[Blameless]

@foxx1337 Have you tried the newest version of OBS with the 21.6.1 drivers? I'm not able to get it to record at all, apparently won't even hook apps via game capture.

 

[foxx1337]

@foxx1337 Have you tried the newest version of OBS with the 21.6.1 drivers? I'm not able to get it to record at all, apparently won't even hook apps via game capture.

Seems to be working here. Here's a test I just made with Hunt: Showdown and Game Capture -

Cat jumped on me mid recording; I'm using OBS but ffmpeg (through OBS' builtin).

The encode string I used for hevc_amf: profile=1 profile_tier=1 preanalysis=true vbaq=true enforce_hrd=true quality=10 rc=0 qp_i=25 qp_p=22

Essential to not crash here with 20.5.2 at least, from ffmpeg:

$ ffmpeg.exe -h encoder=hevc_amf
...
-quality           <int>        E..V....... Set the encoding quality (from 0 to 10) (default speed)
     balanced        5            E..V.......
     speed           10           E..V.......
     quality         0            E..V.......
...

MediaInfo from the RAW before transcoding it for YouTube:

General
Unique ID                      : 84650668878697519963617906773517210663 (0x3FAF203477A887E9E4C3BCA6D8591427)
Complete name                  : E:\temp\2021-06-28 23-26-39.mkv
Format                         : Matroska
Format version                 : Version 4
File size                      : 1.25 GiB
Duration                       : 5 min 3 s
Overall bit rate               : 35.3 Mb/s
Writing application            : Lavf58.29.100
Writing library                : Lavf58.29.100
ErrorDetectionType             : Per level 1

Video
ID                             : 1
Format                         : HEVC
Format/Info                    : High Efficiency Video Coding
Format profile                 : Main@L6.2@High
Codec ID                       : V_MPEGH/ISO/HEVC
Duration                       : 5 min 3 s
Bit rate                       : 32.3 Mb/s
Width                          : 1 920 pixels
Height                         : 1 080 pixels
Display aspect ratio           : 16:9
Frame rate mode                : Constant
Frame rate                     : 60.000 FPS
Color space                    : YUV
Chroma subsampling             : 4:2:0
Bit depth                      : 8 bits
Bits/(Pixel*Frame)             : 0.260
Stream size                    : 1.14 GiB (91%)
Default                        : Yes
Forced                         : No
Color range                    : Full
colour_range_Original          : Limited
Color primaries                : BT.601 NTSC
Transfer characteristics       : BT.601
transfer_characteristics_Origi : BT.709
Matrix coefficients            : BT.601
matrix_coefficients_Original   : BT.709

Audio
ID                             : 2
Format                         : PCM
Format settings                : Little / Signed
Codec ID                       : A_PCM/INT/LIT
Duration                       : 5 min 3 s
Bit rate mode                  : Constant
Bit rate                       : 2 304 kb/s
Channel(s)                     : 2 channels
Sampling rate                  : 48.0 kHz
Bit depth                      : 24 bits
Stream size                    : 83.3 MiB (7%)
Default                        : Yes
Forced                         : No

As a reminder, in the previous driver, back when I was moaning, the fix for preventing the driver timeout was to slightly drop the encode quality in the native OBS encoder (as opposed to ffmpeg here) from "quality" to "balanced" or "speed".

 

[lawson67]

I have a Red Devil Powercolor RX 6800 XT that was running from new out the box in 3DMark TS or any games at about 100c plus on the hotspot, i took it apart and put some Dimond paste i had from years ago and it did nothing to the temps, i thought it was just a **** cooler badly designed even though it looks massive, then i got the balls up to strip it down again armed with some clear nail polish i bought of Amazon and some Thermal grizzly Conductonaut i watched about 3 videos on you tube of people doing LM watch how they applied it then stripped it down again, i really took my time to try and do a clean job, i didn't replace any of the original thermal pad as they looked fine to me, so after i put it back together i took it back to the PC and my heart rate was though the roof hoping i had not buggered it up, i turned on the PSU hit the power button and the powercolor lit up booted to windows and on the desktop HWINFO64 reported it was 26c , i then fired up 3D mark and did a full TS extreme ulta Stress test, normally that would be hitting about 101c - 102c but now max i got was 83c, to say i am happy is an understatement, i am so pleased i did it and very very happy with the results!

