Download: Radeon Software Adrenalin 2020 Edition 20.4.2 drivers

let's say a lite version, i like so

Spoiler: Translate

Article Number N / A

Version-20.5.1 iCafe

Release Date-2020.05.11

Operating system-Windows 10 (64-bit)

Language-Chinese (Simplified), Chinese (Traditional), English

File size-374MB
The dedicated driver for AMD Internet cafes is customized and maintained by AMD ’s domestic R & D team, and it continuously performs various platform compatibility tests based on domestic popular online games, video applications, Internet cafes without disks, billing, game updates and other application and management software A large number of optimizations have been made for the domestic Internet cafe environment, and through continuous cooperation with game developers and Internet cafe management software developers, a compatible driver version customized for the majority of Internet cafe industry users. Compared with the Radeon Adrenalin software package, this version features a simplified driver installation package size, and eliminates some components that are not commonly used in Internet cafes without affecting performance and compatibility.

The driver supports the new RX5600 XT, RX5500 XT, RX5700 XT, RX5700, RX590 GME, RX560 XT, RX560D, RX560, RX570, RX580, RX580 2048SP, RX590, RX460, RX 470, RX 470D and RX 480, and supports Radeon HD 7000, R9 / R7 200, R9 / R7 300 series GPU, and Kaveri, Carrizo series desktop APU.

Certification:

The driver has been double-certified by the simulated Internet cafe laboratory jointly established by "AMD Internet Cafe Solution Lab" and domestic Internet cafe management software developers (Shun Internet®, Cloud Update®, Yiyou®, Xinhaoyi®)

New features

Radeon Chill (Game Intelligent Power Saving):
Innovative game optimization function, dynamically adjust the game frame rate. Radeon Chill can enhance the power consumption performance ratio and reduce the operating temperature of the graphics card. For details, please refer to the following website: AMD's official website Radeon Chill page.
Support DX12, Vulkan.
Already support "World of Warcraft", "Cross Fire", "Overwatch", "League of Legends", "StarCraft 2", "CS: GO" and other online, stand-alone popular games.
This function already supports RX 500 series graphics cards and RX400 series graphics cards
This version does not have Radeon Chill (Game Intelligent Power Saving) function. If you need this function, please download the Radeon Chill Special Edition
This version greatly improves the performance and stability of the game through FireWire.
This version optimizes performance, response time and stability for popular games.
This version disables HDCP encryption by default. Support the direct use of video capture card.
This version already supports the newly released RX5600 XT, RX5500 XT, RX5700 XT, RX5700, RX590 GME
This version optimizes the stability of the cloud update system after the driver installation is restarted.
This version fixes the driver installation problem for RX590GME.
This version optimizes the stability for League of Legends.
This version can optionally install AMD user experience program. Please read the installation instructions.
This version does not include the Radeon Settings installation package.
This version does not support Radeon HD 5000, HD 6000 series GPUs, and Trinity, Richland series desktop APUs.

Main supported video card models:

AMD Radeon ™ RX5700 XT / RX5700 / RX5600XT / RX5500XT

AMD Radeon ™ RX560 / RX560D / RX560 XT / RX 570 / RX580 2048SP / RX 580 / RX 590 / RX590 GME

AMD Radeon ™ RX 460 / RX 470 / RX 470D / RX 480

AMD Radeon ™ HD 7000 series graphics card models
HD7750 / HD7770 / HD7850 / HD7870 / HD7950 / HD7970 / HD7990

AMD Radeon ™ R9 / R7 200 series graphics card models
R7 240 / R7 250 / R7 250X / R7 260X / R9 270 / R9 270X / R9 280 / R9 280X

AMD Radeon ™ R9 / R7 300 series graphics card models
R5 340 / R7 350 / R7 360 / R9 370 1024 SP / R9 370X / R9 380 / R9 380X / R9 390 / R9 390X

AMD Radeon ™ series APU integrated graphics models
R5 series / R6 series / R7 series

if it changes number, something different?

meantime

Spoiler: 27.20.100.8236

 

Yeah, there is something different but compared to the "normal" branch they are behind with features. Even if it says 20.5.1 this driver is NOT the driver we are expecting to have.
This looks to me as if someone would have removed all languages except default (en) and chineese, removed some features and components, slapped an 20.5.1 CN label (or we could call it icafe or whatever) on a 20.4.2 driver.
Probably they've had also an 20.4.1 CN driver which was based on an earlier (20.3.x ?) branch too...

