RX 9000 series GPU Owners Club

[LittleBro]

And fan helps to keep the connector cool. Just in case, you know ...

 

[rusty caterpillar]

No. You followed up with answers to your own question.


Fine, I did that - with one modification. LACT wouldn't let me set a fan curve (I'm on Linux, I guess the card is too new even for MESA 25), so I left it on Auto, but lowered the power limit to 280 W as you asked.

My VRAM temperature still peaked at 88 ˚C. So what was the point of this test?

Not sure you can do the Superposition 8k optimized test on Auto fans, will be inaccurate cause I fixed mine at before the test 1400 RPM. On auto will just go up and down or will stay too low. Didn't know you are on Linux.

And fan helps to keep the connector cool. Just in case, you know ...

And the mesh stops the airflow coming from the third fan

 

[Pan Ogr]

06.03.2025, PowerColor Red Devil, mediaexpert.pl, €937 VAT included, Poland

even now there are no cards in msrp and never was I wanted this model so ... other models/cards (even "standrad" ones ) are cheaper 50-100€. yeah. something is wrong here in PL

 

[spacer00ster]

Thread has now fallen off with all the back and forth cussing. Happened earlier than I was expecting SMH

In other news, congrats @Fluffmeister

Okay sign me up to the club! On Aprils fools day my wallet accidentally opened by mistake and ordered a Sapphire Nitro+ Radeon RX 9070 XT from Overclockers UK for the princely sum of £719 Inc VAT. I laughed... I cried and then had a cold shower.

Been using it for a few days and it's very very good indeed, I'm getting a new case as I needed a crowbar to get it inside my trusty but aging Antec 1200.

View attachment 393266
View attachment 393267
View attachment 393268

 

[AusWolf]

Not sure you can do the Superposition 8k optimized test on Auto fans, will be inaccurate cause I fixed mine at before the test 1400 RPM. On auto will just go up and down or will stay too low. Didn't know you are on Linux.

Ah, you wanted the 8K optimized test! Well, that gets my VRAM up to 92 ˚C, but keep in mind, we're still talking about junction, not case temp.

Fan speed doesn't really fluctuate on my card, Powercolor did a good job with its VBIOS.

I've also got an emergency Windows installation besides Linux, but I rarely ever use it to be fair.

 

[rusty caterpillar]

Ah, you wanted the 8K optimized test! Well, that gets my VRAM up to 92 ˚C, but keep in mind, we're still talking about junction, not case temp.

Fan speed doesn't really fluctuate on my card, Powercolor did a good job with its VBIOS.

I've also got an emergency Windows installation besides Linux, but I rarely ever use it to be fair.

So you have done Superposition 8k optimized 25 min? Care to share your screen with HW info like this one? no need for GPUz



The only advantages my card has: is a middle plate and my modifications(deshrouding, T putty and extra fan) but, you have to bare in mind is a card from 2017 and I'm the secondary owner.
However IMO, 8 years newer cards should really surpass mine in cooling despite the position of the sensors.


Until AMD and SK Hynix share some more info I wouldn't put my trust in VBIOS that much. IS your card and is your choice but, it wouldn't hurt to ad at least a 92/15mm fan on your backplate.

We know already 1 issue : Thermal Resistance on GDDR6 is high, so the heat traversing layers of the chip towards surface or the other way doesn't met Thermal Pads fast enough.
2 issue: PCB density is high on this boards resulting in heat by proximity where a middle plate would have helped but are no 9000 series with such thing yet.

 

[TSiAhmat]

So you have done Superposition 8k optimized 25 min? Care to share your screen with HW info like this one? no need for GPUz

View attachment 393314

The only advantages my card has: is a middle plate and my modifications(deshrouding, T putty and extra fan) but, you have to bare in mind is a card from 2017 and I'm the secondary owner.
However IMO, 8 years newer cards should really surpass mine in cooling despite the position of the sensors.


Until AMD and SK Hynix share some more info I wouldn't put my trust in VBIOS that much. IS your card and is your choice but, it wouldn't hurt to ad at least a 92/15mm fan on your backplate.

We know already 1 issue : Thermal Resistance on GDDR6 is high, so the heat traversing layers of the chip towards surface or the other way doesn't met Thermal Pads fast enough.
2 issue: PCB density is high on this boards resulting in heat by proximity where a middle plate would have helped but are no 9000 series with such thing yet.

what? I am so confused by this post.

