AMD Radeon RX 5700 XT Revisit: 3 Years Later

neeyik

Posts: 2,270   +2,743
Staff member
Which model do you have? Some of the lower-end models have an electrical connector that is insufficient for the power needs of the RX 5700 XT. My XFX RX 5700 XT Triple Dissipation crashed the entire system quite often. I sent it to XFX for RMA and they sent me back a THICC-III. The THICC-III never had that problem but I noticed an interesting difference between them. I believe that this is the difference that caused the system crashes because a driver crash can't cause a system reboot, only a hardware issue can do that, usually something to do with power distribution.

Now, if you look at the power connectors, you'll see that the Triple Dissipation uses an 8+6 plug configuration while the THICC-III uses an 8+8 plug configuration. I believe that the 8+6 is insufficient for the power needs of the Navi 10 GPU and causes the system to crash when the card tries to draw more power than the 8+6 can deliver. This causes the card to stop functioning and so the whole system crashes violently. OTOH, the THICC-III with its full 8+8 power connectors never had a problem. This is exactly how cards would react to a lack of needed power.

Another aspect of this situation that makes me think this is likely is the fact that reviewers NEVER experienced this problem. When an AIB sends a card to a reviewer, they ALWAYS send their top-line model. Steve's Powercolor Red Devil is a good example of this. The top-line models would ALL have the same 8+8 connector like the THICC-III, one of XFX's upper-tier models. As a result, those cards would never experience that same inability to deliver enough electricity to the GPU and so the reviewers would never experience those violent crashes. At the same time however, the general public tends to buy the lower-tier models because they're less expensive and deliver almost the same performance (like maybe 1-2% slower). Therefore, the general public would be experiencing these crashes all the time with reviewers having no idea why.

I think that Steve Walton himself would call my theory sound.
I think there's a little more going on than just a limitation of the power connectors, or even if that's part of the issue at all. The standard AMD RX 5700 XT uses an 8+6 system, which is rated to offer 225W (150+75) of power, precisely what the card's TDP is. The Triple Dissipation also uses the same setup, the same clocks, and the same TDP as the standard model. The Thicc III Ultra, as you pointed out, uses 8+8 (300W) but thanks to its higher clocks, it's naturally going to draw more power under peak loads - which one can see, using TechPowerUp's card power tests:

Standard RX 5700 XT = 227W (peak gaming)
XFX Thicc III Ultra = 288W (peak gaming)

Given that the Triple Dissipation has the same clocks and same power system as the standard one, surely all normal RX 5700 XT would be just as unstable as the card you had? And yet they're clearly not. Also, the GeForce RTX 2080 Super has a TDP of 250W, uses an 8+6 pin power connector setup, and its peak gaming draw was 275W (in TechPowerUp's testing). It's not a model renounced for having stability issues and yet its clearly drawing more power than the connection system is typically rated for.

Edit: I’d forgotten about the PCIe slot too, as it provides up to 75W too. So an 8+6 pin system can draw up to 300W in total. That’s more than enough for any RX 5700 XT. So why use an 8+8 system? Better to draw lots of current on well grounded, thick wire rails than slot pins.

And there's no obvious common link between the various models produced and their power management systems either - AMD used a 7 phase system controlled by an IR35217, the Thicc III Ultra had a 6+1 system and same controller, the Red Devil had 10 phase, the Asus TUF EVO had 7+1, the Sapphire Nitro+ SE used 8+1, and so on. Not sure what the Triple Dissipation uses, unfortunately, but it's clearly going to be one of the above!

In short, if it was just down to it using 8+6, there would be a whole host of graphics cards out there crashing systems, not just a specific RX 5700 model.
 

Steve

Posts: 2,931   +3,082
Staff member
Which model do you have? Some of the lower-end models have an electrical connector that is insufficient for the power needs of the RX 5700 XT. My XFX RX 5700 XT Triple Dissipation crashed the entire system quite often. I sent it to XFX for RMA and they sent me back a THICC-III. The THICC-III never had that problem but I noticed an interesting difference between them. I believe that this is the difference that caused the system crashes because a driver crash can't cause a system reboot, only a hardware issue can do that, usually something to do with power distribution.

