Is RX 9070 VRAM temperature regular value or hotspot?

[RejZoR]

I'm running XFX RX 9070 XT Mercury OC and in more demanding games like Dying Light 2 with ray tracing fully on, VRAM temperature is 92°C which is unreasonably high for such an overbuild card with massive cooler that's also touching VRAM modules. Granted, it's not full contact, but even with partial like it is on Mercury OC, it should be lower.

I heard from someone that AMD changed VRAM temperature reporting to now be hotspot by default instead of regular temperature. Anyone has any concrete info to confirm that? All I can find online are reports of "unreleased AMD card with very high VRAM temperatures" and "RX 9070 having issues too" reports all over news sites and nothing else. If VRAM temperature sensor is now indeed reporting hotspot by default for VRAM, that would explain all those leaked reports and it would also explain why VRAM temperature is so high even on Mercury OC. But I don't know for sure and I can't find any reliable resources for it online. Any reliable info on this matter would be highly appreciated.

 

[AusWolf]

Based on the fact that the limit is set to 108 ˚C by default on many cards, and Micron specifies a max operating temp of 95 ˚C, I assume it's hotspot.

 

[droopyRO]

VRAM temperature is 92°C

What rpm are your GPU fans running at ? A small bump of 300 rpm might be the solution. Also try with the side panel off to see if your case cooling is adequate.

 

[Epaminombas]

Up to 95C is perfectly normal for CPU and GPU to operate, but if it goes above 98C then it could be a problem.

 

[R-T-B]

I'm running XFX RX 9070 XT Mercury OC and in more demanding games like Dying Light 2 with ray tracing fully on, VRAM temperature is 92°C which is unreasonably high for such an overbuild card with massive cooler that's also touching VRAM modules. Granted, it's not full contact, but even with partial like it is on Mercury OC, it should be lower.

I heard from someone that AMD changed VRAM temperature reporting to now be hotspot by default instead of regular temperature. Anyone has any concrete info to confirm that? All I can find online are reports of "unreleased AMD card with very high VRAM temperatures" and "RX 9070 having issues too" reports all over news sites and nothing else. If VRAM temperature sensor is now indeed reporting hotspot by default for VRAM, that would explain all those leaked reports and it would also explain why VRAM temperature is so high even on Mercury OC. But I don't know for sure and I can't find any reliable resources for it online. Any reliable info on this matter would be highly appreciated.

OT but holy shit longtime no see man.

 

[boppingXQWB]

It's asinine how warm many 9070 XT fan curves seem to tuned. Both GPU hot spot and VRAM temps are crazy high this gen. It's not even summer yet and if your VRAM is hitting 92, hot spot or not, I would bump that fan speed if I were you.

(Tinfoil hat) I wonder if this is some sort of planned obsolescence move to cook VRAMs to death by the time warranty end. Users will be encouraged to upgrade when that happens, since GPU improvements are rolling out slower and people are holding onto their cards longer. Just 'cause VRAM manufacturer specs state you are OK to run at 95, it doesn't mean you should be running against it under normal use when a couple hundred RPMs that make no appreciable difference to the computer's noise profile can mitigate it.

 

[freeagent]

(Tinfoil hat)

Those are terrible temps, speaking as an Nvidia user. Those temps are out of control, and just plain lazy.

Gross.

 

[Zach_01]

I'm running XFX RX 9070 XT Mercury OC and in more demanding games like Dying Light 2 with ray tracing fully on, VRAM temperature is 92°C which is unreasonably high for such an overbuild card with massive cooler that's also touching VRAM modules. Granted, it's not full contact, but even with partial like it is on Mercury OC, it should be lower.

I heard from someone that AMD changed VRAM temperature reporting to now be hotspot by default instead of regular temperature. Anyone has any concrete info to confirm that? All I can find online are reports of "unreleased AMD card with very high VRAM temperatures" and "RX 9070 having issues too" reports all over news sites and nothing else. If VRAM temperature sensor is now indeed reporting hotspot by default for VRAM, that would explain all those leaked reports and it would also explain why VRAM temperature is so high even on Mercury OC. But I don't know for sure and I can't find any reliable resources for it online. Any reliable info on this matter would be highly appreciated.

