AMD RX 7000 series GPU Owners' Club

[GamerGuy]

Card is dead...
crashed a LOT yesterday (nothing overclocked) and then it was no longer recognized by the drivers. while DDU the whole screen was full of artifacts and the PC shut off and refuses to power on with the card in the system.
View attachment 338115

It should be under warranty, right? RMA that sucka!

 

[b0uncyfr0]

It seems i made a little boo boo with my 7800xt nitro

My card is vertically mounted and initially (with higher/default voltages) - the hotspot was close to a 100 which is not ideal. I wanted to try and reduce it, so i took the backplate off. I was thinking my giant 30cm fans directly below it would help. Turns out this made it worse!

I benched all the time with the backplate off and decided to put it back on & undervolt once more. And voila, it is now significantly better

• Im currently at 1050mv at 2600 core and memory. It hovers around 63 and 72 hotspot

Because its so cool, it's actually boosting past 2600 core in most cases. Ive seen it hold 2625 ish.

Im thinking i can squeeze more out of the core, maybe 2650.

Is it worth it to drop the memory OC (maybe even reduce it) and aim for a higher core? I game at 1440p/120-144hz ideally.

 

[Ferrum Master]

I am saying this particular Powercolor 7900XTX Liquid Devil model is intended to work this way. It works and works well but with a very high delta to the GCD hotspot.

Holy cow. Just reread this thread... my watercooled card hits max ~85C hotspot. There is something wrong with pressure, thermalpad thickness or whatever, don't kill the card, it should not be that hot on water, ever.

Count me in

Warranty seal seems broken. But you can order some on aliexpress lol.

 

[AusWolf]

It seems i made a little boo boo with my 7800xt nitro

My card is vertically mounted and initially (with higher/default voltages) - the hotspot was close to a 100 which is not ideal. I wanted to try and reduce it, so i took the backplate off. I was thinking my giant 30cm fans directly below it would help. Turns out this made it worse!

I benched all the time with the backplate off and decided to put it back on & undervolt once more. And voila, it is now significantly better

• Im currently at 1050mv at 2600 core and memory. It hovers around 63 and 72 hotspot

Because its so cool, it's actually boosting past 2600 core in most cases. Ive seen it hold 2625 ish.

Im thinking i can squeeze more out of the core, maybe 2650.

Is it worth it to drop the memory OC (maybe even reduce it) and aim for a higher core? I game at 1440p/120-144hz ideally.

Wow, 2600 on a 7800 XT is very nice! My Sapphire Pulse only goes up to 2500-ish max.

As for lowering memory, I don't know, but might be worth having a go. It's usually the memory power consumption that's hindering the GPU from achieving higher clocks.

 

[mama]

Holy cow. Just reread this thread... my watercooled card hits max ~85C hotspot. There is something wrong with pressure, thermalpad thickness or whatever, don't kill the card, it should not be that hot on water, ever.



Warranty seal seems broken. But you can order some on aliexpress lol.

The only component of the die that is unusually hot (and maintains unusually high temps on average is the GCD hotspot). I have repasted twice. Including the original paste, which looked okay to me, that's three coverage attempts. Water pressure in the loop is not an issue (pump speed is monitored). The issue is the card lets in near 600W at times (even on the tamer BIOS). I can't change that despite undervolting and reducing power. To be honest I have given up trying.

If you search around you'll find others with this card having similar issues. Hence my conclusion that this particular card is meant to do this.

 

[AusWolf]

The only component of the die that is unusually hot (and maintains unusually high temps on average is the GCD hotspot). I have repasted twice. Including the original paste, which looked okay to me, that's three coverage attempts. Water pressure in the loop is not an issue (pump speed is monitored). The issue is the card lets in near 600W at times (even on the tamer BIOS). I can't change that despite undervolting and reducing power. To be honest I have given up trying.

If you search around you'll find others with this card having similar issues. Hence my conclusion that this particular card is meant to do this.