I'll illustrate:

Here's the XFX RX 5700 XT Triple Dissipation:
7834-front.jpg

And here's the XFX RX 5700 XT THICC-III:
XFX-RX-5700-XT-THICC-III-ULTRA-1.jpg

Now, if you look at the power connectors, you'll see that the Triple Dissipation uses an 8+6 plug configuration while the THICC-III uses an 8+8 plug configuration. I believe that the 8+6 is insufficient for the power needs of the Navi 10 GPU and causes the system to crash when the card tries to draw more power than the 8+6 can deliver. This causes the card to stop functioning and so the whole system crashes violently. OTOH, the THICC-III with its full 8+8 power connectors never had a problem. This is exactly how cards would react to a lack of needed power.

Another aspect of this situation that makes me think this is likely is the fact that reviewers NEVER experienced this problem. When an AIB sends a card to a reviewer, they ALWAYS send their top-line model. Steve's Powercolor Red Devil is a good example of this. The top-line models would ALL have the same 8+8 connector like the THICC-III, one of XFX's upper-tier models. As a result, those cards would never experience that same inability to deliver enough electricity to the GPU and so the reviewers would never experience those violent crashes. At the same time however, the general public tends to buy the lower-tier models because they're less expensive and deliver almost the same performance (like maybe 1-2% slower). Therefore, the general public would be experiencing these crashes all the time with reviewers having no idea why.

I think that Steve Walton himself would call my theory sound.
I bought and tested every single 5700 XT model, never had a single issue.
 

Neatfeatguy

Posts: 1,025   +1,886
Thanks for the revisit Steven. The 5700 XT was definitely good value for money, often available for below msrp. The 6700XT seems like a regression here.

Being able to sell it for a great price due to its mining performance would have been a plus.


Just before the mining started to drop off, I had come across a 3060Ti for around $400. I gave it to my brother because he had been wanting to get away from the 5700XT (no issue with the card itself, but when he tried to update drivers he'd run into some issue with some game he liked to play so he was using old drivers from 2020).

He installed the 3060Ti and then he got to sell his 5700XT off for almost $1k - told him to take the money from the sale and put it towards their surrogate bills. The 5700XT ran games for him just fine (with the older driver he was using) and he was happy with the performance the card put out, but he was disappointed with the drivers for it.
 

poltevo

Posts: 45   +27
So what about FSR? Can that be ignored? If it gets to a point where every game is being tested with various upscaling methods how do you objectively select an apples to apples comparison? You obviously can't, it becomes highly subjective testing which is why what you're suggesting is wrong. Native is king.

Thanks for your response. The solution I think is to standardised image quality tests, and test for common artefacts such as ghosting, shimmering and issues with transparency. If an upscaling method gets so close to native quality that its practically indistinguishable, then it should be included in the benchmarks. I don't think this is highly subjective, but I do think it requires a good knowledge of how rendering/upscaling technologies work and what to look out for.

Consumers need to know FPS at particular quality setting regardless of the technologies being used. The only way to do that IMO is to include upscaling solutions when applicable. I understand that this is not a popular opinion, as it makes benchmarking more complicated, but I can't see a way round it.

Not including upscaling solutions is a subjective judgement in favour of native. A judgement which is probably true for lots of games where native *is* better. However, when the upscaling solution is very good and extremely close to native, then not including it seems arbitrary and to go against the spirit of a benchmark.

Is then your judgement that upscaling solutions don't get close enough to native? Or is it that standardised image quality tests are not feasible?
 

Steve

Posts: 2,931   +3,082
Staff member
Thanks for your response. The solution I think is to standardised image quality tests, and test for common artefacts such as ghosting, shimmering and issues with transparency. If an upscaling method gets so close to native quality that its practically indistinguishable, then it should be included in the benchmarks. I don't think this is highly subjective, but I do think it requires a good knowledge of how rendering/upscaling technologies work and what to look out for.

Consumers need to know FPS at particular quality setting regardless of the technologies being used. The only way to do that IMO is to include upscaling solutions when applicable. I understand that this is not a popular opinion, as it makes benchmarking more complicated, but I can't see a way round it.

Not including upscaling solutions is a subjective judgement in favour of native. A judgement which is probably true for lots of games where native *is* better. However, when the upscaling solution is very good and extremely close to native, then not including it seems arbitrary and to go against the spirit of a benchmark.