Fan control of any GPU is bind to GPU temp. If GPU temp is staying relatively low then fans drop rpm.
I’m not sure if on all the 9070series the fan control is on the “GPU temp” or the “GPU Hotspot temp”.
For example on 7900XTX fan control is on GPU Hotspot temp. I’ve set a custom fan curve from adrenaline in such a way so that VRAM temp peaks at low 80s (~82C). VRAM temp limit is 108C.

I suggest to set a custom fan curve, a bit more aggressive from stock.

 

[AusWolf]

Fan control of any GPU is bind to GPU temp. If GPU temp is staying relatively low then fans drop rpm.
I’m not sure if all the 9070series the fan control is on the “GPU temp” or the “GPU Hotspot temp”.
For example on 7900XTX fan control is on GPU Hotspot temp. I’ve set a custom fan curve from adrenaline in such a way so that VRAM temp peaks at low 80s (~82C). VRAM temp limit is 108C.

I suggest to set a custom fan curve, a bit more aggressive from stock.

Fan speed seems to be bound to hotspot on my Powercolor Reaper.

 

[blinnbanir]

This is purely anecdotal but AMD seems to have turned their Hardware up to another level. I bought a 9900X3D for $150 less than the 9950X3D and I am absolutely amazed at how fast this chip is. It seems the same thing was applied to these 9000 GPUs. They are very impressive. If you have one congratulations. What Japanese Gamers already realize is that PS5 Games are being developed on Ryzen and Radeon. The PS5 is like a 7700XT with 16GB, so anything faster would be felt. For there to be a 45% market share in Japan already more than just word of mouth obviously made these GPUs popular. The temp issue for me is indeed the shroud design and PCB reuse by GPU vendors so that things are not optimized. I mean it should be criminal for Memory to run at 90+ C on a GPU that is closer to 3 slots than 2.

 

[Zach_01]

Fan speed seems to be bound to hotspot on my Powercolor Reaper.

And what is the GPU Hotspot, VRam temp and fan % rpm on heavy load?
Thanks in advance!

 

[NattyKathy]

Seems like most/all Navi48 users are experiencing this from what I've seen. There's some speculation that it could be a thing with the epoxy Hynix uses for the packaging since the 20GTs Hynix on RDNA3 boards also runs kinda hot (but not as bad as 9070)

I haven't seen specific info on the sensor layout on GDDR6 chips but I kinda doubt they're using a sensor network like modern GPUs and CPUs. More likely there's a single temp sensor strategically placed in each module and the memory temp reported by the driver is the highest among all the chips.

 

[AVATARAT]

Only undervolt + additional cooling works here, + good airflow in the case.

 

[Assimilator]

VRAM temperature is 92°C which is unreasonably high for such an overbuild card with massive cooler that's also touching VRAM modules

According to who? You?

Are you the engineer who designed the card? No? Then why do you think you know better than that engineer as to what is "unreasonably high" or not?

This is exactly the reason why NVIDIA removed the hotspot sensor on the 5000 series, because uninformed users who think they know better than engineers whined incessantly about it without bothering to understand it. Considering AMD had the same problem with Zen 4 and 5 and idiot users, I'm honestly surprised they didn't follow NVIDIA's lead.

And no, I am not advocating for removing sensors, but unfortunately the world has more people suffering from Dunning-Kruger than people who are willing and capable of rubbing two brain cells together. So NVIDIA optimised for the lowest common denominator.

 

[Vayra86]

According to who? You?

Are you the engineer who designed the card? No? Then why do you think you know better than that engineer as to what is "unreasonably high" or not?