If I ever wanted to repaste my card, I'd try the TG Kryosheet. Most pastes are famously bad with large GPUs, especially these chiplet ones.

 

[Ferrum Master]

The only component of the die that is unusually hot (and maintains unusually high temps on average is the GCD hotspot). I have repasted twice. Including the original paste, which looked okay to me, that's three coverage attempts. Water pressure in the loop is not an issue (pump speed is monitored). The issue is the card lets in near 600W at times (even on the tamer BIOS). I can't change that despite undervolting and reducing power. To be honest I have given up trying.

If you search around you'll find others with this card having similar issues. Hence my conclusion that this particular card is meant to do this.

First. If you do water cooling, do not under-volt, it beats the whole idea being on water. Being on Air, yes, for sure UV is a good idea, but not water, I use a shoddy Bykski waterblock that didn't even have proper screws and lacks right mounting holes, but the temps are absolutely fine. Not my first rodeo on water. Your BIOS is also tailored to eat as much it can as it designed with a large cooling budget. All our attempts to interfere with AMD boost algo in reality often brakes something and card starts to underperform.

Aren't you clogged up? The Pump RPM won't indicate a clogged loop, I saying from my own experience, just like an engine it will sound different, more strained.

My card peaks past 550W and have absolutely no problems with it. I bet there is something wrong, like thermal pad thickness for vram, so the pressure ain't enough. Try Kryosheet as last straw. It adds thickness, witch you may need. PTM79xx aint good for water cooling as usually the loop temps are under 40C so it doesn't work, it should with your soaring 100C+ tho.

Try looking how even are the surfaces of the block also. It would be great to give it another change, try taking pictures and post it here, more eyes may help finding the problem cause. For pads order some thermal putty, it will eliminate any problems with pad sizes and thickness, but only if you card is mounted in classic manner, if you use a riser and mount it vertically for the show, then putty is a no go, it drips down, we can't fight gravity.

 

[Kapone33]

First. If you do water cooling, do not under-volt, it beats the whole idea being on water. Being on Air, yes, for sure UV is a good idea, but not water, I use a shoddy Bykski waterblock that didn't even have proper screws and lacks right mounting holes, but the temps are absolutely fine. Not my first rodeo on water. Your BIOS is also tailored to eat as much it can as it designed with a large cooling budget. All our attempts to interfere with AMD boost algo in reality often brakes something and card starts to underperform.

Aren't you clogged up? The Pump RPM won't indicate a clogged loop, I saying from my own experience, just like an engine it will sound different, more strained.

My card peaks past 550W and have absolutely no problems with it. I bet there is something wrong, like thermal pad thickness for vram, so the pressure ain't enough. Try Kryosheet as last straw. It adds thickness, witch you may need. PTM79xx aint good for water cooling as usually the loop temps are under 40C so it doesn't work, it should with your soaring 100C+ tho.

Try looking how even are the surfaces of the block also. It would be great to give it another change, try taking pictures and post it here, more eyes may help finding the problem cause. For pads order some thermal putty, it will eliminate any problems with pad sizes and thickness, but only if you card is mounted in classic manner, if you use a riser and mount it vertically for the show, then putty is a no go, it drips down, we can't fight gravity.

When I turned up my 7900XT it was pulling 407 Watts (Max) but maintained a 2989 MHZ GPU clock. I agree with the undervolting sentiment though. These cards are (7900) super fast if you let them run unleashed.

 

[3x0]

Im currently at 1050mv at 2600 core and memory. It hovers around 63 and 72 hotspot

From my experience, you'll end up finding a game that is unstable at those settings. For me, certain stress tests crashed even at 2300MHz and 1100mV. So be prepared for disappointment

 

[Vya Domus]

First. If you do water cooling, do not under-volt, it beats the whole idea being on water. Being on Air, yes, for sure UV is a good idea, but not water, I use a shoddy Bykski waterblock that didn't even have proper screws and lacks right mounting holes, but the temps are absolutely fine. Not my first rodeo on water. Your BIOS is also tailored to eat as much it can as it designed with a large cooling budget. All our attempts to interfere with AMD boost algo in reality often brakes something and card starts to underperform.

