Can anyone else replicate this crash on the 9070?

[Dr. Dro]

@Dr. Dro @r.h.p

1. Tested the -30% power limit, GPU power peaked at 170W, no crashes for a while but eventually game/adrenaline crash. This is a good sign because the GPU didn't crash, just the software.
2. I set the power limit to -20% to fix the instability. Worked like a charm, played at maximum settings, Ultra + Path-Tracing, for 15 minutes fighting in heavy combat against the police. GPU never went above 197W.
View attachment 391001
3. I returned everything to default, GPU instantly started consuming 240W and I crashed after 10 seconds. This time it damn-near bricked my system...One display stopped working so I reset my PC. But then neither display was working. Tried the motherboard display since the 7600X has a integrated graphics, still not working. Had to clear CMOS and it booted up again.


PSU issue it seems but the 12v cable is actually doing just fine. I opened up the backplate last night after (it's magnetic) and I see no sign of damage or smell anything burning.
View attachment 391002


I'm going to order a better PSU and keep my GPU power-limited until that arrives. Didn't realise I was playing this close to fire (literally).

Yup, called it. It's the power supply that's running at full and even a bit beyond capacity if you look at the current output capability - past a certain point the delivery won't be reliable and that's when your card croaks.

Set aside the money for a decent 750 to 850W PSU, I'd look at the Seasonic Focus Gold or the Corsair RMe, not too pricy, good warranty, decent efficiency, will run great. Remember to get an ATX 3.1 model that has native 2x6 cable for your graphics card, so you won't need to use the adapter.

 

[Prognatodon]

12v cable

I kind of wonder how the other end of it is wired in your case. With A550BN, one unpopulated 8-pin receptacle would already be kinda sketchy, even if the two connectors you have weren't daisy-chained.

 

[NSR]

Yup, called it. It's the power supply that's running at full and even a bit beyond capacity if you look at the current output capability - past a certain point the delivery won't be reliable and that's when your card croaks.

I just hope my GPU hasn't been taking damage each crash, lol. It's not even been 1 week since I got it. I suppose if the problem is a lack of power as opposed to too much then components should be safe?

Set aside the money for a decent 750 to 850W PSU, I'd look at the Seasonic Focus Gold or the Corsair RMe, not too pricy, good warranty, decent efficiency, will run great. Remember to get an ATX 3.1 model that has native 2x6 cable for your graphics card, so you won't need to use the adapter.

Alright, I'll go looking! Will be a short few weeks before I can purchase but I'll keep things under light load until then. Appreciate the help from everyone.

I kind of wonder how the other end of it is wired in your case. With A550BN, one unpopulated 8-pin receptacle would already be kinda sketchy, even if the two connectors you have weren't daisy-chained.

One of the PSU 8-pins is daisy-chained. I've had this set up for years since my 6800 required x3 8-pins. I just connected them to the adapter for the 9070. The 12v cable is the one that came with the GPU but two of the 8-pins going into the adapters are from a daisy-chain. @r.h.p Is this what you were referring to, btw?

I know this set-up can get you sent to the gulag but as long as the cables are fine I've been told this isn't a problem.

 

[Prognatodon]

two of the 8-pins going into the adapters are from a daisy-chain

So a 3*8 to 12+4 adapter over another old 8 to 2*8 adapter?

 

[Dr. Dro]

I just hope my GPU hasn't been taking damage each crash, lol. It's not even been 1 week since I got it. I suppose if the problem is a lack of power as opposed to too much then components should be safe?

Alright, I'll go looking! Will be a short few weeks before I can purchase but I'll keep things under light load until then. Appreciate the help from everyone.

One of the PSU 8-pins is daisy-chained. I've had this set up for years since my 6800 required x3 8-pins. I just connected them to the adapter for the 9070. The 12v cable is the one that came with the GPU but two of the 8-pins going into the adapters are from a daisy-chain. @r.h.p Is this what you were referring to, btw?