 

[apainlessa]

I have a Red Devil Powercolor RX 6800 XT that was running from new out the box in 3DMark TS or any games at about 100c plus on the hotspot, i took it apart and put some Dimond paste i had from years ago and it did nothing to the temps, i thought it was just a **** cooler badly designed even though it looks massive, then i got the balls up to strip it down again armed with some clear nail polish i bought of Amazon and some Thermal grizzly Conductonaut i watched about 3 videos on you tube of people doing LM watch how they applied it then stripped it down again, i really took my time to try and do a clean job, i didn't replace any of the original thermal pad as they looked fine to me, so after i put it back together i took it back to the PC and my heart rate was though the roof hoping i had not buggered it up, i turned on the PSU hit the power button and the powercolor lit up booted to windows and on the desktop HWINFO64 reported it was 26c , i then fired up 3D mark and did a full TS extreme ulta Stress test, normally that would be hitting about 101c - 102c but now max i got was 83c, to say i am happy is an understatement, i am so pleased i did it and very very happy with the results!

Hm... I'm tempted, I have the same card and I hit 95C Hotspot temp in winter

 

[gtz]

Hm... I'm tempted, I have the same card and I hit 95C Hotspot temp in winter

Is the red devil cooler really that bad? Is it because the fan curve is not aggressive? My XFX Merc cools extremely while, I have not even bothered installing the waterblock and putting it in my loop.

 

[apainlessa]

Is the red devil cooler really that bad? Is it because the fan curve is not aggressive? My XFX Merc cools extremely while, I have not even bothered installing the waterblock and putting it in my loop.

Dunno, with the fan-curve aggressively adjusted it sounds like a jet engine under heavy load. I'm not really bothered about the temp, it's not like it's throttling the speed at 95C. Card is way too expensive to do surgery on, maybe down the line.

 

[bloot]

Man, that CPU score though... crazy! I wish I could use MPT but anymore juice into my card just raises the junction temperature too much. What settings are you running in MPT?

Also, if you take off all the junk before and after the actual url it will display correctly; I edited your post to display correctly in my response.

Ah thanks, I was struggling to show the link correctly, I did not use MPT at all this time, it's at stock power limit values +15%.

Using MPT just adding 50W to max power limit got this score using 21.3.1 drivers

I scored 19 824 in Time Spy

I'm sure it would score better with the newest drivers, but it's too hot now here to try to push this stock cooler :S

 

[8800GT]

Ah thanks, I was struggling to show the link correctly, I did not use MPT at all this time, it's at stock power limit values +15%.

Using MPT just adding 50W to max power limit got this score using 21.3.1 drivers

I scored 19 824 in Time Spy

I'm sure it would score better with the newest drivers, but it's too hot now here to try to push this stock cooler :S

2710 average core frequency and 21,110 gpu score? That's mighty impressive my friend. I'd love to see what your card could do with max power when it's colder, on newest drivers!

 

[gtz]

Ah thanks, I was struggling to show the link correctly, I did not use MPT at all this time, it's at stock power limit values +15%.

Using MPT just adding 50W to max power limit got this score using 21.3.1 drivers

I scored 19 824 in Time Spy

I'm sure it would score better with the newest drivers, but it's too hot now here to try to push this stock cooler :S

Damn that is an impressive score.

 

[tolis626]

Hello peeps! I am really sorry if this has been talked about before, but I really don't have enough time right now to read the whole thread. I promise to do so in the future if I do get a 6800XT so that this is my last stupid question.

So, I may be able to get a decent deal on a Sapphire 6800XT Nitro+ SE. Somewhere around 1000€ (which is nuts, but this is the world we live in right now), but I'll also be able to sell my 1.5 year old 5700XT for like 500€. I'm thinking that if I wait until GPU prices come down and the 6800XT goes to its normal EU price, it'll still be like 800-900€ for the AIB cards and I won't be able to sell the 5700XT for more than 300€ then, so 500€ out of my pocket is probably about as good as it's gonna get. The responsible adult thing to do would be to sell the 5700XT right now, pocket the money and buy the 6800XT when it's available at a lower price, but I haven't been responsible for as long as I've been an adult when it comes to PCs, and my rig would be down until I got a new GPU (5900x here, so no iGPU) since I don't have a spare one.