Undervolting gets you more performance, there is no reason to not do it.

Try Kryosheet as last straw. It adds thickness, witch you may need. PTM79xx aint good for water cooling as usually the loop temps are under 40C so it doesn't work, it should with your soaring 100C+ tho.

Just a day ago I applied PTM7950 on my watercooled 7900XT and temperatures are markedly better, it definitely does work.

 

[remekra]

I have my Nitro+ on 460W. If we look at hwinfo maximum power draw reports it's hitting over 500W easily and yet at 40% fan speed set to max it runs at 88-90C on hotspot and delta is only 20C to edge.
So I would say that there is something wrong with the card or mounting of the waterblock.

 

[Ferrum Master]

Undervolting gets you more performance, there is no reason to not do it.

Watercooled cards have BIOS with most high set power limits, like Asrock Aqua has. You don't need to play around have more power headroom that limited cards have. That's the logic. For air cooled cards, yes do UV job, it helps a lot as those are limited. I have my card flashed with SPI programmer and have my limits upped unlike you I guess and mama has a pure WC card with a native high power BIOS, some call those Tier0. For watercooled ones just leave the voltage table alone and increase the clocks... the power budget is enough and cooling capacity also to unleash full power.

Regarding PTM7950, it will work... but this stuff has been around for a while but some still tend to replace those in their laptops, thinking it would improve something, some times it does actualy, some not. I saw one Youtube muppet taking apart RoG Ally, can't remember who, but in the end he concluded that the stock paste was better than all calssic ones he tried... that retard could have known that Asus uses PTM7958 for a while already too... on the serious note, I've seen pumpouts also of PTM paste in laptops... well I see those much more around and taken apart daily, as it is my job, but my words stand firm. It aint good for temps under 40C, the phase change magic just doesn't work under that temperature and the material becomes solid. It is miles better regarding pump out phenomena, but not totally immune to it, thus why I opted too for Kryosheet as soon it came out. I have a theory that in laptops crap happens when the paste is in cold state and the device is carried over, bent etc being a mobile device, it creates some movement and thus we see some anomalies time to time.

I have my Nitro+ on 460W. If we look at hwinfo maximum power draw reports it's hitting over 500W easily and yet at 40% fan speed set to max it runs at 88-90C on hotspot and delta is only 20C to edge.
So I would say that there is something wrong with the card or mounting of the waterblock.

Yes 20C delta is the average you should see no matter what cooling you use. If it is more, it means something is wrong.

 

[Vya Domus]

Watercooled cards have BIOS with most high set power limits, like Asrock Aqua has. You don't need to play around have more power headroom that limited cards have. That's the logic. For air cooled cards, yes do UV job, it helps a lot as those are limited. I have my card flashed with SPI programmer and have my limits upped unlike you I guess and mama has a pure WC card with a native high power BIOS, some call those Tier0. For watercooled ones just leave the voltage table alone and increase the clocks... the power budget is enough and cooling capacity also to unleash full power.

These cards are always limited by power, undevolting always gives you higher clocks. Are you suggesting you get the same performance with your card whether you UV or not ? Because that I do not believe.

 

[Ferrum Master]

Because that I do not believe.

Who cares about that? Please crossflash your card and look, not sure there is actually such high power BIOS for an XT, I haven't looked, so far I see you operate from things you haven't tested yourself.

You can't UV if you do Core overclock, as most people use unstable systems actually and YOLO and never see highest boost states unless you use water, high power bios with enhanced current duration limits(no slider cures that, except using Elmor EVC to fool AMD Firmware). That's a different game going after ~3200MHz clock with UV the card will crap out. Its around 20-30% performance boost, that is the fair reason using water cooling for me, it has always been.

You really think those benchmark records are done using UV and I am all wrong on my approach?

 

[Vya Domus]

Who cares about that?