Is then your judgement that upscaling solutions don't get close enough to native? Or is it that standardised image quality tests are not feasible?
If you understand these technologies you know what you're suggesting isn't remotely possible. The quality of upscaling technologies changes from game to game, resolution to resolution and even the quality settings used to begin with. All that aside, how you do standardized something that's entirely subjective?
 

Avro Arrow

Posts: 3,104   +3,993
TechSpot Elite
I bought and tested every single 5700 XT model, never had a single issue.
Ok, so there's that theory out of the window! Holy hell Steve, you must have money to burn! :laughing:

Does the theory at least sound plausible based on what happened? I can't think of a purely software issue that can cause a hard reset but a power distribution issue would always have that result. Nevertheless, I readily admit that my level of expertise, while substantial, is easily eclipsed by your own.

Do you have any theories as to the cause?
 

Steve

Posts: 2,931   +3,082
Staff member
Ok, so there's that theory out of the window! Holy hell Steve, you must have money to burn! :laughing:

Does the theory at least sound plausible based on what happened? I can't think of a purely software issue that can cause a hard reset but a power distribution issue would always have that result. Nevertheless, I readily admit that my level of expertise, while substantial, is easily eclipsed by your own.

Do you have any theories as to the cause?
It's a work expense :) We spend a huge amount of money on PC hardware each year, it's not all 'free'.

As far as I could tell most of the issues reported online were from those with the cheaper models, but all the budget models we purchased were fine with the exception of one that needed a higher quality cable than the one that came with the monitor.
 

Avro Arrow

Posts: 3,104   +3,993
TechSpot Elite
It's a work expense :) We spend a huge amount of money on PC hardware each year, it's not all 'free'.
Oh I know that. I've seen your Harbour On Box [:D] video where you bought up a crap-tonne of B550 and X570 motherboards for testing. Then there was that Biostar video card that had a "cooler" that didn't actually cool (that one was so badly made that it blew my mind). That's why I made the joke about money to burn. I know that you don't just review free hardware because your impartiality is important to you and that's something I've always respected about you.

The first article of yours that I read was in 2008 and it prompted me to choose a Radeon HD 4870 over the GeForce GTX 260. I didn't recognise your name until over a decade later when I was looking for the article that I read about the HD 4870 and saw your name there. I was kinda surprised and I said to myself "So THAT'S why he's so good! He's been doing it long enough!" :laughing:
As far as I could tell most of the issues reported online were from those with the cheaper models, but all the budget models we purchased were fine with the exception of one that needed a higher quality cable than the one that came with the monitor.
I think perhaps that because of the backlash the AIB partners fixed the problem by the time you got your hands on them. Of course, by then, the damage to the card's reputation was already done. I still think that it had something to do with power delivery because it was causing a system reset.
 

Avro Arrow

Posts: 3,104   +3,993
TechSpot Elite
In short, if it was just down to it using 8+6, there would be a whole host of graphics cards out there crashing systems, not just a specific RX 5700 model.
That's not necessarily true. It may be that the first-released RDNA1 GPUs needed more than what 8+6 could provide when they peaked. There also might have been a manufacturing error. I think that it was power delivery because it caused a full system reset, something that I've never seen a simple driver error cause. Driver crashes that I've seen only crash to the desktop and possibly go into Windows basic resolution. A system reset (like what my card caused) was always a power delivery issue.
 
What sort of crashing are we talking about here and which model? It's also worth trying HDMI and seeing if that solves the issue. Ideally you should have palmed if off to a miner 6 months ago for top dollar :D
I have the Sapphire Pulse 5700XT. This is how the crash usually looks: application freezes for a second, then both my displays go black, then several seconds later I'm back at my desktop and whatever application I was running is still open but unresponsive and I have to force quit it. The AMD bug report popup automatically appears afterwards. It mostly happens in games, but I've had it happen while just watching a youtube video.

I'd rather not use HDMI if I can help it, because my main monitor (AOC G2460PF) maxes out at 120Hz on HDMI and 144Hz on DP.

Still getting driver crashes?
Maybe? I'm not sure what makes a crash a driver crash or not. Does what I've described above sound like a driver crash?

