[Bug]: iGPU - AMD Encoding(s) "Quality" vs Intel

Describe the bug amd drivers !

To Reproduce Steps to reproduce the behavior:

install any amd drivers
Initialize the Encoder
Try to have some decent quality

Setup (please complete the following information):

OS: [Windows 10 iOT LTSC 21H2 x64]
Driver Version [23.19.02 or any other version are the same]
iGPU [2017 Vega 7 rebranded into 2021 Renoir]
Which component has the issue [Encoder, of course both hevc and avc]

Debug Log (please upload or paste):

try amd drivers !

Expected behavior When something is sold as an Video Encode Capable i expect that to have Quality, and not to encodings without QUALITY !

Screenshots exibit a: original source: a original source - has film grain and quality

exibit b: amd igpu "quality" hevc b vceenc_x64 v8 16 custom + ffmpeg v6 0 build 24-sept-2023 + amd amf v1 4 30

exibit c: intel igpu avc intel igpu avc LA-ICQ

exibit d: intel igpu hevc intel igpu hevc ICQ

Additional context Draw your own conclusions ! Is AMD selling broken products ?! Is AMD good to be sued for this, sure it is !

here is the original video sample.. please be good and share your results with any amd "quality". https://www.mediafire.com/file/16v8cny5ujanme3/bluray.sample.original.hevc-10bit-cpu.x265.mp4/file if your settings can make a better quality, like more crisp more detailed then share with us the discovery.

ps. intel and nvidia and apple are most welcome to compare with the amd "quality" ! thanks

and again please take a look bellow at my results: note: original shots where in WEBP format, but github is not accepting WEBP so i had to convert them into PNG !

a. original hevc-10bit-cpu.x265 a original hevc-10bit-cpu x265

b. intel-igpu.apollolake.qsv_avc-LA-ICQ b intel-igpu apollolake qsv_avc-LA-ICQ

c. intel-igpu.apollolake.qsv_hevc-ICQ c intel-igpu apollolake qsv_hevc-ICQ

d. amd-igpu.renoir.amf_avc-hqvbr d amd-igpu renoir amf_avc-hqvbr

e. amd-igpu.renoir.amf_hevc-hqvbr e amd-igpu renoir amf_hevc-hqvbr

f. amd-igpu.renoir.amf_hevc-hqcbr f amd-igpu renoir amf_hevc-hqcbr

original screen shoots in WEBP format (hmm github doesn't support WEB(P) ! intel.v.amd.zip

The tests where done on my laptop Lenovo with IPS screen 0 pwm (pulse width modulation, meaning no headaches no eyes balls hurt), zen 2 5500U with multi rebranded igpu renoir aka vega 7 cores units. total laptop power consumption 55 watt ! drivers at time of writing this, Driver Version 23.19.02-230831a-396094C-AMD-Software-Adrenalin-Edition AMD Windows Driver Version 31.0.21902.5 and the testings where done on: a. TranscodeHW.exe ( AMD AMF 1.3.40 ) b. VCEEncC64.exe ( VCEEncC 8.16 custom build with FFMpeg 6.0 snapshot 25-sep-2023 and AMD AMF 1.3.40 ) c. ffmpeg.exe ( FFMpeg 6.0 snapshot 25-sep-2023 with AMD AMF support ) OS is Windows 10 iOT LTSC 21H2 x64

Presets used: ALL ! but mostly HQ-CBR, HQ-VBR, CBR with AVC --preset slower and HEVC --preset slow + pre analize enabled and tons of other vpp enhancements or without and so on......................

Now on the other laptop i had to borrow it for some hours.

ACER 2016 Apollo Lake, N4200 intel pentium quad core with igpu HD Graphics 505. Software used latest intel drivers DriverVer=05/24/2023,31.0.101.2125 Windows 11 22H2 QSVEncC 7.48

Presets used: LA-ICQ ( Look Ahead - Intelligent Constant Quality) AVC ICQ ( Intelligent Constant Quality) HEVC

Remark is that on default on Intel 2016 system take a very good look at the preset LA-ICQ for AVC and compare the quality of the shot to the original sample. The result is even better than the original in details. Simply A M A Z I N G i might say for Intel superior quality !