The real question is this. Is the Sapphire card worth it? I know I've been extremely happy with my 5700XT Nitro+, but the 6800XT is a completely different beast. Reviews seem to be positive, but all of them seem to skip on actual overclocking, short of barely moving some sliders and calling it a day. What I'm more concerned about is the average OC headroom these cards have. The 5700XT Nitro+ would usually be a better than average card for OC, but I can't find any data points for the 6800XT and would like some input, if possible.

Pardon my possibly incomprehensible mess of a post, I'm writing it as fast as possible as the missus is already shouting because things need to be done around the house!

PS : Really hope my PSU can handle the thing if I do end up getting it.

 

[Exostenza]

I have a Red Devil Powercolor RX 6800 XT that was running from new out the box in 3DMark TS or any games at about 100c plus on the hotspot, i took it apart and put some Dimond paste i had from years ago and it did nothing to the temps, i thought it was just a **** cooler badly designed even though it looks massive, then i got the balls up to strip it down again armed with some clear nail polish i bought of Amazon and some Thermal grizzly Conductonaut i watched about 3 videos on you tube of people doing LM watch how they applied it then stripped it down again, i really took my time to try and do a clean job, i didn't replace any of the original thermal pad as they looked fine to me, so after i put it back together i took it back to the PC and my heart rate was though the roof hoping i had not buggered it up, i turned on the PSU hit the power button and the powercolor lit up booted to windows and on the desktop HWINFO64 reported it was 26c , i then fired up 3D mark and did a full TS extreme ulta Stress test, normally that would be hitting about 101c - 102c but now max i got was 83c, to say i am happy is an understatement, i am so pleased i did it and very very happy with the results!

EDIT: I see, so it is only aluminum that is corroded by liquid metal so I guess my GPU would be ok. Hey, I might do this today!

Deleted my original post after doing a quick google. Sorry.

 

[lawson67]

EDIT: I see, so it is only aluminum that is corroded by liquid metal so I guess my GPU would be ok. Hey, I might do this today!

Deleted my original post after doing a quick google. Sorry.

Let me know how you get on

 

[Blameless]

Seems to be working here.

You mentioned 21.5.2. I don't have any issues with OBS and 21.5.2 either, but I still can't get OBS to capture anything at all with the 2.6.x drivers.

 

[Exostenza]

Let me know how you get on

I was about to do it but then I realized that my GPU doesn't have a nice flat cold plate like they all should but direct heat pipes instead. This makes it so the contact with the die is not flat at all and would likely cause the liquid metal to not make good contact. I decided to hold off for now, unfortunately.

 

[foxx1337]

You mentioned 21.5.2. I don't have any issues with OBS and 21.5.2 either, but I still can't get OBS to capture anything at all with the 2.6.x drivers.

Those are settings I used because it otherwise crashed on 21.5.2. Keeping the settings, never keeping the obsolete driver.

 

[Blameless]

Those are settings I used because it otherwise crashed on 21.5.2. Keeping the settings, never keeping the obsolete driver.

Alright, something else must be wrong with my configuration then. Thanks for the clarification.

 

[Blameless]

@foxx1337

Figured it out. FFMpeg via OBS was working all along, but it was buffering so much data during longer recordings that I though it was freezing OBS...it could take several minutes to save the files I was recording after stopping it. However, there does seem to be an actual problem with the 21.6.x drivers with the default AMF OBS plugin, one that did not exist in prior versions.

Anyway, I've settled on the following FFMpeg settings for 1440p60:
profile_tier=high enforce_hrd=true quality=quality qp_i=22 qp_p=22

Interesting that the default plugin worked on my 6800XT in 21.5.2 and earlier and I can use the high quality mode without issue in FFMpeg. Perhaps your sample has a weak SoC?

Also, setting the rate control mode when specifying qp is redundant, as setting profile=1 (which is the only profile). I'm also 99% sure that preanalysis=true vbaq=true don't do anything with cqp rate control.

I'm sure it would score better with the newest drivers, but it's too hot now here to try to push this stock cooler :S

Might be a wash as newer drivers don't tend to OC as well as the 21.4.x and earlier branches.

I get the same GPU score at ~2.6GHz with 21.6.2, but that's about as far as is stable on this card with these drivers. With pre-21.5.x drivers I could do about 100MHz more in Time Spy at the same voltage.

 

[StrudelMan]

@foxx1337

Figured it out. FFMpeg via OBS was working all along, but it was buffering so much data during longer recordings that I though it was freezing OBS...it could take several minutes to save the files I was recording after stopping it. However, there does seem to be an actual problem with the 21.6.x drivers with the default AMF OBS plugin, one that did not exist in prior versions.