I know this set-up can get you sent to the gulag but as long as the cables are fine I've been told this isn't a problem.
View attachment 391008View attachment 391009

Would not be worried about damage for now, although I wouldn't keep on running it at the breaking point either. Lowering the power limit to minimum possible will minimize risks and crashes for the time being.

 

[NSR]

So a 3*8 to 12+4 adapter over another old 8 to 2*8 adapter?

Not sure if I follow your breakdown. I'll repeat it how I understand it but I think that sounds about right.

In my pictures you can see the 3*8 slots coming from the 12vhpwr cable. One of those slots is occupied by an 8pin directly from the PSU. The other two are occupied by a splitter cable that is powered by the other 8pin from the PSU. I think that's basically what you said.

 

[TechBuyingHavoc]

When I was buying my PSU, I used a guide like this recommendation list that describes different tiers of PSU quality (based on internal topology, component selection, etc) as a rough guide. That list is a bit old now so you can supplement it with something like this Google document which is updated as recently as March 7 2025.

Dr. Dro has the right recommendations for the best budget PSUs, I myself have a 2022 Corsair RM750x and I am perfectly happy with it. In hindsight, I would have gone with the 850x model based on how GPU power keeps going up and up.

 

[Prognatodon]

I think that's basically what you said

Yup.
Now I wonder what's gonna happen if you ditch that old splitter cable, the whole thing technically runs on one 8 pin cable with another integrated splitter already.

This kind of adapter centipede is not incapable of causing extra voltage drop. But, depending on how the Sapphire cable is wired internally, it might allow running the card on just 2 8pins as pictured. Although, if my stuff was getting flaky to the point of monitors not detecting (if it's what that was) I'm not sure if I'd proceed with the experiments.

 

[NSR]

Yup.
Now I wonder what's gonna happen if you ditch that old splitter cable, the whole thing technically runs on one 8 pin cable with another integrated splitter already.
View attachment 391027
This kind of adapter centipede is not incapable of causing extra voltage drop. But, depending on how the Sapphire cable is wired internally, it might allow running the card on just 2 8pins as pictured. Although, if my stuff was getting flaky to the point of monitors not detecting (if it's what that was) I'm not sure if I'd proceed with the experiments.

Interestingly, the 9070 Nitro+ is documented on here as being a 2*8 GPU even though Sapphire provides a 3*8 adapter.
Sapphire NITRO+ RX 9070 Specs | TechPowerUp GPU Database

I won't be the one to test it though, lol.

 

[LabRat 891]

IDK if mentioned already...
Newest ATX v3.1 PSUs are specified to deal with the high transient currents modern GPUs often pull (for milliseconds at a time).

While Navi48 does run on average *much* more efficiently than Navi31, it does transient spike to about double the 300-340W 'power limit'.

Interestingly, the 9070 Nitro+ is documented on here as being a 2*8 GPU even though Sapphire provides a 3*8 adapter.
Sapphire NITRO+ RX 9070 Specs | TechPowerUp GPU Database
View attachment 391043
I won't be the one to test it though, lol.

TPU's Db is largely automated, IIRC.
'Reference' Navi 48 cards, use 2x 8-pins.

The Red Devil, and a few others use 3x 8-pins.
The Nitro and Taichi OC with 12-pins(IIRC) belong to the same power class as the Red Devil.

3x 8-pin native Navi 48 cards and 12-pin native cards, are comparable.
2x 8-pin Sapphire Pulse, and other 2x 8-pin cards, have Reference power limits.AFAIK, the 3x8-pin and 12-pin cards have higher power limits.

 

[r.h.p]

Interestingly, the 9070 Nitro+ is documented on here as being a 2*8 GPU even though Sapphire provides a 3*8 adapter.
Sapphire NITRO+ RX 9070 Specs | TechPowerUp GPU Database
View attachment 391043
I won't be the one to test it though, lol.