This is an 1080ti with gddr5x which (I assume) measures edge temp of the memory at best. Are you sure you can compare the temps of VRAM on your card to the rx7000, rx9000 or even rtx5000 series?

 

[SirKeldon]

06.03.2025, PowerColor Red Devil, mediaexpert.pl, €937 VAT included, Poland

even now there are no cards in msrp and never was I wanted this model so ... other models/cards (even "standrad" ones ) are cheaper 50-100€. yeah. something is wrong here in PL

Wow, launch day and those prices already? Definitely you're right about pricing in Poland, jeez! Hope you're happy with your purchase and that it performs like a beast, adding you to the list ASAP

Thread has now fallen off with all the back and forth cussing. Happened earlier than I was expecting SMH

I wasn't expecting so many reactions either, but that's just how public forums are, some people are more passionate, others prefer to stay quiet.

The VRAM topic has already been discussed at length in another thread, and I don't mind it being brought up here again, as long is a respectful conversation. Just to clarify, my fan curve suggestion was meant as a precaution, not to stir panic like in the Rejzor thread finally happened. These cards are only a month old, and so far there's only one officially reported malfunctioning case that's temp-related, but it's unclear if it's actually connected to this issue.

Most of us have between 1 and 3 years of warranty (3 in my case), and RMA processes are something manufacturers try to avoid, so IMHO we're pretty safe at least during that period. I adjusted my fan curve more out of habit and obsession than concern. Until VRAM failures become widespread and consistent, it's all just speculation, no reason to panic for now, cause none has appeared yet; cinemaware's 9070, despite running terribly hot, it still works ok AFAIK, correct me here if I'm wrong.

So yeah, my conclusion is: use your cards as normal, cause temps are within specs as stated, tweak them if you like (at your own risk, of course), and let's only worry if there's really something to worry about.
In the meantime, let's enjoy our GPUs

 

[lexluthermiester]

Ah, you wanted the 8K optimized test! Well, that gets my VRAM up to 92 ˚C, but keep in mind, we're still talking about junction, not case temp.

Yeah, that might still be a problem. That VRAM shouldn't be getting above 85C, even the T-Junction. What happens when you increase your fan curve to compensate?

 

[AusWolf]

So you have done Superposition 8k optimized 25 min? Care to share your screen with HW info like this one? no need for GPUz

No HWinfo or GPU-Z. I'm on Linux, remember? The only measurement I have is in the system monitor. I could swap my Windows SSD, but I think that would be pointless because...

View attachment 393314

The only advantages my card has: is a middle plate and my modifications(deshrouding, T putty and extra fan) but, you have to bare in mind is a card from 2017 and I'm the secondary owner.
However IMO, 8 years newer cards should really surpass mine in cooling despite the position of the sensors.

You want to compare the GDDR6 hotspot sensor on my 9070 XT to the unknown GDDR5 sensor (I assume, edge temp) on your 1080 Ti? In which twisted parallel universe does this make sense?

Until AMD and SK Hynix share some more info I wouldn't put my trust in VBIOS that much. IS your card and is your choice but, it wouldn't hurt to ad at least a 92/15mm fan on your backplate.

We know already 1 issue : Thermal Resistance on GDDR6 is high, so the heat traversing layers of the chip towards surface or the other way doesn't met Thermal Pads fast enough.
2 issue: PCB density is high on this boards resulting in heat by proximity where a middle plate would have helped but are no 9000 series with such thing yet.

Micron states that their chips have a max operating edge temp of 95 ˚C. The 9070 XT only reports hotspot temp on VRAM. So as long as my hotspot temp is lower than the max operating edge temp, I'm fine.

Yeah, that might still be a problem. That VRAM shouldn't be getting above 85C, even the T-Junction. What happens when you increase your fan curve to compensate?

Why is that a problem? Do you have any source on it? If it really was a problem, then AIBs wouldn't set a 108 ˚C max VRAM temp in the vBIOS.

I cannot increase my fan curve, as LACT still doesn't have full functionality with my card.

Guys, again... every single 9070 and 9070 XT runs with hot VRAM. Every. Single. One. Please don't tell me that there isn't any AMD AIB who knows how to build a card, or that they just suddenly forgot.

 

[lexluthermiester]

Why is that a problem? Do you have any source on it? If it really was a problem, then AIBs wouldn't set a 108 ˚C max VRAM temp in the vBIOS.