Also, setting the rate control mode when specifying qp is redundant, as setting profile=1 (which is the only profile). I'm also 99% sure that preanalysis=true vbaq=true don't do anything with cqp rate control.

The AMD encoder was abandoned because Jim (OBS developer) had personal conflicts with Xaymar (AMD encoder, StreamFX, FFMPEG encoder [now deprecated for StreamFX], and VoiceFX developer [also created SLOBS, before Streamlabs decided that scamming naive web developers was better, and created/maintains Own3d.tv plugin]) and also because AMD makes developing for their hardware a royal pain. AMD, generally being short for cash compared to NVIDIA and Intel, likes to use Microsoft (and also Sony) and "it's in the consoles so you have to code for it" to create developer support for their stuff. The downside of this approach is that HW accel for anything that isn't a Mantle derivative (d3d12, vulkan, GNM) doesn't exist.

The best option for using the AMD encoder is overclocking the core as much as you can (which will not only overclock the actual GPU, but also the infinity cache (which is why RDNA2 scales so well with core freq) and the video acceleration blocks) and using the StreamFX plugin to enable zero copy support. This should help improve performance of the encoder dramatically, so you actually will have enough power for more than ~1440p30. The quality settings usually decrease quality (from defaults), and there is exactly zero documentation on the settings from AMD, so defaults are recommended with the Preset appropriate for your usecase (the dropdown should say what is recommended for streaming and what is recommended for recording). To increase quality, your best bet is to switch to HEVC/H.265 (doesn't work for streaming because MPEG and Fraunhofer are greedy af) and crank the bitrate as high as your disks allow (50 mbit for acceptable quality, 500mbit for something you can compare to NVENC or QSV, only NV12 supported).

 

[Blameless]

The AMD encoder was abandoned

Whatever the history behind it, the included OBS plugin worked well enough until a few weeks ago, when AMD changed something in their drivers.

The best option for using the AMD encoder is overclocking the core as much as you can (which will not only overclock the actual GPU, but also the infinity cache (which is why RDNA2 scales so well with core freq) and the video acceleration blocks) and using the StreamFX plugin to enable zero copy support.

Are you sure about this?

I'm certain the video encoder is part of the SoC and seems to be influenced by SoC clock a lot more than core clock...as one of the few ways I've been able to induce encoder crashes is by undervolting or overlclocking the SoC too much.

The Infinity Cache may well be coupled to core clock somehow, but I'm also certain it has it's own clock domain, because jacking up GPU core clock does not significantly increase memory copy speeds in benchmarks that largely fit within the cache (AIDA64's GPGPU bench, for example). Other sources have also mentioned a separate cache clock domain, which would need to be the case if cards with disparate core clocks and highly variable boosts are all targeting the same cache bandwidth.

This should help improve performance of the encoder dramatically, so you actually will have enough power for more than ~1440p30.

Encoder performance has never been an issue for me on RDNA or RDNA2. On the later I've always been able to use the HEVC encoder via OBS+game capture to grab 4k60 @ CQP 20-24 (200-400Mbps in high motion scenes with a lot of detail) without any perceptible reduction in game performance. As of 21.6.x I have to use FFMpeg rather than the standard method, but the end result is the same.

The quality settings usually decrease quality (from defaults)

I've read that 'balanced' gives the best SSIM in bitrate constrained rate control settings, but in my testing 'quality' did narrowly produce the lowest bit rates at a given QP and I sure can't identify any perceptible difference at the QPs I'm using between presets.

Edit: Tested again, using a highly repeatable scene, and my previous observations seem sound. At 4k60, CQP 24, there is no subjective visual difference between recordings, but as expected the bitrate falls as quality goes up. "Balanced" is ~2% less bits than "speed" and "quality" is ~4% smaller still. Performance hit while recording was about 4% for "quality" (114 fps average without recording 109 fps while recording) and a bit over 3% for both "speed" and "balanced" (110 for both). GPU also consumed 1-2% less power while recording, indicating that the overhead of the capture was reducing load more than the capture and encode process was adding to it.

and there is exactly zero documentation on the settings from AMD

This has definitely been frustrating.

 

[ZealotKi11er]

Whatever the history behind it, the included OBS plugin worked well enough until a few weeks ago, when AMD changed something in their drivers

Could it be the latest update of AMF: Advanced Media Framework - GPUOpen