Nothing more....nothing less...

 

yes absolutely, is 20.5.1 iCafe
well so something is different and very important thing (for me), is lite and can use this driver with old panel setting
For now i am fine with this
waiting for real 20.5.x waiting for real WDDM2.7 waiting for real Windows 10 with support this feature or maybe waiting for real Update also in older version, it is not said that it is together... time will tell

the synchrony of the delay suggests this but it is not said
Windows 7 support (iCafe)
https://www.amd.com/zh-hans/support/kb/release-notes/icafe-windows-7-driver

EDIT:

Spoiler: Curiosity, a secret message?

楗摮睯⁳敒楧瑳祲䔠楤潴⁲敖獲潩⸵〰਍਍䡛䕋彙佌䅃彌䅍䡃义居潓瑦慷敲䅜䥔䅜䕃䅜獳浥汢敩嵳਍䌢䥌䄮灳捥⹴敄楶散䍌⹄片灡楨獣䐮獡扨慯摲㴢䌢䥌䌮浯楢敮⹤片灡楨獣䄮灳捥獴⸱慄桳潢牡≤਍਍䡛䕋彙佌䅃彌䅍䡃义居潓瑦慷敲䅜䥔䅜䕃䅜獳浥汢敩嵳਍䌢䥌䄮灳捥⹴敄楶散剃⹔片灡楨獣䐮獡扨慯摲㴢䌢䥌䌮浯楢敮⹤片灡楨獣䄮灳捥獴⸱慄桳潢牡≤਍਍䡛䕋彙佌䅃彌䅍䡃义居潓瑦慷敲䅜䥔䅜䕃䅜獳浥汢敩嵳਍䌢䥌䄮灳捥⹴畃瑳浯潆浲瑡敓敬瑣潩⹮片灡楨獣䐮獡扨慯摲㴢䌢䥌䌮浯楢敮⹤片灡楨獣䄮灳捥獴⸱慄桳潢牡≤਍਍䡛䕋彙佌䅃彌䅍䡃义居潓瑦慷敲䅜䥔䅜䕃䅜獳浥汢敩嵳਍䌢䥌䄮灳捥⹴楄灳慬獹潃潬牵⸲片灡楨獣䐮獡扨慯摲㴢䌢䥌䌮浯楢敮⹤片灡楨獣䄮灳捥獴⸱慄桳潢牡≤਍਍䡛䕋彙佌䅃彌䅍䡃义居潓瑦慷敲䅜䥔䅜䕃䅜獳浥汢敩嵳਍䌢䥌䄮灳捥⹴楄灳慬獹慍慮敧㉲䜮慲桰捩⹳慄桳潢牡≤∽䱃⹉潃扭湩摥䜮慲桰捩⹳獁数瑣ㅳ䐮獡扨慯摲ഢഊ嬊䭈奅䱟䍏䱁䵟䍁䥈䕎卜景睴牡履呁屉䍁居獁敳扭楬獥൝∊䱃⹉獁数瑣䐮癥捩䱥䑃䜮慲桰捩⹳畒瑮浩≥∽䱃⹉潃扭湩摥䜮慲桰捩⹳獁数瑣㉳刮湵楴敭ഢഊ嬊䭈奅䱟䍏䱁䵟䍁䥈䕎卜景睴牡履呁屉䍁居獁敳扭楬獥൝∊䱃⹉獁数瑣伮敶䑲楲敶⸵片灡楨獣刮湵楴敭㴢䌢䥌䌮浯楢敮⹤片灡楨獣䄮灳捥獴⸲畒瑮浩≥਍਍䡛䕋彙佌䅃彌䅍䡃义居潓瑦慷敲䅜䥔䅜䕃䅜獳浥汢敩嵳਍䌢䥌䄮灳捥⹴畍瑬噩啐⸲片灡楨獣刮湵楴敭㴢䌢䥌䌮浯楢敮⹤片灡楨獣䄮灳捥獴⸲畒瑮浩≥਍਍䡛䕋彙佌䅃彌䅍䡃义居潓瑦慷敲䅜䥔䅜䕃䅜獳浥汢敩嵳਍䌢䥌䄮灳捥⹴畍瑬噩啐⸴片灡楨獣刮湵楴敭㴢䌢䥌䌮浯楢敮⹤片灡楨獣䄮灳捥獴⸲畒瑮浩≥਍਍
TRANSLATE by Google:
You are the only one who can do this, you can get it, you can get it, you can get it, you can get it, you can get it, you can do it, you ca n’t do it, you ca n’t do it. ⹴ Wu Jie San 䍌 ⹄ piece Lan Zhen Veterinary 䐮 Xi Ren Shang mow 㴢 䌢 䥌 䌮 Seigo Wu Xia ⹤ piece Lan Zhen Veterinary 䄮 Huiwan mongoose ⸱ Ben Huang Li Mu ≤਍਍ 䡛 䕋 exchange wretched 䅃 Mi 䅍 䡃 home of righteousness Hui You can't do it, you can't do it, you can't do it, you can't do it, you can't do it, you can't do it, you can't do it at all. ⸱Lihu Lake ≤ ਍਍ 峛 䕋 汇 佃 亃 亃 亃 亃 亃 七 尃 knocking generously 兜 䕜 兜 戳 汢 敽 峳 爽 戳 刳piece Lan Zhen Veterinary 䐮 Xi Ren Shang mow 㴢 䌢 䥌 䌮 Seigo Wu Xia ⹤ piece Lan Zhen Veterinary 䄮 Huiwan mongoose ⸱ Ben Huang Li Mu ≤਍਍ 䡛 䕋 exchange wretched 䅃 Mi 䅍 䡃 home of righteousness Hui Wu Kang knock 䅜 䥔 䅜How to do it? How to do it? How to do it? Do you know how to do it? Lake ≤ ਍਍ 佛 䕋 汇 佌 䅃 Mi 免 䡃 The righteousness is generously knocked down 奜 䅔 䕜 䅜 獳 浥 汢 敩 嵳 ਍ 䌢 䥌 䄮How to do it? How to do it? How to do it? How to do it? Do you know how many things you want to do? You can't do it in this way, you can see how it works, if you don't know how to do it, if you want to know how to do it or not, it's not easy to figure out how to do it ∽ 戰 氹The problem is that the number of cases is very low, and the number of cases is very low. You can't get it right or wrong. You can't do it if you don't know how to do it. If you don't know how to do it, don't miss it. What's the difference? It's a piece of paper, a piece of paper, a piece of paper, a piece of cake, a piece of cake, a piece of cake, a piece of cake, a book, a book, a book, a book, a book, a book, a book, a book, a book, a book, a book, a book, a book, a book, a book, a book, a book, a book, a book, a book, a book, a book, a book, a book, a book, a book, a book. What are you doing? How are you doing this? It ’s not a good idea. You ca n’t do it. You ca n’t do it. You ca n’t do it.瑒 瑮 浩 ≥਍਍