Not so much a source as just simple electronics thermal basics. When IC's(of any kind) get that hot, it equals bad. Always. I don't care what the thermal throttle temp is set to. Anything above 85C will cause long term degradation. This is IC physics 101.

I cannot increase my fan curve, as LACT still doesn't have full functionality with my card.

Can you adjust your clocks/voltage down a bit to see if that will improve the situation?

Guys, again... every single 9070 and 9070 XT runs with hot VRAM. Every. Single. One.

Whether it does or not is unimportant, the temps need to come down if you want that card to go the distance for you.

 

[AusWolf]

Not so much a source as just simple electronics thermal basics. When IC's(of any kind) get that hot, it equals bad. Always. I don't care what the thermal throttle temp is set to. Anything above 85C will cause long term degradation. This is IC physics 101.

Then why does my 6750 XT still work perfectly with a 105 ˚C GPU hotspot 3 years after I bought it?

Can you adjust your clocks/voltage down a bit to see if that will improve the situation?

All I can set is a custom power limit for now, but that doesn't change much regarding VRAM temps.

Where it does or not is unimportant, the temps need to come down if you want that card to go the distance for you.

So far, there's no evidence for or against this claim.

 

[lexluthermiester]

Then why does my 6750 XT still work perfectly with a 105 ˚C GPU hotspot 3 years after I bought it?

It hasn't degraded enough yet? It's not a matter of if, but when a heat related failure will happen with those kinds of temps. It's best to keep thing cool.

All I can set is a custom power limit for now, but that doesn't change much regarding VRAM temps.

That sucks. Hopefully that fine-grained level of control will come.

So far, there's no evidence for or against this claim.

There is, just not specific evidence to that card. There is a reason why thermal throttle schemes exist. That condition along with the known science involved is enough evidence to prove that such high temps are inherently bad.

 

[AusWolf]

It hasn't degraded enough yet? It's not a matter of if, but when a heat related failure will happen with those kinds of temps. It's best to keep thing cool.


That sucks. Hopefully that fine-grained level of control will come.


There is, just not specific evidence to that card. There is a reason why thermal throttle schemes exist. That condition along with the known science involved is enough evidence to prove that such high temps are inherently bad.

I'm not doubting that high temps are bad. I'm only questioning what "high" actually means. If Micron states that their GDDR6 chips are fine up to an operating temp of 95 ˚C Tcase, then who am I to question it?

 

[TSiAhmat]

This VRam Temperature Topic is getting exhausting, just circling around with neither site having a definitive answer to it.

Just tested some more vulkan_memtest & I had 0 thermal throttling after 18 ~ 19 minutes (same temps of vram as 4 minutes in and no change in speed)

Spoiler: 4 ~ 5 Min

Spoiler: 18 ~ 19 Min

Can we lay the thematic to rest for now? We (the 9000 series owner) [or at least I, can't really speak for others] will report if anything is amiss with the cards/vram.

(If you already getting 105ish hotspot/Vram you should check your card -> Are the fans on/spinning, Is the case airflow appropriate [Case-fans in the wrong direction, maybe weird air turbulence or simply not enough fresh air] or bad pressure/paste contact with the die/vram modules [would personally RMA it if that were the case])

If feel like only time will tell if the temps are a real (and general) problem of the 9070 (xt) cards (especially after some years in the summer/hottest season)

 

[spacer00ster]

This VRam Temperature Topic is getting exhausting, just circling around with neither site having a definitive answer to it.

Can we lay the thematic to rest for now? We (the 9000 series owner) [or at least I, can't really speak for others] will report if anything is amiss with the cards/vram.

100% in agreement with @TSiAhmat.

In other news, @SirKeldon could you remove me from the XT list please? as I canceled my order. I've been getting great framerates (remember I'm coming from a GTX 1070), especially in my long gaming session yesterday evening/night and can't ask for more. Also didn't realize that I would have to get a new PSU . Good thing the card wasn't shipped yet.

I did a Monster Hunter benchmark just now with the 9070 at 1440p with results 23421 and avg. 136 FPS. This was with -100mv, +300MHz GPU clock & +10 PL.

 

[lexluthermiester]

If Micron states that their GDDR6 chips are fine up to an operating temp of 95 ˚C Tcase, then who am I to question it?

That's the limit. What they also state(and I can't find where) is that optimal nominal operating temps should kept under 80C.

However, I don't want to seem antagonizing to you, so I digress.

