106406258 >>106407116
>>106403437 (OP)
9800X3D can process 32 32-bit float mults and another 32 add/sub in parallel with AVX512 code.
Double the numbers if you use 16 bit floats.
It can also do 32 32- bit FMAs, the up to 64 figure is when executing separate ADDSUB/MUL AVX512 instructions
It cannot be beaten, AVX512 is the future and ARM is a shit architecture.
7800x3d is only slightly worse, it has to goku vegeta fusion its 256bit units so it can only execute one 512bit fma or one mul & one add/sub at a time, still benchmarks indicate a slight boost over 256bit AVX
>>106403451
Eat mushrooms, it will teach you why "anime" artstyles attracts people so much(round shapes + colorful things)
Anonymous No.106406426 >>106406736
Anonymous No.106406477
>>106403451
Making retards seethe.
Anonymous No.106406736
>>106406426
Don't care, he revived english and french dubs(those french belgian women accent, gold for the ears) and he makes both antis and fanboys sneed, give him that at least
Only thing i'll agree on is censorship and piss poor woke "localization" somehow exclusive to english, fuck that
Anonymous No.106407103
>>106403437 (OP)
Native Instruments Massive X
Anonymous No.106407116
>>106406258
>at mushrooms, it will teach you why "anime" artstyles attracts people so much(round shapes + colorful things)
he's too retarded and obsessed with trannies, mushrooms won't help
Anonymous No.106408777
>>106405328
No, all these posts are ffmpeg posts end up being nonsense. They disable autovectorization by default and tend to be incredibly cherry-picked benchmarks. One of the previous times it was a news story, the filter was entirely different
Anonymous No.106408875 >>106408906
>>106403946
Is it not possible to video encode on gpu?
t. pogeet !!b2oSUmilA2N No.106408906
>>106408875
Even if it is, it won't be usable by anything other than the binary the GPU encoding implementation sample ships with, making it an existence that only servers to prove theoretical benchmarks and not actually aid with practicality development.
In the end, we have no choice but to write our own GPU encoder/decoder if we wish to actually leverage their potential and that is really really a lot of work.
Anonymous No.106408924
Probably for vector operation and shit.
Anything that's more intensive for SSE to handle should be offloaded to the GPU anyway.