On AMD i tried and tried and tried to desperation for something equal to an 2016 Apollo Lake, but nothing, zero quality on AMD. More even the desktop screen recordings are so bad that i just wanna donate my laptop for a freak show museum of technology !

here are the transcodings created on a 2016 intel laptop apollo lake n4200 with igpu hd graphics 505

LA-ICQ AVC https://www.mediafire.com/file/nv0chyzjwczf74c/a.+intel.apollolake.avc.LA-ICQ.mkv/file

ICQ HEVC https://www.mediafire.com/file/ose96tfhdpu3e8g/b.+intel.apollolake.hevc.ICQ.mkv/file

software used QSVEncC 7.48

use your whatever amd gpu or igpu and try to match the intel 2016 quality ! PS: The transcodings where made without any modification(s) or alteration(s), now imagine what more intel could have done with enhancements ! A M A Z I N G !

The bitrates were not set for comparison for each case. The bitrates vary and this would explain the corresponding quality differences. For example, in the two AMD/Intel HEVC cases, you cannot expect to have the same quality for both cases when one of the cases is given more than twice as many bits to encode the same information as the other case.

AMD HEVC: 2.3 Mbits/s Intel AVC: 11.5 Mbits/s Intel HEVC: 4.9 Mbits/s

Unsupervised answering.

Did you tested yourself with aprox bitrates ? In my tests those posted here are few from many, many, many more "bits/profiles.. ..a.s.o" Various enough to open a candy shop. I have even with hevc 25mbps made on amd that are completely useless against Intel arithmetic's.

So here is one of the so many tests on amd trying to fix the the unfixed Clipboard01 See the bitrate on that one, and remember that is one example of the sooo many tests way, way over 100. I can't post them all here.

Use the original sample from bluray and do check it by yourself on any amd igpu or gpu, fine set your bitrate to 150mbps and see if your 150mbps match the Intel LA-ICQ Avc or ICQ Hevc.

I am very curious to see your response, and screenshot and video too.

And here is the answer from the email that you send it to me about about ReLive desktop screen recorder clipboard_(26-Sep-23 _ 20_03_42) as can been seen i don't have relive since amd decided that it should not be used on "lower end devices" since amd want to capitalize expensive products...if i can be understandable enough saying..

now my laptop has removed also the video profiles, from efi bios by lenovo itself as lenovo blocked the profiles since they had on older drivers version 2021 the very best colors for multimedia videos. that's why i bought this laptop in first place. but like i said drivers from 2021..it's a loong time mate. so about video profiles i can't do nothing since is locked from efi bios. since again lenovo tweaked perfectly for this ips with 0 pwm screen.

clipboard_(26-Sep-23 _ 20_02_45) this is what i use from AMD AMF 1.3.40 the file DVR.exe custom source code patched by me to record from AVC to HEVC with profile Quality and 2600000 bits as around 2.9 mbps. It records also audio by default too.

Now it utilizes ONLY 30% of the igpu Video Codec 0 and the whole system task manager indicates ONLY 2 % ( two percent ) of Total CPU usage ( meaning all running windows,os,whatsoever) Amazing isn't it ?! And still AMD decided to remove the ReLive since "lower end systems can't record" what a joke.

Because of my custom experienced with DVR.exe i deleted,nuked for ever the OBS, since OBS utilizes way, way too much resources the CPU and 3D for the same video desktop recording and is no match against DVR.exe from AMD AMF. Again OBS is 200mbs in size it's big and worst where DVR.exe is only 670 kb (kilobytes) and i placed the file DVR.exe into c:\Program Files\AMD\CNext\CNext\ to be alongside the ffmpeg filters of AMD Software Center.

About player i use PotPlayer clipboard_(26-Sep-23 _ 20_33_54) the player is special configured by me to use only DX11. I play any HEVC 10-Bit 1080p BluRay remux FANLESS at only 11 WATT total laptop power consumption ( speakers consume 3 watt both ) The power consumption is measured via a special adapter via the wall, that indicates total devices power draw. Amazing isn't it !

You might want to update your Video driver that is way outdated and the encoder has received upgrades. but I'm curious why would you do these type of test with outdated drivers? Non written rule, before asking for help on something make sure you have everything UP TO DATE, as in most cases just by upgrading to the current ones fixes issues. Then if you still have the same problem, then you ask for help. And can you explain how is that a BUG after you update everything? I think is more human error but could be wrong.

I think is more human error but could be wrong.

Based on his screenshot he uses convolution, smooth and unsharp filters [--vpp-convolution3d, --vpp-smooth, and --vpp-unsharp] on re-encode and compaints that film grain is not being preserved. After he basically attempted to smoothen it 3 times. Moreover, he uses CQP [aka ICQ and LA-ICQ] for Intel, but CBR or VBR for AMD. Don't you think that it sounds kinda not same or analogous in terms of parameters?.