I simply don't understand what you are trying to say. UV gives you more performance, no matter what configuration you have, if you are trying to make the point that it's not worth it because of instability OK I guess but your point about it being OK with cards on air vs cards on water still makes no sense to me.

RDNA3 cards don't stay at their frequency limits for any extended periods of time because of power, UV allows the card to get closer to that limit but it will still never really run at 3.2 Ghz consistently.

 

[Ferrum Master]

RDNA3 cards don't stay at their frequency limits for any extended periods of time because of power, UV allows the card to get closer to that limit but it will still never really run at 3.2 Ghz consistently.

3.2GHz is load possible, you can do Timespy on 3.2GHz with no problems when set correctly.

I am asking you again, those people who make best scores on this card use UV? If you go for higher clocks you need voltage to be stable on RDNA3. To sustain that load, you are usually on water and pretty much with unlocked BIOS. So you are so locked in on your setup and completely ignore what others may have. I tried to explain why it works for you, because you are limited, tier0 BIOS is less constrained.

 

[Vya Domus]

I am asking you again, those people who make best scores on this card use UV?

For sure, if you want to maximize the performance you'll need to UV. As I already said these cards are always limited by power, even if you have a BIOS with a higher board limit you'll still have to UV if you want more performance.

3.2GHz is load possible, you can do Timespy on 3.2GHz with no problems when set correctly.

It's not going to run at 3.2 Ghz in all benchmark/games, lets be real, in fact in most it wont.

 

[KedsDead]

I have a 7900xt Red Devil, that I have repasted... just got it and did because I thought the temps were off. This thing on OC Bios, on air, with -100 undervolt and 0% PL, it still clocks to 3100+ MHZ, had to have the entire system on full roar, to even get the temps under 85c hotspot.

I have it running now at -5%pl and the clocks get to 2800 to 2900 and I can now run my fans at 65% and the hotspot be 82c.

I'm pretty sure the Red Devil OC bios is just moving the slider.. quite bios is 0-15% and OC bios is 15 - 30% or something like that..

 

[mama]

First. If you do water cooling, do not under-volt, it beats the whole idea being on water. Being on Air, yes, for sure UV is a good idea, but not water, I use a shoddy Bykski waterblock that didn't even have proper screws and lacks right mounting holes, but the temps are absolutely fine. Not my first rodeo on water. Your BIOS is also tailored to eat as much it can as it designed with a large cooling budget. All our attempts to interfere with AMD boost algo in reality often brakes something and card starts to underperform.

Aren't you clogged up? The Pump RPM won't indicate a clogged loop, I saying from my own experience, just like an engine it will sound different, more strained.

My card peaks past 550W and have absolutely no problems with it. I bet there is something wrong, like thermal pad thickness for vram, so the pressure ain't enough. Try Kryosheet as last straw. It adds thickness, witch you may need. PTM79xx aint good for water cooling as usually the loop temps are under 40C so it doesn't work, it should with your soaring 100C+ tho.

Try looking how even are the surfaces of the block also. It would be great to give it another change, try taking pictures and post it here, more eyes may help finding the problem cause. For pads order some thermal putty, it will eliminate any problems with pad sizes and thickness, but only if you card is mounted in classic manner, if you use a riser and mount it vertically for the show, then putty is a no go, it drips down, we can't fight gravity.

Thanks for the reply. Check my previous posts for HWInfo64 observations. Vram cooling is fine. In fact, everything is nice and cool except the GCD observation. The pads don't need changing. They are in good condition and obviously working. The loop is super clean. I've changed the liquid twice now in a matter of two months. BTW I use EK CryoFuel (not the solid stuff). I don't use a riser.

You may be onto something with the paste/pad thickness on the die. I will say though the card came like this (slightly worse in fact). My attempts to have it exchanged where not successful so I thought there was nothing to lose to try a repaste (and another as it happened). As I say, if you look around the net you'll find others with the same issue for this card. At the moment I don't feel motivated or inclined to empty the loop again, repaste and refill the loop. Maybe later. I just thought it was a public service to let people know about the Powercolor 7900 XTX Liquid Devil or my experience with it at least.