 

How is performing for you? Is Wattman available? Is the only thing i mostly use for undervolting.

 

That driver is a Chinese malware sponsored by AMD. It isn't even signed for my windows 10 x64 pro 2004 build. Beware!

If you want something lightweight vs normal version, use the radeon pro variant of the driver. I use this without any issue.

 

yes, why no need setting panel by another driver pack because in this there is not, or you can symple use Memory tweak or OvedriveNTool if you only need this wattage

Spoiler: wattman

I installed it without problems and for is clean, if for you is a amd chinese malware....well i think a bit, with all the problems there are with amd drivers where they waste his time... in driver for iCafe
sometimes it is praised for espionage, is chinese and chinese, we lived with the american one until now in all side, there are those who are considered the richest man in the world thanks to this, windows (open the window and breathe), think about smartphones.... all rich with our data so when you go somewhere else the others say you are spied on
if they spy on it's okay, if others do it's not okay... a little fantastic story
I deactivated the log service (i had not seen it first) now i have removed it... simple
And again i repeat, in these drivers i have not found anything strange

Pro driver doesn't have my card yet... so amd style i arranged, this + pro panel setting

 

So wednesday I got a new monitor .. ASUS TUF VG249Q 144Hz Freesync

I am back to using display port again as my old screen had only HDMI

Anyway return of the black screens :/ first 1 in months

 

pasting from changelog (already translated): "this driver disables HDCP encryption[...]" !!!

so...lolz

btw, seriously, it's big deal....I used for a long while nV's iCafe driver (also cos it did support XP beyond its EOL - heh, at that time I had a spare secondary HDD runnin' XP for retrogaming ) on different windozes and good performances overall [waaay better than WHQL/Hotfixes]

I wonder if nV still releases iCafe cos I switched to Vulkan Beta branch some time ago (same ol' story: waaay better than WHQL/Hotfixes)

 

What is missing IMHO is /// let's build 2020 RIGHT and then the point
--finally nailed down all the black screen and HDR issues and FIXED THEM

 

Im sure they are working to stabilize them as much as they can.
They are not Gods.