ICQ: intelligent constant quality algorithm Description: A better version of the CQP mode. Recording with a constant quality, similar to the CRF mode of x.264, that makes better usage of the bandwith than CQP mode.

And look at this user name. He specifically targets AMD and AMF based on all complaints i read. He doesn't even read answers from developers properly.

Ok, now to breakdown. Did everything myself here In general film grain is quite hard to preserve, and seems especially so in B/W videos. It is first thing to be smoothened out by compression, just to reduce output bitrate.

Anyways, did my own transcoding session. From his own source sample [source bitrate 6186 kbps]:

Normal 6200/18600 (limiter is 3x from average) [output bitrate 6096 kbps]: I mean, yeah, PSNR could've been better (i guess that is where all that lost film grain went), but if you able to see general quality loss at 97 (94 at 1 percentile) VMAF from 100, you have some fancy eyes with incredible zoom and attention. [You will notice only in direct comparison to source and only knowing where to look]

And VBR-HQ 6200/18600 (3x limiter) [output bitrate 6156 kbps]:

Sadly, general HQVBR benefits are not suitable for short samples, it allows to equalise quality in long run, not instantaneously. (it uses not actve scene --> be more conservative --> accumulate average bitrate buffer --> intensive scene --> spend buffer to increase bitrate for harder moment. approach)

As you can see, less bitrate AND less compression than you managed to get. Compression is reduced by quite a lot, actually.

I can get similar to his compression levels (still slightly better based on I-frames though, if i set my bitrate to 1800/5400 [Got 5.62 MB file] Now VMAF drops below 90 on average, and 83.8 in 1% percentile.

And yes, most significant result of the compressionis that it removed more and more film grain compared to other details. CQP 30 also eats up part of the film grain (not as much though), but resulting bitrate is 3777 kbps

Now... What about files with Intel made AVC LA-ICQ and HEVC ICQ he sent? AVC LA-ICQ one have 11.2 mbps bitrate, and HEVC ICQ one is a 4980 kbps. Oops. Doesn't seem like he compressed much from that original 6186 kbps (and AVC is having almost twice bitrate source have)... Anyways finished side by side comparison now:

Source vs AVC LA-ICQ

Source vs HEVC ICQ

Sooo. After this quick review i have 3 questions:

He also lost first almost 10 seconds from each file. How did he not notice that to compare video frametimes must be offset? Moreover, in screenshots he provided, he only lost 5 seconds, and not almost 10 (9s with something around 600 ms).
Where is same level of film grain as original had? Or same level he provided on screenshots? Had it REALLY not been compressed or ingrained in Intel's captures as he claims? [I can confirm that samples he provided were made with QSVEnc though]
What happened with color in his QSV encoded videos? (And how is this one not more important topic here). It looks more bleak in comparison!

I WISH i could compare them with VMAF score here. But as he lost 10 seconds i am unable to equalise files to source. Even if i am able to reproduce 10 second loss myself, source still have them, so i cannot compare outputs to source.

Hmm... Let me approximate his 4980 kbps HEVC output with CQP... Done, ended up with CQP 29 and 4856 kbps bitrate. Granted i used HEVC, and not HEVC10, which could've sligtly improved efficiency, but i compensated that with preanalysis (sadly you cannot use PA for 10 bit encodings for now) Oh, what is this... 96 VMAF with 92 1% percentile score? Doesn't look too bad. It means generally picture is barely changed

Now let's compare it with source and his result. Yeah, some grain is still smoothened, it is inevitable. But there is still grain

Now what about comparison to his HEVC sample? Hmm... Yeah, maybe on his one grain is SLIGHTLY more pronounced, but not that much. On other hand, colors for his are just different compared to original.

So... Autor gets stuck on topic that AMD quality is worse because film grain is being more compressed than other parts. But he also assumes that Intel provide you with more quality even if it completely wrecks color as long as slightly more grain was preserved? Moreover i couldn't even see same levels of film grain on provided by him videos compared to his screenshots (which are also in different color range compared to what i captured myself). [If anything, i may even guess that this color difference is EXACTLY why more grain was preserved. It is just more pronounced on bleaker video with lesser black and white].

I will even slightly help him to guess with what had happened with color! Not that he will have control over that easily (most likely).

But i cannot make colors look same even if i use 0-255 to 16-235 shader in MPC-HC. His video looks slightly more yellowy compared to simple full->limited compression

@DimkaTsv You do know that on gpu (dedicated/integrated) you can NEVER compare intel with amd, if you do then you are doing a very big mistake. Any pro coder use cpu encodings or special studio sets as in hardware, but on gpu (dedicated/integrated) intel are the best. Regardless about amd, is that no sane person will ever encode on anything that has amd gpu or igpu. However i do not agree with you @DimkaTsv and i might say that intel ICQ/LA_ICQ are by far superior to anything that amd had, have and will have for many, many years to come.