 

[skipdaddy911]

sapphire 9070 pulse PCDIGA in portugal for €750 so basically msrp. undervolt to -40mv offset +10 to power limit and mem to 2620mhz

 

[TSiAhmat]

I did a Monster Hunter benchmark just now with the 9070 at 1440p with results 23421 and avg. 136 FPS. This was with -100mv, +300MHz GPU clock & +10 PL.

View attachment 393511

Monster Hunter Wilds is quite inconsistent/buggy not the best indication if a gpu performs well or not, also the game is quite cpu bottlenecked (Just look at your max gpu Clock, holy sh*t)

That's the limit. What they also state(and I can't find where) is that optimal nominal operating temps should kept under 80C.

However, I don't want to seem antagonizing to you, so I digress.

just for your information (was in the sapphire 9070 xt pulse review):

and adrenalin/sensors reports tjunction not tcase

(my last post to that topic)

 

[lexluthermiester]

just for your information (was in the sapphire 9070 xt pulse review):View attachment 393523
and adrenalin/sensors reports tjmax not case

(my last post to that topic)

Yes, they can do it 24/7, but how long.. That does NOT mean it is optimal. I'm not blowing messiness out my backside here. I speak from experience and have seen system parts burn themselves out from running at such temps. CPU's, motherboards, PSUs and yes, GPUs. It happens and it's preventable. Those temps are NEVER good for any length of time, regardless if the parts can run at those temps. Just because it can does not mean it should.

 

[TSiAhmat]

In other news, @SirKeldon could you remove me from the XT list please? as I canceled my order. I've been getting great framerates (remember I'm coming from a GTX 1070), especially in my long gaming session yesterday evening/night and can't ask for more. Also didn't realize that I would have to get a new PSU . Good thing the card wasn't shipped yet.

A friend of mine swapped from a dying 1080 system to an 9070xt+9800x3d (complete system overhaul, expect 2 ssds) and he is also very happy with the performance.

Just look at the performance jump

 

[Ruru]

My friend has that classic Finnish enthusiasm as he took the no1 place.

3DMark.com search

3DMark.com search

www.3dmark.com

 

[SirKeldon]

In other news, @SirKeldon could you remove me from the XT list please? as I canceled my order. I've been getting great framerates (remember I'm coming from a GTX 1070), especially in my long gaming session yesterday evening/night and can't ask for more. Also didn't realize that I would have to get a new PSU . Good thing the card wasn't shipped yet.

Done! And yes, luckily you won't have to hassle with a return etc, I guess if you feel out of performance at some moment, you can always undervolt a lil bit and squeeze some more.

A friend of mine swapped from a dying 1080 system to an 9070xt+9800x3d (complete system overhaul, expect 2 ssds) and he is also very happy with the performance.

Just look at the performance jump

Kind of same movement here, moved from a GTX 1080 + 9900K to 9070 + 9700X, this thing is a complete beast

 

[AusWolf]

That's the limit. What they also state(and I can't find where) is that optimal nominal operating temps should kept under 80C.

Sure, but that's still Tcase, not Tjunction. Since 9070 cards do not report Tcase, we have no way of knowing how far they are from that number.

However, I don't want to seem antagonizing to you, so I digress.

I don't feel antagonised.

Yes, they can do it 24/7, but how long.. That does NOT mean it is optimal. I'm not blowing messiness out my backside here. I speak from experience and have seen system parts burn themselves out from running at such temps. CPU's, motherboards, PSUs and yes, GPUs. It happens and it's preventable. Those temps are NEVER good for any length of time, regardless if the parts can run at those temps. Just because it can does not mean it should.

I remember when CPUs ran with a max operating temp of 60-65 ˚C. Now even 95 ˚C is fine. Times change, our experiences get challenged by new hardware. Whether 90 ˚C VRAM hotspot on these cards is fine, only time will tell. But considering what we've discussed above, I'm giving AMD and its AIBs some trust on the matter.

Edit: Most importantly, sensor placements change, too.

 

[spacer00ster]

Monster Hunter Wilds is quite inconsistent/buggy not the best indication if a gpu performs well or not, also the game is quite cpu bottlenecked (Just look at your max gpu Clock, holy sh*t)

Yea I see that now. What do you think would be the best benching software? - 3DMark, Furmark, OCCT etc. I find Steel Nomad good to crap on your card settings and send you back to the drawing board.