 

I know, but..... VEGA Arch is for ... how long now ? Even Polaris has still some issues with 2020.... And RDNA1 is existing since 1,5 years already....
Tell me just HOW you can regress sooooo badly on a new branch... This is really hard for me to understand.... Plus they actually make some money now.... actually since over 2 or 3 years... basically since ZEN+ so with that I mean... they could really find some good guys/developers (who are also AMD andor open source "AMD" way fans ) who would gladly join them to help develop even better software.
I know their team was small.... and I have accepted that for a given time... just bit my teeth and said to myself -- calm down, you are supporting the good guys....

But current there is little to no excuse for this faux pas called 2020 branch...
...at least I am happy that Jonas and some other guys were able to hint me in the right direction to find a "dirty workaround" (cause you can't call it different) to stabilize my vega64 by locking to P1 state for GPU at 852/810 (MHz/mV that is). Which is really a dirty way cause of the thermals, power usage and the wear of the chip itself...

 

For Vega you do have the option to use older drivers whereas the 5700 and especially the 5600 kinda locks you into the more problematic late half 2019 and 2020 drivers.
It's mostly the randomness and uncertainty though there's possibly some variable of hardware problems as well for a number of Navi cards particularly memory stability but there's so much I don't know and there's also a lot of interesting developments for the Linux driver side like recent code commits allowing GFX10 (RDNA1/Navi) to have a soft reset same as GFX9 (GCN/Vega and earlier.) instead of doing a full driver reset.

Confirmed issues with the hardware playback particularly RDNA1 and VP9 plus potential issues with connectivity thus black screens and the whole mess with performance it's just a mess and very hard to keep track of because there's so much and a lot of it is mostly community guesswork.
(Witcher 2 as a example and hitting 10 - 15 FPS throw in DXVK and oh it's 60+ FPS, there's no way to get a 4 - 6x performance increase without something with the D3D drivers being seriously off even with how Navi downclocks and hits throttling for any small hiccup and wait period between the CPU and GPU.)


And then there's Radeon VII, Vega and also Polaris and earlier. Whatever the code base actually is here these should never have regressed at all and having outstanding issues going ignored for a year or more is pretty bad.

Way I'm looking at it unless Ampere forces AMD to to actually focus on the software side of development and the driver situation or RDNA2 sees some real improvement it's going to be tough not to go for a NVIDIA GPU as the next upgrade, RDNA2 being a major focus once Navi20 launches might see these regression just keep adding on and remaining unfixed too which would be pretty terrible.


There's a chance the branch jump and driver development improving but I'm not really seeing that any more, critical issues for stability and then performance should have landed within the first quarter with major focus on code and maintaining functionality on earlier hardware models.
We'll see how things are once 20.5.1 lands I suppose, would be nice to be wrong but even if the hardware with RDNA2 ends up being good the software is also important and if that's going to be a problem again or see ongoing issues set aside to focus on other priorities well there's not much to choose from for the next upgrade if AMD has really dropped things this badly on the GPU side of things.

But then there's this entire randomness too where there's a ton of cards where everything is fine from performance to general stability, would be fun to have a nice little lab and just do a mass of testing across multiple CPU's, boards and GPU's and drivers not that I'd ever have the resources for that but then many of these issues have filed and confirmed reports and can be replicated although the black screen issue continues to be a complete unknown for what's going on here and the performance is both highly API dependent and also more severe for some games or game engines even if these are using D3D9 or D3D11 some just perform far worse and figuring out the core cause here would be interesting.
(Could be CPU limited but what kind of hardware design is that where it drops to idle speeds waiting on updates on the CPU anyway so I don't think so plus it drops far more on Windows than it does on Linux and with wrappers or DXVK rather it continues to work at acceptable performance levels even if clock speeds do drop here too.)


Another lengthy post but I'd like to see things change but it's taking far too long and it's making it hard to seriously recommend AMD even if the hardware is good if the software is going to be so unpredictable plus RDNA2 / Navi20 is a whole set of new questions from how the launch situation will be like here to what's going to happen with existing hardware and ongoing issues.
(NVIDIA has their own share of problems too so that's also something to catch up more with and see how that develops post the 450.xx branch jump when that sees a public release in time for Win10 2004 and all that.)