This is exactly the reason why NVIDIA removed the hotspot sensor on the 5000 series, because uninformed users who think they know better than engineers whined incessantly about it without bothering to understand it. Considering AMD had the same problem with Zen 4 and 5 and idiot users, I'm honestly surprised they didn't follow NVIDIA's lead.

And no, I am not advocating for removing sensors, but unfortunately the world has more people suffering from Dunning-Kruger than people who are willing and capable of rubbing two brain cells together. So NVIDIA optimised for the lowest common denominator.

Engineers don't work in a vacuum; they work in a world where a product needs to meet corporate targets.

Due diligence applies here, and no you can't just trust any company's 'engineers' to know best.
We have a product history full of fuckups, from batteries catching fire, to melting connectors, to coolers simply not even touching the GPU die proper.

While I don't disagree the vast majority of users overreact to temps.... security through obscurity really isn't the way here. Planned obscolescence is real and yes we have a right to know what our product is doing under the hood. We bought it, after all.

 

[AusWolf]

Engineers don't work in a vacuum; they work in a world where a product needs to meet corporate targets.

Due diligence applies here, and no you can't just trust any company's 'engineers' to know best.
We have a product history full of fuckups, from batteries catching fire, to melting connectors, to coolers simply not even touching the GPU die proper.

While I don't disagree the vast majority of users overreact to temps.... security through obscurity really isn't the way here. Planned obscolescence is real and yes we have a right to know what our product is doing under the hood. We bought it, after all.

The reason I agree with @Assimilator on this one is that nearly every single 9070 (XT) seems to be running high VRAM temps. I can't imagine that every single AIB botched every single card.

 

[AVATARAT]

According to who? You?

According to this as example:

If this happens in a well ventilated case, what do you think would happen in a normal consumer case with enough dust?

And yes, the industry, as always in recent years, is just making "things" that will help the device work as little as possible after the warranty ends.

 

[Vayra86]

The reason I agree with @Assimilator on this one is that nearly every single 9070 (XT) seems to be running high VRAM temps. I can't imagine that every single AIB botched every single card.

So educate users with a convincing article why the temp is what it is, why you report it, and why there is no problem.

Its how you build trust. You can't on one hand say people are stupid and on the other never educate them so they remain stupid. True stupidity is the unwillingness or incapability to learn.

 

[Event Horizon]

I have a 10 year old gtx 970 and a 4 year old rtx 3060 ti. Both are fully functional, in regular use, and free of issues. I'd like to get similar usage out of my 9070. I will be keeping a close eye on the temps (especially throughout the summer).

 

[Assimilator]

So educate users with a convincing article why the temp is what it is, why you report it, and why there is no problem.

This education happened in literally every review of a Zen 4 CPU. It happens on the product page for those CPUs. Yet we still got, and I am sure will get in future, multiple threads of idiots asking "why my cpu tmprture so highhh?" when they could've got that answer with a single Google search. That is the type of advanced stupidity that companies fight against every day.

True stupidity is the unwillingness or incapability to learn.

Which describes most consumers. Therefore it makes the most business sense for any company to optimise for the majority case of "don't scare the idiots", by not exposing scary temperatures at all. You can't fix stupid, but you can design products that don't trigger it as often.

 

[RejZoR]

According to who? You?

Are you the engineer who designed the card? No? Then why do you think you know better than that engineer as to what is "unreasonably high" or not?

This is exactly the reason why NVIDIA removed the hotspot sensor on the 5000 series, because uninformed users who think they know better than engineers whined incessantly about it without bothering to understand it. Considering AMD had the same problem with Zen 4 and 5 and idiot users, I'm honestly surprised they didn't follow NVIDIA's lead.

And no, I am not advocating for removing sensors, but unfortunately the world has more people suffering from Dunning-Kruger than people who are willing and capable of rubbing two brain cells together. So NVIDIA optimised for the lowest common denominator.

