Is RX 9070 VRAM temperature regular value or hotspot?

[freeagent]

If AMD says it’s ok, then it has to be ok.. right? I am no engineer.. but how does the other team get their ram to run 30c cooler is beyond my pay grade.

 

[TSiAhmat]

If AMD says it’s ok, then it has to be ok.. right? I am no engineer.. but how does the other team get their ram to run 30c cooler is beyond my pay grade.

Is the vram temps on the 50 Series Hot-spot or average/something else?

 

[TechBuyingHavoc]

Maybe. Or maybe this guy in the centre.

Awesome film, by the way, highly recommended.

The Menu is such a great movie. Fantastic satire for anyone who enjoys fine dining.

How are people with the 9070 XT changing their fan curves? Is the common wisdom something close to what this Reddit post says? I will summarize it for those who don't want to click the link.

Adrenalin Settings:
· Max Frequency Offset: -500 MHz
· Voltage Offset: -85 mV
· VRAM Memory Timing: Fast Timing
· VRAM Max Frequency: 2700 MHz
· Power Limit: -30%

Cyberpunk 2077, Hell Let Loose, theHunter: Call of the Wild, Kingdom Come: Deliverance II, and Marvel Rivals is the games tested. Resolution was 2560×1440 on max settings (no FSR or frame generation).

The OP says that it leads to a 4.5% performance loss with actual FPS (6% loss on 3DMark and 25% loss on Furmark benchmarks) but the power usage drops by >30%.

 

[RejZoR]

A question has been asked to HWiNFO author in this thread about RX9000 VRAM temp reading

Post #4

GIGABYTE RX 9070 XT ELITE HWINFO SUPPORT COMING?

Hi Martin, I just wanted to confirm if there is a future release that will support the rx 9070 xt? I didnt see it in the future upcoming changes. Thanks

www.hwinfo.com

Martin didn't respond with exact mention of "hotspot", but he did mention highest temperature which is what hotspot is suppose to be. So, hotspot it is apparently.

 

[Event Horizon]

Sounds like the temps are well within spec for most models.

If I'm not mistaken, the 5070 Ti's max VRAM temp is 88C, whereas the 9070's max VRAM temp is 108C (both XT and non-XT). I wish vendors would offer individual memory module temps for advanced users.

Anyway, for long term longevity (far beyond the warranty), it's probably a good idea to: adjust the fan curve, tweak voltage offset and power limit, improve case ventilation, cap the FPS.

 

[rusty caterpillar]

it has it's the same conversation. Someone designed a stupid cable, and Nvidia and AMD are using it.
What isn't smart is claiming "they know what they are doing, so it must be fine", that's some delusional stuff

Is not just the cable, power gate to is problematic.

 

[Bomby569]

Who talked about a cable here? The topic is about safe VRAM temperatures. (So far), there's no indication of anyone's graphics card catching fire, or killing itself in any other way due to high VRAM temps, unlike the cable example, of which there's plenty of.

the talking was about engineers know what they are doing. They clearly don't most of the time.

 

[rusty caterpillar]

Sounds like the temps are well within spec for most models.

If I'm not mistaken, the 5070 Ti's max VRAM temp is 88C, whereas the 9070's max VRAM temp is 108C (both XT and non-XT). I wish vendors would offer individual memory module temps for advanced users.

Anyway, for long term longevity (far beyond the warranty), it's probably a good idea to: adjust the fan curve, tweak voltage offset and power limit, improve case ventilation, cap the FPS.

By "max VRAM temp" you mean safe temp threshold?
GDDR6 present in 9000 series are recommended at 95 C works on 1.35V while GDDR7 works on 1.2 V and got roughly the same temp threshold.

For both I would set my own threshold to 85 C but, is no way I'm gonna run any of this cards at 95 C VRAM temps. Why would I play some silly lottery?

Nvidia released cards for decades without cables burning

Someone designed a stupid cable, and Nvidia and AMD are using it.

Sry I quoted wrong message and just added the fact that burning cables and stupid design is just the tip of the iceberg , the power gate on the PCB is more problematic on long term IMO.

 

[Event Horizon]

By "max VRAM temp" you mean safe temp threshold?
GDDR6 present in 9000 series are recommended at 95 C works on 1.35V while GDDR7 works on 1.2 V and got roughly the same temp threshold.

For both I would set my own threshold to 85 C but, is no way I'm gonna run any of this cards at 95 C VRAM temps. Why would I play some silly lottery?

I agree on staying below 85 C. That's a good rule of thumb. My source for the 9000 series is this hwinfo thread: https://www.hwinfo.com/forum/threads/vram-thermal-limit-on-9070-xt.10391/

Someone with a 5070 ti could perhaps chime in and let us know what they get for this value.

 

[rusty caterpillar]

I agree on staying below 85 C. That's a good rule of thumb. My source for the 9000 series is this hwinfo thread: https://www.hwinfo.com/forum/threads/vram-thermal-limit-on-9070-xt.10391/

Someone with a 5070 ti could perhaps chime in and let us know what they get for this value.

I have no clue what HWinfo reads from that card and why is set to 108C... when the known threshold GDDR6 is 95C, maybe is a different new revision of GDDR6 or AIB mistakenly set the threshold, IDK We have very little data atm, till a clear more image is 85 C max in my book.

 

[AusWolf]

I have no clue what HWinfo reads from that card and why is set to 108C... when the known threshold GDDR6 is 95C, maybe is a different new revision of GDDR6 or AIB mistakenly set the threshold, IDK We have very little data atm, till a clear more image is 85 C max in my book.

Because 108 °C is the max hotspot temp and 95 °C is tJmax. At 90 °C hotspot (what you see in GPU-Z), your junction temp is way lower. You're basically 18 °C below the max hotspot temp.