@amdvce i understand your frustration, 2 months ago i moved from always had intel cpu with integrated graphics and nvidia gpu to amd cpu zen 4 and amd gpu 7900xt. The greatest buyer remorse ever i had was when i tried from past december to encode video in 7900xt, in one word HORRIBLE. Even my ancient 1080 nvidia was way way better. So i saw that you have a lenovo notebook with "vega 7" and i would say that based from your sampled screenshots & video you kindly provide the quality compared to mine are the same, meaning HORRIBLE.

@DimkaTsv as i am member of several forums with specialized video encoding(s) one example doom forum, i can say that no one talks amd gpu or igpu encoding regarding quality but only fps, as new members prefer higher fps per encoding than quality since they don't understand mathematics. And again on here too you see that is silence as no sane person again will use amd gpu or igpu to encode video so that's why no one engage in here since heck the amd amf are for amateurs encoders. However some with very expensive amd gpu like my self would probably thinking how such a powerful gpu can do on video encoding too, well the answer for video quality for 7900xt are in the 1'st post.

Now @DimkaTsv i would assume that @amdvce is a programmer since he compiled both amd amf and vcenc with version 6.0 of ffmpeg and that was way ahead of time since none of them as both amd amf or vcenc where using version 5 of ffmpeg. Again almost every amd fans common remark is "AMD DRIVERS", since the zen hype people complain about one thing only "AMD DRIVERS" and here is the catch, why amd don't hire better developers, especially for amd amf department.

now comparing the only two brave and bold ones that lost, wasted their time like me too trying to squish,squash quality from doing encodings on amd gpu @DimkaTsv with @amdvce i will start in alphabetical order to not offend anyone

@amdvce seems to use only ffmpeg 6.0, as he compile both amd amf and vcenc with ffmpeg 6.0, the again uses potplayer based on ffmpeg 6.0 for video playback.

@amdvce you also got my attention with your little project dvr hevc desktop recording and i would very much like that if you are kindly enough to share it, please i use too hevc but after seeing your little project and compared with obs i don't use obs anymore and my relive from amd software is buggy and uses way more % than your's 30% at 1080p.

@DimkaTsv seems to use a mixture of ffmpeg with the worst ever video player such as mpc-be or mpc-hc are all based on lavfilters. any decent video encoder knows that you can't mix ffmpeg with lavfilters. ffmpeg are state-of-the-art in video, where lavfilters are based on ffmpeg but lavfilters are worse than ever. lavfilters lacks quality and style and they are using way too much energy as in wattage compared to original ffmpeg. anyway any pro user knows that the ffmpeg is pro for pro's and lavfilters are just for the masses. and again any pro video player user know that potplayer has no equal as again pro users use potplayer since potplayer is pure ffmpeg 6 enhanced where the masses use mpc-be or mpc-hc both based on low quality lavfilters and use brain fart madvr that is no energy friendly at all, ask [Greta Thunberg]. btw @DimkaTsv if you are russian that will explain a lot.

@rhutsAMD he can't even provide a quality test either, that say much about amd software, AMD DRIVERS or better amd lacks proper people to design drivers.
he most probably feel the new normal of 100º heat and self shut's down since knows that quality in video encoding(s) on amd gpu or igpu are as good as his programming skills. But it's ok every one know that intel pays way better for best software designers where amd can only afford the worst ones.

@DimkaTsv You do know that on gpu (dedicated/integrated) you can NEVER compare intel with amd, if you do then you are doing a very big mistake

He did exactly that, so i reversed this comparison back to him. Why am i, who is wrong in this case? [Also why comparison isn't allowed at all. We should aspire for improvement, don't we? Granted it also provides disappointment as ignorance is bliss sometimes.] Moreover, he wrapped it up as AMF github repo issue and basically blamed AMD drivers and AMF devs for differences in HW encoder produced results.

Don't get me wrong, AMF can be improved (and constructive requests are welcomed), but you also do understand that HW encoders are basically functions which are fixed in silicone. You cannot just change them. And older iterations will have even less leeway for improvement with time.