That's a rather stupid assumption. I've worked with computers and especially cooling for long enough to know a thing or two about cooling, especially since I've been modding the cooling solutions for like 25 years now, with total redesigns in recent years where I've re-engineered entire coolers to make them more effective.

Also there is a great difference not knowing and knowing what is normal temperature and what is hotspot temperature. I know what each means, but if I just get presented with a temperature value with no information if that's a hotspot temperature or regular, that's a rather important, critical difference I have to question. VRAM running at 92°C measured normal way is very poor. VRAM running at 92°C measured from hotspot sensor, totally different thing. So, excuse me if I'm being pedantic and wanting to know what kind of measurement AMD is doing on VRAM modules on RX 9070 XT.

You're talking about regular users who have no idea what hotspot even means and then screaming how temperatures are high. I'm not that regular user. I know what hotspot is. What I don't know is what measurement AMD is using. GPU-Z states just regular temperature with no mention of hotspots. Igor from Igor's Labs has once mentioned that AMD is using hotspot measurement. Apparently. But I have no solid confirmation on that.

 

[BoggledBeagle]

Somewhat connected to the subject - I find the current trend of silence above all to be perverse.

You have a graphic card working hard at 330W - why should it be dead silent?! What a dumb idea.

Some options here:

• a silent card with baking RAM at 90°C and 1200 RPM fans
• a slightly noisy card with RAM at 80°C and 2000 RPM fans
• more noisy card with RAM at 75°C and 2500 RPM fans

Out of these options, I would NEVER choose option 1.

Every single machine or home appliance is expected to make some noise, when it operates. Why should a computer component consuming hundreds of watts be dead silent? What is so wrong about some weak wind noise?

I am missing a third option on the BIOS switch - good cooling.

 

[Zach_01]

According to this as example:

View attachment 390335

If this happens in a well ventilated case, what do you think would happen in a normal consumer case with enough dust?

And yes, the industry, as always in recent years, is just making "things" that will help the device work as little as possible after the warranty ends.

Look at those speeds people...
Anything below 1500 at full load is too conservative for my taste.
Dont get me wrong… I’m not saying work them at 2500.
But 1000-1300rpm ain’t cutting it for 300-350W no matter how big is the cooler.

Adrenalin software is also a tool.
Use it!

Play around with your custom fan curve at the point you feel the noise desired, tolerable or whatever.
Find your personal balance between noise and temps.
its that simple


Edit:
typo

 

[boppingXQWB]

Somewhat connected to the subject - I find the current trend of silence above all to be perverse.

You have a graphic card working hard at 330W - why should it be dead silent?! What a dumb idea.

Some options here:

• a silent card with baking RAM at 90°C and 1200 RPM fans
• a slightly noisy card with RAM at 80°C and 2000 RPM fans
• more noisy card with RAM at 75°C and 2500 RPM fans

Out of these options, I would NEVER choose option 1.

You don't even need to bump RPM to over 2x for that. A couple/few hundred additional RPMs should do, which is why I find it so incredulous how warm the fan curves are tuned.

While trying to intentionally cook my XTX to 90C+ (there's a thread explaining why in this very forum), I noticed that going from a reasonable 1500ish RPM to the minimum-allowed fan speed on quiet vBIOS (something like 1200) made the hot spot/vram spike from less than 85 or so to 95C+: numbers are not exact as I did not record them, but it should be thereabouts.

9070 XT owners who have tuned their fan curves should be able to chime in on the exact (and probably small) change they had to make to improve the temps.

 

[AVATARAT]

Look at those speeds people...
Anything below 1500 at full load is to conservative for my taste.
Dont get me wrong… I’m not saying work them at 2500.
But 1000-1300rpm ain’t cutting it for 300-350W no matter how big is the cooler.

Adrenalin software is also a tool.
Use it!

Play around with your custom fan curve at the point you feel the noise desired, tolerable or whatever.
Find your personal balance between noise and temps.
its that simple
View attachment 390354

I don't have a problem with my 9070 XT but I know how to mod it.