And for Vega AMD also altered OverDrive / Wattman behavior two or three times so changing it up a bit and resolving the idle problems should have been fixed already but instead there's still a number of users who set the P1 state as min or just push the P0 state a bit higher to work around it.

Hah yeah 11+ months soon, definitively gotten a bit more cynical and negative for how the situation has been handled, liking the hardware still but can't recommend it although for those were it works it does work really well.
(The 50/50's kinda stupid after all it either works fine or you have a completely destabilized system what sort of odds are those.)



EDIT: Logic of it means AMD's struggling 11 months after RDNA1/Navi10 and there's still several issues to fix just for this let alone regressions for GCN Gen5 and earlier so adding in RDNA2 / Navi20 would add a huge amount of work.

Thus I see the launch for this as a very large question mark for what's going to happen although there's still the low chance the 20.5.1 or later drivers could really clean up these ongoing issues and turn things around better late than never but the amount of outstanding issues is a huge mass of work and code for almost every aspect of the driver and software suite.


RDNA2 having little to no issues might also push the theory about RDNA1 hardware flaws more but that's just guesses and some data from the more open Linux driver code on how some aspects of the code works.

Still doesn't do much for the other regressions either for current Vega and Polaris GPU owners and others.

So many questions and possibilities, dozens of major outstanding issues to fix within a five - six month period seems unlikely but we'll see.

Wattman and Radeon Settings open issues of all kinds, Polaris, Vega and Navi open issues of all kinds and general OS issues like the mouse stuttering problem, hardware like the GPU video issues and so on plus API problems and what do I know.



Negative pessimist or just cynical realist ha ha.
Well there's a chance but I'm not liking how it's looking so far but it could change.

 

And that's a lot of text again, suppose it's mostly just guesswork, personal opinions and reading up on the situation over almost a year now with still a lot of problems being there much as there's been improvements to as to not forget about those or anything.

Let's see what the next driver does and then well hopefully some of the bigger outstanding issues can start being closed that'd be a great start(?) for it I suppose.

 

As always --- totally agreeing with you Jonas. First I've had mixed feelings, but then (reading further) t was like... oh...yeah I can agree to that... oh! he is right! etc.
Just one comment on my side--- I was FORCED to do P1 cause P0 is not controllable by any means I know... And I've already tried afterburner, overdrive, soft power tables etc. The idle clocks (P0) seem to be controlled by the driver/BIOS only .... except I am wrong and you can correct me (and in that case I would be veeeeeeeeeeery glad to tweak only P0 to save on some electricity costs, thermals and of course gpu life)

 

Ah right P0 is probably grayed out as a minimum purely idle mode so it would be the first state and not the zero one that gets adjusted.

Wattman and the entire interface has changed a bit since I switched to the 5700 and then I also had a Vega 56 with Hynix VRAM and some differences in how those are configured when compared to the full Vega 64 both clocks and the voltage it gets.
Speaking of that I wonder if some of the stability issues are from stress to the VRAM although more so for the Navi black screen issues although AMD had some issues with the Vega64 as well so maybe the binning and overall variance is greater than thought and a bit extra is needed even for the stock models to retain stability.

Although that goes for the GPU core too I suppose hopefully mostly for overclock modes but then the boost behavior sometimes gets funny and exceeds what should be the max value which could be a problem if it doesn't have enough voltage.
Or it starts throttling or doesn't boost correctly at all, more theories and more work that needs to be done on the driver side of things.

Vega VII in particular had (has?) a lot of problems with that being something in-between what Vega and Navi used for it's power play mode with fewer states but less dynamic than what Navi's doing.
(Issues ranging from locked to mid-speed memory clocks to not boosting to boosting poorly when using manual clock values even when kept close to the stock settings as I recall.)


EDIT: Sigh.

"Yeah the GPU's are really good."
"There's just this little situation with the software." *Long list of outstanding issues.*

Ha ha, well that does need to improve now doesn't it.
Not that it hasn't there's been plenty of fixes since 19.7.1 too but it's just that there's still plenty of issues still being reported and many of the problems still remain or have only been fixed in part.

 

At least there's a chance the increase in revenue will funnel into the AMD GPU division as well and the additional staff and resources can start working out the driver even if it takes a year or so, would really be interesting what a large scale overhaul could do but it's a extreme task so it might not even be feasible even if the GPU team gets a significant increase in manpower but that alone could help speed up the bug fixes and get more organized bug reporting and feedback going and see better communication and also ideally testing and replicating issues improve.