Any pro coder use cpu encodings or special studio sets as in hardware, but on gpu (dedicated/integrated) intel are the best

Really? Intel? Not Nvidia as you mentioned dGPU, and not iGPU only? Well, i guess it will depend on source and codec... I just refreshed information and did saw that Intel is actually pretty good at HEVC. I always thought that QSV was just slightly but behind. But afaik currently NVENC is superior to others if we count average use case. AVC is way worse for AMD, though, this one i will agree on without question. But HEVC and AV1 aren't bad at all. [Correction: just rechecked. Yes, AMF still slightly lags behind in aggressive compression case, i will give you that, but difference is not that drastic, imo. And if you don't pursue highest compression level you usually either have redundant bitrate or can spare slightly higher bitrate limit. I think 5-10% difference. Not much overall, but enough to compensate]

Anyways, topic starter points out HEVC in particular. Again! Do not use AVC encoder on VCN if you want quality. And quality will be pretty much fine [well, except one oopsie with AV1 having 1082 alignment instead of 1080, and lack of AV1 codec level of crop support. It makes VMAF comparison not easy at 1080p. But AMF devs are cooperating to solve it through ffmpeg]. Sure, you can do better, but it isn't that bad as people picture it to be. Also note that AMF doesn't currently support preanalysis with 10-bit encoding, meaning quality can potentially suffer here (on scene transition). But, again, question at topic was normal HEVC, not 10-bit profile.

However i do not agree with you @DimkaTsv and i might say that intel ICQ/LA_ICQ are by far superior to anything that amd had, have and will have for many, many years to come.

You may agree or not, but i don't say anything about it being good or bad. My point was that comparison is not quite properly made as used rate control methods are widely different. It does not reduce fact that ICQ/LA_ICQ can produce better result compared to simple VBR.

However some with very expensive amd gpu like my self would probably thinking how such a powerful gpu can do on video encoding too, well the answer for video quality for 7900xt are in the 1'st post.

Excuse me? His encodes are made on Vega 7 iGPU, mine are on RDNA3 dGPU. If anything i provided data that with same exact set bitrate RDNA3 does way less compression on average, at least from data i have. Meaning that either VCN improved noticeably between Vega 7 and RDNA3, or he is messing something up. Difference between QP 30/33.5 and QP 24/27 is very significant.

Now @DimkaTsv i would assume that @amdvce is a programmer since he compiled both amd amf and vcenc with version 6.0 of ffmpeg and that was way ahead of time since none of them as both amd amf or vcenc where using version 5 of ffmpeg.

Except VCEEnc doesn't require precompilation. Neither does QSVEnc he used for Intel. So it wasn't necessary. But you may be right that he compiled it himself. VCEEnc 8.16 release didn't come out at 25-th September... It was released 24-th June.

I did not recompile it myself, even though i probably also can.

now comparing the only two brave and bold ones that lost, wasted their time like me too trying to squish,squash quality from doing encodings on amd gpu @DimkaTsv with @amdvce

Ok, yes, i slightly did. As did he in his "best attempts", as he stated that he tried everything. (Just look at number of vpp filters he used). Even though most of command line parameters i used are for passing through metadata for HDR, chaptering and audio (i use drag and drop onto .bat file with generalised command. Currently even made command heavily customisable via variables)

Wanna see it with barebones settings (aka no preanalysis)?. Pretty sure it still will be much better compared to his quantisation levels simply due to RDNA3 having newer VCN.

He also uses older drivers. Meaning older runtime. Even though I don't quite think they are old enough to affect his GPU.

@DimkaTsv seems to use a mixture of ffmpeg with the worst ever video player such as mpc-be or mpc-hc are all based on lavfilters

Why should i care, if i just want to play video. Most people won't. Do you want to gate me off topic just because i use MPC-HC and not Pot Player? Don't you think it is dumb take that someone can't participate in discussion or argue at all just because he doesn't use specific player that you think everyone should use? If MPC-HC shows his video with incorrect color, but mine with correct ones, should it matter for me that it is because of LAV Filters? You encode video to play properly everywhere, and not selectively. Sure, fine, i can do PotPlayer comparison (i will also have few more things to test out), when i return back to home.

Btw don't you know what LAV filters are based on? Oh right... libavcodec and libavformat libraries. Aren't they part of ffmpeg? Oh, they are. Sure it is different product, but it uses same root. With that i mean that separating LAV filters from ffmpeg to such extend is simply wrong.

lavfilters lacks quality and style and they are using way too much energy as in wattage compared to original ffmpeg.