SAPPHIRE NITRO+ AMD Radeon™ RX 9070 GPU
came across this for info. Did you get the bundled power accessories and sag bracket? @LabRat 891 is correct, the new 3.1 atx psu is designed for this card. The 12v cable should slot straight in

 

[NSR]

IDK if mentioned already...
Newest ATX v3.1 PSUs are specified to deal with the high transient currents modern GPUs often pull (for milliseconds at a time).

While Navi48 does run on average *much* more efficiently than Navi31, it does transient spike to about double the 300-340W 'power limit'.

Someone earlier recommended I buy ATX 3.1 but didn't add this. Good to know!

SAPPHIRE NITRO+ AMD Radeon™ RX 9070 GPU
came across this for info. Did you get the bundled power accessories and sag bracket? @LabRat 891 is correct, the new 3.1 atx psu is designed for this card. The 12v cable should slot straight in

Yeah, I got everything. I think it's all the same as the XT. I'm using their 12vhpwr adapter but a different sag bracket with better support.

 

[Hyderz]

I’d suggest not playing anymore until psu is swapped lucky no other parts have been damaged especially when parts are expensive these days to replace

 

[eidairaman1]

Not sure if I follow your breakdown. I'll repeat it how I understand it but I think that sounds about right.

In my pictures you can see the 3*8 slots coming from the 12vhpwr cable. One of those slots is occupied by an 8pin directly from the PSU. The other two are occupied by a splitter cable that is powered by the other 8pin from the PSU. I think that's basically what you said.

The more pigtails you have the less power supplied to each port, use direct lines, not pig tails

 

[Bomby569]

this is were your you could be wrong
...Furmark only stresses the GPU and playing a Game / wateva blender or what , uses CPU/GPU/RAM/MOBO. IF ur psu isnt delivering enough juice it WILL crash to Black screen and still power the basics like
cpu fan etc but lose the ability to control itself EG: restart from case or in worse case psu switch

my bad, i forgot about adding something to also stress the CPU, the rest is really irrelevant here.

For a start, Furmark and Heaven benchmarks don't use path tracing or raytracing, which it seems initiates the issue, neither do they display the sort of transients power spikes that Cyberpunk does in my own system. I understand you have no idea of my background, but as a 64 year old with over 30 years of pc building experience, my best guess is his his PSU is unable to supply a stable 12v to his GPU under extreme load.

like i said about i did forgot to add something to also stress the CPU. Transients are irrelevant if you are already maxing CPU and GPU, you can't go over the max.
Not sure what path or ray has to do with this. if you max the GPU wattage it can't go any higher, the limit is the limit no matter what you are making it do.

 

[AusWolf]

like i said about i did forgot to add something to also stress the CPU. Transients are irrelevant if you are already maxing CPU and GPU, you can't go over the max.
Not sure what path or ray has to do with this. if you max the GPU wattage it can't go any higher, the limit is the limit no matter what you are making it do.

I'm not sure what RT or PT has to do with this, either, but you graphics card can certainly go over its max limit for a fraction of a second. Hence diagrams like this:

 

[Dr. Dro]

my bad, i forgot about adding something to also stress the CPU, the rest is really irrelevant here.
like i said about i did forgot to add something to also stress the CPU. Transients are irrelevant if you are already maxing CPU and GPU, you can't go over the max.
Not sure what path or ray has to do with this. if you max the GPU wattage it can't go any higher, the limit is the limit no matter what you are making it do.

No, that's not how it works. The ATX 3.0+ spec specifically accounts for the momentary power excursions caused by load transients at up to 2x the PSU's rated capacity, specifically because they can and will cause ATX 2 PSUs to shutdown. Even high quality units aren't immune, there was a batch of Seasonic Prime power supplies a few years ago that had their OCP point set incorrectly, and the RTX 30 series GPUs caused them to shutdown - Seasonic exchanged these iirc.

While you can run a 12VHPWR (H+) or 2x6 (H++) connector graphics card off an older ATX 2 power supply without issue, generally only the highest capacity models with gargantuan 12V rail capacity will withstand them, preference for models that exceed 1000 W on 4090, 1200 W on 5090. The RX 9070's and lesser Nvidia GPUs should get away with a 750 W ATX 2 PSU just fine.