That's a more long term change though, Corona isn't helping either far as having this being optimal and everything being properly staffed and working at full capacity which is also a factor until well whatever happens with that situation and also importantly when that starts changing and improving.

 

You have less things in Linux (for example your desktop might not even use OpenGL, there are some compositors that use OpenGL some that don't and in some cases you can make it use OpenGL or not based on your settings; you also have only OpenGL and Vulkan (only if you play games that support Vulkan else you won't see it used) to deal with in Linux case (sure there is gallium nine but you will only see it used if you use wine, else you won't see it)). That makes things a bit less complicated and easier to debug. There is also a high chance to see something totaly disable because they couldn't get it stable so they just decided to keep it disabled until they will figure out how to get it stable (if it's even possible).
You lack video hardware decode in Linux web browsers (in most case, there is a way to have it but it's really buggy to be honest, at least from my testing it had a passion to fill the vram, don't get me wrong but I find it abnormal to see the vram to 640+ MB usage when I only have 1 tab opened and an youtube video @720p running in that single tab and no other program opened...).
Sure there is video hardware decode in VLAN in linux but in some cases it proves to be buggy (i have a video that looks like total crap with hardware decode in linux, looks normal with software decode in linux and looks normal with hardware decode in windows; fun part is that the issue with hardware decode in linux is limited to that particular video, rest work fine with hardware decode).

Regarding the issues. It might easily be a software issue. I don't have any recent AMD GPU (and don't plan to buy one), but from my experience with my HD 7750 I can say for sure that some issues that I experience with win 8.1 I never experienced with win 7 (I dislike so much win 10 that I don't install it and with me playing games less and less a Linux distro looks more and more appealing, to be honest if I exclude games even ARM can do it for me) (for example I'm forced to use a screensaver in win 8.1 because if I set the display to turn off after x minutes sometimes it just stay black and never wake up (it actually loses the display in device manager for some reasons), yet I don't have this problem in win 7 and the only diference is the windows version, hardware is the same, even the gpu driver is the same ).
There is also a chance that some gpus come with a vbios that is basicaly a total mess (won't be a first to see a vbios that is a total mess and well the name of the manufacturer to be a known one and that vbios to be the original one). I put the vbios in the software side of things. That will make a particular gpu made from manufacturer A work fine while same model from manufacturer B cause issues.

Isn't the iGPU from Ryzen cpus also affected by some of the problems? (I don't have a Ryzen cpu so I can't test this is why I'm asking. Even if I want to buy one I will have major problems getting the right ram, all the online sellers in my country don't list the model of the ram they sell, all I can see is something like Crucial 4GB ddr4 but not the model so kinda impossible for me to know until I receive the ram if it's on the list with ram tested on that mb or not, and I read way too many cases when the ryzen cpu + mb just refused to boot with ddr4 that was not on the list of ram that mb was tested with...)

 

Amd has serious problems. I believe two and a half months over the last Whql isn't a good sign. I hope they can "fix" their demons. At 10 weeks approximately with no stable software. What the hell team, give us a certified wddm 2.7 already.

 

Yeah the situation is a bit different with the drivers and then Linux compared to Windows so it's not directly comparable I suppose but it's a interesting observation plus it's a good showing of what AMD's additions to the kernel and core driver does along with the open source driver code and utilities allowing for functionality like what's found in Wattman and the recent projects for D3D compatibility and translation layers what with Proton, DXVK and more.

Also been interesting to read up on how the GPU operates on Linux and compared to PC though it did take AMD a while for some of the DRM and core updates to land and there's still more to come and sites like Phoronix frequently lists many of the upcoming improvements and changes but it's looking up and things seem to be going well overall now.


And then there's the bios situation and overall hardware construction for the 5700 which is a interesting one as usual having AMD's robust PCB although the stock blower cooler hinders things somewhat and then you have some really well built models and some that were so bad the manufacturers made free revisions and replacements available but some of these models can still have problems with for example the fitting and cooler effectiveness as a result of the poor surface contact.

Wattage and the target figures for power draw and the fan curvature is all over the place (Fan defaults tend to be really low speeds and then quickly boosting and going back to low speed as temps drop.) with the stock model I think it was using a target of around 180w and then you have 195 to 220w with a few pushing into 240w and I believe the highest going by Tech Power Up bios data was at 280w one of those triple 8-pin models.