Proof please. I won't say anything about "quality" and "style" as they are depending in eye of the beholder. But power consumption part is questionable statement. By how much is it higher, even if it is? 2? 3 watts more? Maybe 5? Don't forget we are talking about dGPU in my case. I hadn't noticed abnormal power consumption with MPC-HC myself. Just normal HW decode consumption.

and use brain fart madvr that is no energy friendly at all, ask [Greta Thunberg]

I don't like MadVR (moreover, iirc it isn't supported properly anymore?), and use MPC-VR instead. Also, wtf is that tunberg argument?

btw @DimkaTsv if you are russian that will explain a lot

Excuse me? How is that coming into equation? So now, in addition to gating by used software (aka not compiling it myself with ffmpeg 6.0 and not using PotPlayer), you also try to gate me out by throwing everything onto my assumed nationality? WTF is this discussion turning into then?

For context of why i criticised him to that extend.

I can tell you that i personally caught him on making new "issue" containing only statement with double lie, when he tried to frame AMF devs for not supporting grain filter natively. So for now, my trust to his statemenents is always under "double check" everything mark. If you interested his double lie was:

When he explicitly ignored direct answer from AMF dev that natively adding grain is not supported, but they may consider adding such filter in future.Then forcing different answer and creating new topic when he claimed this answer as main without looking 2 comments (his second question and dev first answer) above.
When he tried to frame it as QSV and NVENC both support add grain filter natively. Spoiler: they not. Even ffmpeg doesn't have own filter, except for SVT-AV1 (which supports it by spec). To add grain you must use either libplacebo or custom written filter in command. And i tried to google that hard enough. Funniest part, is that proof for HW encoders not supporting add film-grain for Nvidia and Intel came from rigaya. Which is same dev who develops VCEEnc and QSVEnc he uses. (and for NVENC it also came from Nvidia SDK docs based on what i found).

I suggested him to proof that Nvidia and Intel support hw accelerated film-grain filter, but he didn't do it.

@amdvce, @DimkaTsv and @scotuss Yo all bother for nothing. As about @rhutsAMD, well don't worry as in the near future drivers engineers will be replaced by AI (artificial intelligence). It's common sense since AMD itself has using it since ZEN 3, as zen 3 has been co-designed with AI. So it's obviously that AI will take charge in the very near future of writing THE AMD DRIVERS too. Disney said that in the common years about 500 animators, 90% of the animation team will have to go since AI will replace them and from taking a full 4 to 5 years to do an animation will become weeks.

What's to know is that Qualcomm with their's Snapdragon Oryon X will eat both AMD and Intel breakfast and Apple desert too. Want the proof in AI, sure here check the video bellow. https://www.bilibili.com/video/BV1DJ4m1e7eo/?t=6

So yo all above i suggest to stop arguing and start from now planning saving(s) for the laptops with Qualcomm Oryon and Windows 11 24H2 for Arm64. At 40tops Qualcomm Oryon is besting the nvidia 3090Ti. And this is bad news also for nvidia. Ok so do have an idea about what and Qualcomm Oryon cpu can do in terms of processing, imagine this beast processing video encoding too.

And @amdvce thanks for pointing something out for me..such as AMD promoting something and fixing it after years of promoting. And now @amdvce just of curiosity how did exactly knew that AMD quote "in the near future will have money problems", it seems that Dr. Lisa SU just recently sold a lot of AMD shares, is it for RD department or money trouble?

@DimkaTsv buddy you should be thankful to @amdvce, as @amdvce exposed some quality enhancements that are non existing on AMD gpu's. More @amdvce also shows that AMD is deliberate cutting functions for their video encoding engine such as relive for laptops, as myself have a laptop with vega 8 and i can't use relive for uni or stream since AMD decided tha Vega 8 is old'ish and yeah AMD promotes RDNA 2 and above for RELIVE ! @DimkaTsv ain't that a lie please tell me as @amdvce clearly showed that his RELIVE can record 1080p the entire desktop with only 30% of Video Codec Engine of his Vega 7 and even more his RELIVE is recording in HEVC and such if you don't know HEVC is indeed far better than AVC all above at expense of only 2%, 3% of total cpu usage of his AMD 5500U. Well @DimkaTsv if you don't find this amazing then i don't know your meaning, because of @amdvce i clearly understand that my AMD 5825U with Vega 8 can do it too, but hey AMD decided that is better NOT to do it !

It's clearly a lie from AMD, as AMD say in specs something then after some days it deletes the specs or cripple them. @DimkaTsv since you sir like to google it do google it as it happen recently with AMD 8000G series as AMD re-editing the specs !