I fixed mine by undervolting to 1000mV and restricting the top GPU clock speed to 1815MHz. At first I used the stock Radeon software to do it but I noticed it would randomly reset things.
I did try undervolting and underclocking the GPU. I did also notice that AMD Radeon Software reset/lost this setting which is maybe why it didn't work. I didn't pursue this further. It seems silly to me to spend (at the time) top dollar for a GPU that would only work if you had to install 3rd party software to gut the performance.

Since you guys are curious, I'll expand on some of the things I tried to (unsuccessfully) solve my issues. Other things I tried (not in chronological order):
  • Two power cables from the PSU to the GPU (instead of the the one cable with two 6+2 connectors).
  • Only one monitor plugged in.
  • Running it in a different PC.
  • DDU driver uninstall/reinstall.
  • Every setting I could find in Radeon Software turned off (Radeon Anti-lag, Chill, Boost, Image Sharpening, Enhanced Sync, FreeSync, etc).

I've got it plugged in with a different DP cable I had laying around, so we'll see how that goes. The crash has been happening a bit more than once a month in 2021 and 2022, so I guess I'll report back here in a month to see if the issue stays away 🤷‍♂️
 

Steve

Posts: 2,931   +3,082
Staff member
I have the Sapphire Pulse 5700XT. This is how the crash usually looks: application freezes for a second, then both my displays go black, then several seconds later I'm back at my desktop and whatever application I was running is still open but unresponsive and I have to force quit it. The AMD bug report popup automatically appears afterwards. It mostly happens in games, but I've had it happen while just watching a youtube video.

I'd rather not use HDMI if I can help it, because my main monitor (AOC G2460PF) maxes out at 120Hz on HDMI and 144Hz on DP.


Maybe? I'm not sure what makes a crash a driver crash or not. Does what I've described above sound like a driver crash?


I did try undervolting and underclocking the GPU. I did also notice that AMD Radeon Software reset/lost this setting which is maybe why it didn't work. I didn't pursue this further. It seems silly to me to spend (at the time) top dollar for a GPU that would only work if you had to install 3rd party software to gut the performance.

Since you guys are curious, I'll expand on some of the things I tried to (unsuccessfully) solve my issues. Other things I tried (not in chronological order):
  • Two power cables from the PSU to the GPU (instead of the the one cable with two 6+2 connectors).
  • Only one monitor plugged in.
  • Running it in a different PC.
  • DDU driver uninstall/reinstall.
  • Every setting I could find in Radeon Software turned off (Radeon Anti-lag, Chill, Boost, Image Sharpening, Enhanced Sync, FreeSync, etc).

I've got it plugged in with a different DP cable I had laying around, so we'll see how that goes. The crash has been happening a bit more than once a month in 2021 and 2022, so I guess I'll report back here in a month to see if the issue stays away 🤷‍♂️
I get you don't want to use HDMI, but it's a troubleshooting thing.
 

Neatfeatguy

Posts: 1,025   +1,886
Maybe? I'm not sure what makes a crash a driver crash or not. Does what I've described above sound like a driver crash?

Check the event viewer, see what error it posts.

Look under the Windows Logs > System - should see system related crashes here for drivers
Also look under Windows Logs > Applications - should see any application/game/program that crashes

Hopefully you can see what is crashing....not that you'd really be able to fix the crashes themselves, but at least you'd know what exactly is crashing.
 
Check the event viewer, see what error it posts.

Look under the Windows Logs > System - should see system related crashes here for drivers
Also look under Windows Logs > Applications - should see any application/game/program that crashes

Hopefully you can see what is crashing....not that you'd really be able to fix the crashes themselves, but at least you'd know what exactly is crashing.
I tried running on a different DP cable for a while. When it crashed this time I looked at the Windows Logs > System logs that you described. I see this: "Display driver amdwddmg stopped responding and has successfully recovered." two times within 10 seconds of each other. However, from a user perspective the thing definitely did NOT successfully recover. This time the crash was more permanent. After my monitors went black and green and the game I was playing crashed (sound stopped and I see in Windows Logs > Application that my game crashed 3 seconds after the second amdwddmg crash), my GPU fans started spinning super fast and I had to hold the power button to get anything working again.

I guess I'll try Steve's suggestion of an HDMI cable and see if that makes a difference.
 