On the same vein, the power supply OP has fails even the most basic of power supply elementals like the ATX 2 spec's hold up time, this is specified at 16.6 ms but the bulk capacitor is too small and it only holds 10 ms, for example - and given the double forward/SBR topology, this is a power supply intended for only strictly basic PCs. You can run a GPU off it, sure, something like a basic model RTX 3050/4060 or 6600 XT would be about the limit, best to stick to iGPUs on this one. It isn't an IED (certainly won't go kaboom), but it's not meant to be used on any kind of high performance gaming PC.

 

[Clanseven]

my bad, i forgot about adding something to also stress the CPU, the rest is really irrelevant here.



like i said about i did forgot to add something to also stress the CPU. Transients are irrelevant if you are already maxing CPU and GPU, you can't go over the max.
Not sure what path or ray has to do with this. if you max the GPU wattage it can't go any higher, the limit is the limit no matter what you are making it do.

I'm not sure you understand that transient power demands by the GPU have little to do with its max rated power draw. To see a very good example of this, please watch Aris of Hardware Busters video

showing a 3080Ti card punishing a Corsair AX1000 watt PSU. Aris is the PSU expert behind Cybernetics Labs, the world renowned PSU testing organisation. So, transient power spikes are far from irrelevant.
I owned a 3080Ti, suffering occasional issues similar to NSR's. I upgraded my PSU to a Seasonic Prime 1000w TX, Titanium rated PSU. with rock solid load regulation. Issues solved.

 

[Vayra86]

@Dr. Dro @r.h.p

1. Tested the -30% power limit, GPU power peaked at 170W, no crashes for a while but eventually game/adrenaline crash. This is a good sign because the GPU didn't crash, just the software.
2. I set the power limit to -20% to fix the instability. Worked like a charm, played at maximum settings, Ultra + Path-Tracing, for 15 minutes fighting in heavy combat against the police. GPU never went above 197W.
View attachment 391001
3. I returned everything to default, GPU instantly started consuming 240W and I crashed after 10 seconds. This time it damn-near bricked my system...One display stopped working so I reset my PC. But then neither display was working. Tried the motherboard display since the 7600X has a integrated graphics, still not working. Had to clear CMOS and it booted up again.


PSU issue it seems but the 12v cable is actually doing just fine. I opened up the backplate last night after (it's magnetic) and I see no sign of damage or smell anything burning.
View attachment 391002


I'm going to order a better PSU and keep my GPU power-limited until that arrives. Didn't realise I was playing this close to fire (literally).

Better yet than just a -30% power limit: limit your GPU clock to 2400 mhz (Max frequency slider).

You will see 150W I reckon in Cyberpunk. Max frequency is the real deal here for underclocking.

 

[eidairaman1]

Use this group

RX 9000 series GPU Owners Club

With the release of the Radeon RX 9000 series, it seemed like a great time to start a club for everyone to discuss, compare performance, and share experiences. Pretty sure I’m not the first here. Keep it civil, have fun, and let’s talk GPUs! Looking forward to seeing everyone’s builds...

www.techpowerup.com

 

[NSR]

Two months ago I made this thread about my 9070 crashing with Cyberpunk's path tracing. At the time it seemed like a PSU issue and whilst changing PSU has resolved the impact of the crash (no longer have to reset my system), there is still some type of conflict between the 9070 and this setting.

I have crashed after a fresh Windows install, with a power limit, with an overclock, removing Adrenaline and only reinstalling the drivers, with or without mods, FSR performance from 1080p to 360p.

I know it's not just me as other 9070 owners on the radeon subreddit are experiencing the same thing:

But no reviewers have reported such an issue. Not sure if it's because their benchmarks are too short, if they crashed and disregarded it as an anomaly or if they genuinely just have no issues with this scenario.

To be clear, the 9070 can achieve good stutter-free FPS with path tracing enabled and some upscaling. That's not the problem. The problem is it'll just suddenly crash with zero indication why.