Also @DimkaTsv you sir are using AMD where Intel have B-Frames and their encodings are differently, that's why you sir think that some frames are cut but they aren't as on Intel you can choose B-Frames 2,3,4 whatever. And your yellow-ish might be from you Intel Control Center -> Video Colors. And final @DimkaTsv you sir can't use MPC-BE-HC whatever based on LAVFilters, You sir can't be seriously about that. Use POTPLAYER as @scotuss pointed out POTPLAYER is FFMPEG 6 ENHANCED ! And such there is no player as equal in quality and performance like POTPLAYER !

@DimkaTsv seriously you sir should see the bigger picture here and recheck @amdvce posts, you sir might even see that one of @amdvce post has been closed by @rhutsAMD with a lame excuse.

On Intel forums since mr. Patrick P. Gelsinger has took over the moderators email You for asking details about bugs and they do keep You posted and updated about their current status. Intel indeed has much changed and their drivers too have evolved so much.

On AMD You won't find this benevolence from it's developers or moderators. I can conclude that @scotuss pinned it with the remark "But it's ok every one know that intel pays way better for best software designers where amd can only afford the worst ones."

@DimkaTsv buddy you should be thankful to @amdvce, as @amdvce exposed some quality enhancements that are non existing on AMD gpu's.

Like? I don't know what is so special that he showed me, that i didn't know about VCEEnc except that QSV has ICQ mode. NVENC also doesn't have it, but is it because Nvidia cut their users from it? I would like you to pass opposite challenge. Find HQVBR or HQCBR mode on QSV for me. Spoiler - you may even not attempt to try, there are none for either QSV or NVENC. Or are talking you about vpp-filters? Then again, you missed because i use same exact VCEEnc as he does, so of course i know about them. And spoiler. There are more of them for VCEEnc compared to number available for QSVEnc.

So is it Intel now who cuts down functionality or AMD? See how easily i turned this argument inside out?

More @amdvce also shows that AMD is deliberate cutting functions for their video encoding engine such as relive for laptops, as myself have a laptop with vega 8 and i can't use relive for uni or stream since AMD decided tha Vega 8 is old'ish and yeah AMD promotes RDNA 2 and above for RELIVE !

Oh my... You clearly didn't study topic well before claiming.

ReLive developers are separate from AMF devs.
ReLive have requirements for 2GB of VRAM. And since iGPU don't have a dedicated VRAM, it doesn't show up for users with only iGPU. If you have problems report to it's developers, not here. Reporting here won't change anything, because these are wrong devs to do so.

So issue with this statement is NOT that AMF is bad, but rather that ReLive devs are morons. I have my own issues with ReLive (multiple of them for now at that), but i don't pour them down here because i know that they and AMF developers are two independent groups.

lso @DimkaTsv you sir are using AMD where Intel have B-Frames and their encodings are differently

You mean that same Intel that doesn't have P-frames for HEVC and uses B-frames instead? [Unless they fixed it in three years?] Sure. Don't get me wrong, Intel may be better, but it doesn't make your argument good.

you sir might even see that one of @amdvce post has been closed by @rhutsAMD with a lame excuse

One when he created a GITHUB ISSUE where he claimed that he initiated 10'000'000 USD lawsuit against AMD, which was properly closed by Rhuts for "not being an AMF related issue"? I saw that one. If you can tell me how that his issue was actually related to AMF development, to not be closed, go ahead.

If you say one part loud, why won't you say another one as well?

Use POTPLAYER as @scotuss pointed out POTPLAYER is FFMPEG 6 ENHANCED ! And such there is no player as equal in quality and performance like POTPLAYER !

Sounds like advertisement. When i last checked PotPlayer only uses DirectShow API and separately installed (but prompted) ffmpeg library is only used for decoding non-supported natively codecs (like HEVC, wow). Yup, only remainder of ffmpeg is that ffcodec library that is installed via Xiph's Open Codecs. [Which also develops VLC, but that is different story] Granted my data may be too old, but i hadn't saw actual mentions of ffmpeg in installer as archive.

Ah i see. Three new accounts created today just to answer me. Should've guessed from commenting style. Smells like multiaccount for me. Ok, now i will check twice. One of your mistakes is that you press on your beloved PotPlayer way too much. As if no other player exists. Thanks for still recommending it though. I won't necessarily say that i will use it daily (interface is good, control needs to be adjusted manually) , but it is good addition to list of players i have and can use for tests.

Quick review:

But overall window responsiveness (not UI itself, just window positioning and stuff) is not as great. It sticks window to places where i didn't tell it to do, and doesn't rememeber to restore alignment after i enable or disable sidebar. 