Neatfeatguy

Posts: 1,025   +1,886
I tried running on a different DP cable for a while. When it crashed this time I looked at the Windows Logs > System logs that you described. I see this: "Display driver amdwddmg stopped responding and has successfully recovered." two times within 10 seconds of each other. However, from a user perspective the thing definitely did NOT successfully recover. This time the crash was more permanent. After my monitors went black and green and the game I was playing crashed (sound stopped and I see in Windows Logs > Application that my game crashed 3 seconds after the second amdwddmg crash), my GPU fans started spinning super fast and I had to hold the power button to get anything working again.

I guess I'll try Steve's suggestion of an HDMI cable and see if that makes a difference.

Driver is crashing - for whatever reason it may be.
Could be software related (conflicting issues with the games and driver you're using)
Could be hardware related (not enough voltage to the card or maybe clock speeds are too high)

I know my younger brother ran a 5700XT up until about 8 months ago when I gifted him a 3060Ti. I don't know if he ever updated the drivers he was using with the 5700XT, but I do know when he got it (was in January 2020) he tried the most recent driver and he kept getting odd crashes in The Division 2.

He came from using a 780Ti that ran things just fine until they didn't (VRAM failure). He choose to grab a 5700XT, best new card in his price range.

He installed the latest AMD driver at the time, if I remember correctly, was version 20.x.x that came out at the time. The Division 2 would crash after it launched. I had him find the driver before that, was an earlier 20.x.x version. The game launched with it, but would crash after a minute or two. I had him go back to the last 19.x.x version and his system ran games just fine, no random black screens (that had been reported with the 5700XT models) and games weren't crashing. He stayed on that driver for as long as I can remember and used the card for almost 2 years.

Long story short, if trying through HDMI doesn't resolve your issue, go back to older drivers and see if that makes a difference.

If drivers don't fix the issue, see if you can bring the clock speeds down 100MHz.
If clock speed adjustment doesn't help, you could see if you could slightly bump up the voltage while keeping the clock speeds down.

I'm not saying that my story below is the same as your issue, but I have had a GPU constantly crash to desktop while gaming:

Years ago I sent a GTX 280 back to BFG for a warranty fix. They sent back a GTX 285. BFG then announced they were shutting down and no longer taking RMAs. I now had a GTX 285 in place of the 280.

I was running 280s in SLI, but I could get by with a slightly better single card so I put the 285 in. Re-installed drivers and loaded a game - crash.
I tried most recent driver - crash
I tried a multitude of older drivers - all crash to desktop, driver fail and recover.
I lowered clock speeds - crash
I raised voltage - crash
I adjusted down clocks more & pushed voltage a bit more - crash

I screwed around with the card for a few days, but everything resulted in a crash. I had a faulty GPU. BFG closed doors and I couldn't send it back. At least I still had my 280 that worked.

It bugged me for a week or so, having a faulty 285. So I did the only thing left I could think of, I flashed the BIOS from the working 280 to my faulty 285 and it worked. The 285 was fully functional on the 280 BIOS. I ran the cards in SLI for almost another 2 years.

Story is, BIOS flash (if there are other compatible BIOSes for your card) could be your very last (emphasis on the words VERY LAST) ditch effort to fix your issue.
 

Avro Arrow

Posts: 3,104   +3,993
TechSpot Elite
I tried running on a different DP cable for a while. When it crashed this time I looked at the Windows Logs > System logs that you described. I see this: "Display driver amdwddmg stopped responding and has successfully recovered." two times within 10 seconds of each other. However, from a user perspective the thing definitely did NOT successfully recover. This time the crash was more permanent. After my monitors went black and green and the game I was playing crashed (sound stopped and I see in Windows Logs > Application that my game crashed 3 seconds after the second amdwddmg crash), my GPU fans started spinning super fast and I had to hold the power button to get anything working again.

I guess I'll try Steve's suggestion of an HDMI cable and see if that makes a difference.
I remember that mine would crash violently, turning my screen green and outputting a loud buzzing through my speakers. Only pressing the reset button could get me out of it which told me that it was a hardware crash (I still suspect that it was something to do with power delivery). Then I tried changing the drivers and all that did was cause the screen to go pink when it crashed instead of green. :laughing:

It definitely wasn't the drivers though because when I got fed up, I pulled the card out, swapped in one of my R9 Furies and the problems ceased immediately. I didn't even change the driver installation, it just worked. That tells me that the card was the problem because when I got a THICC III from XFX as the replacement for my Triple-Dissipation card, I never had a problem again. Weird, eh?