 

[Ferrum Master]

Powercolor 7900 XTX Liquid Devil or my experience with it at least.

I can't remember(overclock.net probably) where and what card, but it isn't that rare that cards are shipped with wrong thickness thermal pads. You have to consider that also. VRAM tems will be fine if the pads are too thick.

Give a Kryosheet a shot. Actually many are complaining about hotspots on these cards when doing water cooling it is not limited to the Vector 2 block you have. You have to treat is a sports car, you have to delicate and high precision.

 

[AusWolf]

For sure, if you want to maximize the performance you'll need to UV. As I already said these cards are always limited by power, even if you have a BIOS with a higher board limit you'll still have to UV if you want more performance.

Call me whatever, but I don't see how undervolting can pass as an all-around thing to do for everyone. With less voltage comes less stability, and maybe your card is stable 99% of the time, so you won't even notice, but you do lose stability when you UV. It's just the nature of things.

The factory assigned set voltage and frequency values for every GPU, and I don't pretend to know any better. Maybe you got a golden sample, and you can do some beautiful UV (while still losing some stability), but all the AMD cards I've recently owned got unstable at -25 mV, which is nothing. I agree about the power limit being the no.1 factor limiting performance, but I don't believe UV is the way around it.

 

[Ferrum Master]

recently owned got unstable at -25 mV, which is nothing. I agree about the power limit being the no.1 factor limiting performance, but I don't believe UV is the way around it.

Just exactly my experience. There are dozens of games that are fine with it. But time tome time a crash occurs and actually for most times during idle. Guess what? Not screwing with UV cured all problems.

 

[gurusmi]

Call me whatever, but I don't see how undervolting can pass as an all-around thing to do for everyone. With less voltage comes less stability, and maybe your card is stable 99% of the time, so you won't even notice, but you do lose stability when you UV. It's just the nature of things.

The factory assigned set voltage and frequency values for every GPU, and I don't pretend to know any better. Maybe you got a golden sample, and you can do some beautiful UV (while still losing some stability), but all the AMD cards I've recently owned got unstable at -25 mV, which is nothing. I agree about the power limit being the no.1 factor limiting performance, but I don't believe UV is the way around it.

One is playing with tolerances of all the parts. There is a frame in which a inbound/outbound signal should be. TTL is known for 5V +/-5%. That means a TTL-circuit will run with 5v+/-0,05V. If every vgoltage is inside this voltage frame the whole circuit is fine. I not it can fail nut it is not said that it will fail. There could be parts on it thatmight have a higher tolerance But also that is not guaranteed.

 

[tpa-pr]

Speaking of undervolting:

This got me 146fps average in CP2077 maxed out! And it hit 3.25Ghz effective clocks sustained in GravityMark. It seems the only part of the card that's touchy is the VRAM so I've kept it at stock with fast timings. Boy this card is great.

Also, the paste job must be phenomenal. My room was 35C ambient temp yesterday and the junction temp still only hit 88C at the max under heavy load. Average at the moment outside of the extreme ambient is 80-82C. I'm actually curious if PowerColor used PTM7950 on this card but I'm too scared to open it up and confirm!

 

[Space Lynx]

Speaking of undervolting:
View attachment 339461

This got me 146fps average in CP2077 maxed out! And it hit 3.25Ghz effective clocks sustained in GravityMark. It seems the only part of the card that's touchy is the VRAM so I've kept it at stock with fast timings. Boy this card is great.

Also, the paste job must be phenomenal. My room was 35C ambient temp yesterday and the junction temp still only hit 88C at the max under heavy load. Average at the moment is 80-82C. I'm actually curious if PowerColor used PTM7950 on this card but i'm too scared to open it up and confirm!

yeah, I feel like VRAM oc has to be kept off, otherwise core won't max itself. VRAM oc doesn't really gain anything anyway in my testing. the gains come from the core undervolt and boost.