AMD put the SOC clocks at 1.2v but every custom model uses 1.05 instead and for amps anything between 12 to 20 with 14 as a common average so again a bit of a wide range.
(Pulse here that I am using has 14 on the normal bios and curiously a value of 15 on the silent bios though the wattage is lower down from 194 to 186 but this means very little overall.)

And then the situation with AMD's continued high transient power spikes (Fury? Maybe even earlier.) and Wattman for the power draw limit slider and settings which allows an extra 50% or if so desired it can be a decrease and thus far as power requirements are concerned you get a pretty wide range plus AMD's stock GPU voltage is also 1.2v and this has remained for a majority of custom bios models.


But then the drivers and Wattman might apply some changes to the values here, generally with the 5700 XT Pulse I end up at around 1157v for GPU core voltage when saving out the default config and checking the values for that is and since 19.7.2 AMD also locked down some of the fan settings which were then further changed in the 19.12.2 drivers and again in the 20.2.2 drivers making them react quicker to temperature changes but locking the idle behavior.

Undervolting and all that can be a bit tough to dial in, memory adjustments in particular can be very sensitive but cutting power draw in half, reducing junction temps by nearly 30 Celsius and maybe losing 3 - 5% performance if even that yeah it's pretty good.


Open source Linux driver changes and kernel commits for later Linux builds that has AMD patches I believe is also where some of the 5500 and 5600 silicon improvements and hardware fixes come in from though I don't know much about these and a lot of these things are also resolved through micro code and firmware updates as part of the display drivers themselves.



Anyways just from the amount of builds for the 5700 and now the situation with the 5600 and it's various bios updates for higher memory clocks there's already a lot of variance for trying to test stability and with how effective and how common suggestions for "Just undervolt it." or similar recommendations are it's also tough to say how many of these issues are driver related and how many are system specific and if there's a larger number of actually faulty GPU's out there or if it's elsewhere.

I thought the VRAM was problematic initially but then once the 19.8.x drivers were done with things stabilized a lot though a few games still had problems and the occasional desktop black screen or shutdown issues but after 20.2.2 most of these have stopped entirely.


Far as Navi goes the big thing is the frequent clock drops, last of these still ongoing black screen crashes would be nice to see resolved as well and then for some API's and the hardware video decoder there's already filed bug reports and AMD has acknowledged many of these but still not resolved them.

Hard to summarize this or get a good overview for what remains and what's possibly still problematic both for remaining confirmed issues and the many unconfirmed and hard to test and replicate problems especially for stability even with some of the more frequent games that do come up. (Popular online games but then these also occupy many of the top monthly player spots on Steam so of course they'd get a lot of posts when something does act up.)


For the Windows driver situation the various ups and downs and regressions particularly for generally very mature and stable GPU models like Polaris is also unfortunate as is the amount of time it's taken to get these resolved it really shouldn't have been almost a year.


Eh I'm bad at trying to do some summary or overview of the situation, there's a lot of variables involved and with multiple teams and then also Linux against Windows the drivers and behaviors across the two isn't strictly comparable either though it makes for some interesting observations and insights.

A ton of variables, potential sensitivity to things like memory and the PCI Express bus plus the display connectivity and far as software and the bios and values here are well they're all over the place.
Trying to do some testing is also difficult, low level API situation should be more demanding but whether it's native D3D12 or Vulkan or using a D3D9 or D3D11 -> Vulkan wrapper via DXVK so far it's stable and performing usually really well whereas native D3D9 or D3D11 can be pretty finicky but some titles also just work for hours without issue and there's little that can be immediately identified as a possible cause for as to what breaks and why.

It's been a interesting nine or so months since I upgraded, 20.2.2 is what I'd say really made a difference for the Windows drivers though performance and more still needs work.



So now what's left is a bit of a worry for what the situation will be like when RDNA2 comes out and how the branch jump from 19.50.x to 20.10.x and WDDM 2.7 / Win10 2004 will go.
Linux wise it sounds like the commits and improvements to 5.7 and 5.8 are going really well at least, efforts like Proton, Wine, Mesa and others are doing a incredible job too along with other projects plus AMD's own core driver code and kernel improvements.

Think that's another lengthy text dump but yeah there's a lot to this and testing and fixing it will be a process so not much to do but wait and see how the situation develops there's certainly more work to be done.