My specs for reference:
9070 Sapphire NITRO+
7600X
32GB 6000MT/s CL30
RM1000e

Game is installed on a games-only 1TB 990 Pro with 500GB free.

Everything updated (drivers, chipsets, windows, game).

 

[Chrispy_]

I've found CP2077 to be the most demanding thing I run in terms of undervolting. Most things are stable with an 80-100mv undervolt, everything except CP2077 with Path Tracing is rock solid with a 75mv undervolt, but CP2077 Overdrive with framegen has crashed like that, and I've dropped it to -65mv, at which point I have another 60-70 hours without a single crash.

The Nitro+ is a factory overclocked card, and that factory overclock may not be 100% stable. For the sake of ruling out the overclock as an issue, go into the Adrenaline driver's performance/tuning page and take 400-500MHz off the maximum clock slider and apply it - no need to change anything else.

You won't actually lose much performance, since CP2077 hovers at around 2.9-3.0GHz in CP2077 and a -400MHz clock adjustment will limit it to 2.9-3.0GHz. Do this just to see if dropping the clocks a bit stops the crashing. If it does, reach out to Sapphire support and explain the situation.

For what it's worth, I occasionally think to myself "man, this looks shit" with path tracing enabled, and I'll just toggle it off to see if it looks like shit with raytracing disabled. Half the time, no - the game is much better looking and properly lit with the fake raster lighting and shadowmaps.

Don't get me wrong, path tracing has the power to make games look absolutely stunning, but it was bolted onto CP2077 as an afterthought so there are plenty of situations where there's not enough lighting and the RT Overdrive preset just looks and runs like garbage. This is my opinion, obviously, but subjectively I feel like the artists that did the lighting and shadows of the game maps put lots of effort into their non-raytraced work, and that means the path traced version hasn't had the same level of care or attention to make it look good.

Oh, and for what it's worth, my two path-tracing crashes on the 9070XT at -80mv and -75mv were after at least a couple of hours in-game, so I don't think it's inherently a RDNA4 or driver problem.

 

[RejZoR]

My RX 9070 XT seems to be very picky with undervolt. I've had -75 for a while and it worked, but crashed every other full moon. Dropped it to -50 and it happened once every few weeks. I'm at -30 now to see what gives. I also use Radeon Chill so GPU clocks really fluctuate A LOT which is probably often the reason.

I might try Cyberpunk 2077 path tracing again. I've done it before and didn't have crashes, but ai only walked inside the starting building for the hacker path.

Path tracing is easy for developers though idTech 8 in Doom The Dark Ages really impressed me with ray tracing. Looks so good and runs insanely fast and without a single stutter. Probably the best and most advanced engine available at the moment.

 

[NSR]

The Nitro+ is a factory overclocked card, and that factory overclock may not be 100% stable. For the sake of ruling out the overclock as an issue, go into the Adrenaline driver's performance/tuning page and take 400-500MHz off the maximum clock slider and apply it - no need to change anything else.

I'll give this a shot and see how long it lasts.

Don't get me wrong, path tracing has the power to make games look absolutely stunning, but it was bolted onto CP2077 as an afterthought so there are plenty of situations where there's not enough lighting and the RT Overdrive preset just looks and runs like garbage. This is my opinion, obviously, but subjectively I feel like the artists that did the lighting and shadows of the game maps put lots of effort into their non-raytraced work, and that means the path traced version hasn't had the same level of care or attention to make it look good.

Personally I think PT provides the most visually impressive image but the base game is also pretty enough that RT and non RT can hold their own weight.

Path tracing is easy for developers though idTech 8 in Doom The Dark Ages really impressed me with ray tracing. Looks so good and runs insanely fast and without a single stutter. Probably the best and most advanced engine available at the moment.

I haven't had the opportunity to play Dark Ages yet but putting aside the performance hit from Eternal, the game looks very good and people seem to be enjoying it.

 

[Vya Domus]

Cyberpunk is an Nvidia tech demo, in other words expect no fix for this.