I personally think that you should RMA the card. Sapphire is legendary for taking good care of its customers.
 
In using a Radeon 5700 flashed with xt firmware with a 5950x ( waiting for 7xxx cards ) FSR/RSR is amazing when you figure out how to set it up.

Agreed native resolution is better.. but till I can get a better gpu ( might settle for water cooled 6950xt ) in still able to play 1440P above 80+ on all titles.

Warzone I can get 120+ with RSR active. 60-70 without.
 
I still have my Nitro+ 5700 XT as my primary GPU, still going strong and is a good performer at 1440p. I don't ever sell or trade up any of my PC stuff even though I probably should have in this case.
 

Avro Arrow

Posts: 3,104   +3,993
TechSpot Elite
I still have my Nitro+ 5700 XT as my primary GPU, still going strong and is a good performer at 1440p. I don't ever sell or trade up any of my PC stuff even though I probably should have in this case.
You still would've gotten screwed as you tried to replace it if you weren't willing to go 6 months without a GPU.
 

Neatfeatguy

Posts: 1,025   +1,886
You still would've gotten screwed as you tried to replace it if you weren't willing to go 6 months without a GPU.

Or at the height of the crypto mining, he could have spent around $1000 on a GPU (might have landed a 6700XT at that price if he went AMD because the 6800s were going around $1200 and the XT models were upwards of $1500) and that would have netted him around a 25-30% performance gain. Or possibly a 3070 (they were going for around $950).

The he could have sold his 5700XT for a grand, that's what they were going for on Ebay for a while. Walked away even in terms of cost of getting a new GPU and gaining 30% performance.

But, you'd have to be able to find a GPU during that time and have the money to spend on it. If you were lucky enough to do this, you could turn around and sell that 5700XT in a matter of days (if not hours). But, I know that not everything is able to be in that position to simply spend $1000 and wait to recoup the money later.

I gifted my brother a 3060Ti and he sold his 5700XT a few months before crypto mining craze ended and got $900 for it.
 

Avro Arrow

Posts: 3,104   +3,993
TechSpot Elite
Or at the height of the crypto mining, he could have spent around $1000 on a GPU (might have landed a 6700XT at that price if he went AMD because the 6800s were going around $1200 and the XT models were upwards of $1500) and that would have netted him around a 25-30% performance gain. Or possibly a 3070 (they were going for around $950).

The he could have sold his 5700XT for a grand, that's what they were going for on Ebay for a while. Walked away even in terms of cost of getting a new GPU and gaining 30% performance.

But, you'd have to be able to find a GPU during that time and have the money to spend on it. If you were lucky enough to do this, you could turn around and sell that 5700XT in a matter of days (if not hours). But, I know that not everything is able to be in that position to simply spend $1000 and wait to recoup the money later.

I gifted my brother a 3060Ti and he sold his 5700XT a few months before crypto mining craze ended and got $900 for it.
Yep, and just imagine what he would've received for the 3060 Ti...
 
Or at the height of the crypto mining, he could have spent around $1000 on a GPU (might have landed a 6700XT at that price if he went AMD because the 6800s were going around $1200 and the XT models were upwards of $1500) and that would have netted him around a 25-30% performance gain. Or possibly a 3070 (they were going for around $950).

The he could have sold his 5700XT for a grand, that's what they were going for on Ebay for a while. Walked away even in terms of cost of getting a new GPU and gaining 30% performance.

But, you'd have to be able to find a GPU during that time and have the money to spend on it. If you were lucky enough to do this, you could turn around and sell that 5700XT in a matter of days (if not hours). But, I know that not everything is able to be in that position to simply spend $1000 and wait to recoup the money later.

I gifted my brother a 3060Ti and he sold his 5700XT a few months before crypto mining craze ended and got $900 for it.
I still have a functional Pentium 4 ATI 9550 rig among other older hardware. Not really looking at PC parts to make money.