Btw already found few problems. 
If i load my 8k60 AV1 sample with HW acceleration, Pot Player it just shows me about 1/20-th of picture in upper-left corner until i alt-tab to something that covers all player and do it back to reload presentation layer in player itself. Even RTSS overlay is being cropped! Doesn't happen if i not use their "Built-in D3D11 Video Renderer"

Easy access to HW acceleration toggle is nice. And fact that you don't need to restart player is even better. But it all breaks if i don't Auto mode renderer selection or  their bulit-in D3D11 Video Renderer again. Because then i am FORCED to use SW decoder as toggle doesn't work in some videos.

And it also lacks "save video" functionality, which i can do with MPC-HC + yt-dlp combination to backup some interesting stuff i encounter. 
And it defaults to AVC 1280x640 resolution when opened 4k video, while MPC-HC yt-dlp defaults to VP9 2560x1440 level. Probably can set it to 4k somwhere even. 

It also either custom EVR presenter or build-in D3D11 renderer by default, which i found out after some testing

So it is not all rainbow for PotPlayer you suggested either. Nice player, but why would i spend shit load of time to tinker with it, if i just want to play damn movie?

Anyways, i still promised. And i do withold such promises. VBR 6400 set. Absolutely ZERO additional arguments except --preset slow which he used in his sample, so i am justified as well.

I see average QP of 25/27.4 vs his 30/33.5. It is 5/5.9 QP levels lower on average than his encoder did. That is A LOT of QP steps for difference to be seen.

[Small UPD:] Oops, forgot that it was 6200 and not 6400. Oh well i will repeat with VBR 6200 then...

Well, as you can see, there is not much difference. 25/27.4 turned into 25/27.55, which made QP level difference down to 5/5.75 from 5/5.9. Still way less QP than his screenshot showed.

CQP 29. No any additional "enhancing steps" (still used modular command though). Preanalysis doesn't work for CQP rate control anyways. Bitrate is about 5000 kbps. but as setting it to 30 would drop bitrate to whopping 3800 i wouldn't dare to do it easily.

And Pot Player visual comparison, as you wished for. Source vs AMF HEVC VBR 6400

Source vs AMF HEVC CQP:29

Source vs QSV HEVC ICQ (file provided by author)

Wait, is it what i saw in MPC-HC again, but this time in reverse? Why AMF one is still properly representing limited color range, while QSV one converted video to full? Wasn't it supposed to be other way around, because it is highly speculated Pot Player diff? It should not be like that, no-no-no. But oh well, he provided file as is, i hadn't touched it. Now you would try to blame me for doing something else wrong, right?

AMF HEVC CQP:29 vs QSV HEVC ICQ

Now. Going from CQP 29 to CQP 30 is very noticeable though. It smoothens out a lot more of a grain.

Source vs AMF HQVBR 4800

AMF HQVBR 4800 vs QSV HEVC ICQ

So my point stands. QSV mangled up color range. And even if PotPlayer reads it differently from MPC-HC, it still not in support of the QSV side. Also AMF HQVBR 4800 is not that far from QSV 4800 ICQ based on how grain looks. But i will give you that QSV sometimes does have slightly more pronounced grain in some specific places in the frame. Maybe due to having mangled color space, as it allows grain to be more visible? I am not encoder expert though. [But, as i mentioned above. Beauty is in an eye of a beholder. I encountered movie recently where film grain was implemented so strongly, that it actually diverted me from watching movie, because sky felt like it's moving due to grain intensity

Now back to multiaccount topic.

From GitHub Terms of Service:

One person or legal entity may maintain no more than one free Account (if you choose to control a machine account as well, that's fine, but it can only be used for running a machine).

I understand your frustration, 2 months ago i moved from always had intel cpu with integrated graphics and nvidia gpu to amd cpu zen 4 and amd gpu 7900xt. The greatest buyer remorse ever i had was when i tried from past december to encode video in 7900xt, in one word HORRIBLE. Even my ancient 1080 nvidia was way way better. So i saw that you have a lenovo notebook with "vega 7" and i would say that based from your sampled screenshots & video you kindly provide the quality compared to mine are the same, meaning HORRIBLE.

GPU encoders are meant for realtime encodings. You compensate any quality loss by using a high bitrate (100mbps) / CQP 23 HEVC / CQP 26 AV1

If post-processing is required, a cpu encoder with high quality / slow preset is used for the final archival output.

GPU encodes should never be used for archival purposes.

Most real professionals understand this.

GPUOpen-LibrariesAndSDKs / AMF

[Bug]: iGPU - AMD Encoding(s) "Quality" vs Intel #420