 

[Zach_01]

I was sure the Samsung GDDR6 chips on my XTX have a rated TJmax of at least 105C, likely 110C.
Totally forgot that apparently not all GDDR6 are not created equal. Oops.

Looking at OP’s card and a few 7900XTX and 9070XT models, they use the same model of Hynix chips according to TPU’s reviews.
Now, that exact model of VRAM chips is rated at 0-85C, but is it the TJmax? Is the source reliable? The review here for AsRock Taichi 9070XT has VRAM temps gone as high as 90C.
I would love to check how other 7900XTX models fare, but the reviews here don’t show VRAM temps for the XTXs.

That said, the “all these cards should have been dead” point still stands, but I feel the ground here is a bit shaky now.

7900XTX Nitro+ here. Has Samsung chips.
According to HWiNFO readings at 78C VRAM temp the thermal limit is around 72%. By extrapolation the max limit is at ~110C.

At full load TBP 375W while VRAM runs at 2600MHz, temp is max at 78-80C, avg around 72C and fans rpm avg 1900-2000 (max rpm 3200).
The card’s air feed is 22C for the above thermals.
And don’t forget the cooler size on XTX Nitro+.
It’s ~3.5 slots.

Said it already but I guess it’s needed again.
A 300+W card needs some serious airflow through it to keep any hotspot temp under 85C while inside a case. Also very good case ventilation.

AIBs are trying to keep noise at minimum for marketing purposes and so they choose to run thermals closer to limit than most users are comfortable with. Hence the 1000-1300rpm that most 9070s run.

Use your adrenaline to set your fan curve/noise preference. End of story…
At this point the subject is over discussed I believe.

 

[rusty caterpillar]

Engineers know what to do, but marketing screws everything up and this has been well known for years.

"Engineers know what to do" but are forced or coerced to do something else, cause low budget and self embellishment logos on big plastic shrouds has to be big enough to cover the exhaust of the heatsinks.
I understand Marketing but, hurting the product on the long run is not Marketing rather, planned obsolescence.

​

7900XTX Nitro+ here. Has Samsung chips.
According to HWiNFO readings at 78C VRAM temp the thermal limit is around 72%. By extrapolation the max limit is at ~110C.

At full load TBP 375W while VRAM runs at 2600MHz, temp is max at 78-80C, avg around 72C and fans rpm avg 1900-2000 (max rpm 3200).
The card’s air feed is 22C for the above thermals.
And don’t forget the cooler size on XTX Nitro+.
It’s ~3.5 slots.

Said it already but I guess it’s needed again.
A 300+W card needs some serious airflow through it to keep any hotspot temp under 85C while inside a case. Also very good case ventilation.

AIBs are trying to keep noise at minimum for marketing purposes and so they choose to run thermals closer to limit than most users are comfortable with. Hence the 1000-1300rpm that most 9070s run.

Use your adrenaline to set your fan curve/noise preference. End of story…
At this point the subject is over discussed I believe.

View attachment 390774

Is not over discussed IMO.
They haven't done much in designing the thermal solutions. However I gonna run it quiet and low VRAM temps and I'll have to mod the shit out of it and spend some more money.

My GDDR5X hottest VRAM sits at 56 C when GPU is 60 C and H Spot is 74 C. 3x 92mm @1400 +1 x92mm @ aprox 800-1000 Rpm.
See for yourself pwr draw 316 W max avg 256W



Not sure I've seen my VRAM going to 65 C yet.

How can I without AC and an air cooled card in a shitty modified Lian Li able to achieve that and AIB can't??
Show me a stock air cooled 1080 Ti running that VRAM temps out of the box. I personally don't know but, if you found one I'll be very interested in that particular AIB.

7900XTX Nitro+ here. Has Samsung chips.
According to HWiNFO readings at 78C VRAM temp the thermal limit is around 72%. By extrapolation the max limit is at ~110C.

At full load TBP 375W while VRAM runs at 2600MHz, temp is max at 78-80C, avg around 72C and fans rpm avg 1900-2000 (max rpm 3200).
The card’s air feed is 22C for the above thermals.
And don’t forget the cooler size on XTX Nitro+.
It’s ~3.5 slots.

Said it already but I guess it’s needed again.
A 300+W card needs some serious airflow through it to keep any hotspot temp under 85C while inside a case. Also very good case ventilation.

AIBs are trying to keep noise at minimum for marketing purposes and so they choose to run thermals closer to limit than most users are comfortable with. Hence the 1000-1300rpm that most 9070s run.

Use your adrenaline to set your fan curve/noise preference. End of story…
At this point the subject is over discussed I believe.

View attachment 390774

As long as I know 9070XT XFX Merc is the heaviest also expensive but, is louder than others and not impressive on cooling followed by Sapphire Nitro with the silly secondary back plate and no TIC in between.

 

[AusWolf]

"Engineers know what to do" but are forced or coerced to do something else, cause low budget and self embellishment logos on big plastic shrouds has to be big enough to cover the exhaust of the heatsinks.
I understand Marketing but, hurting the product on the long run is not Marketing rather, planned obsolescence.

View attachment 390775

Those are useless, definitely. But those are a couple of cards with a bad design from an AIB. The thing is, VRAM runs hot on almost every 9070 XT. If someone wants to tell me that every single AIB botched every single 9070 XT, and they're all gonna fail soon, go ahead.

 

[Zach_01]

My GDDR5X hottest VRAM sits at 56 C when GPU is 60 C and H Spot is 74 C. 3x 92mm @1400 +1 x92mm @ aprox 800-1000 Rpm.
See for yourself pwr draw 316 W max avg 256W

View attachment 390776View attachment 390777

Not sure I've seen my VRAM going to 65 C yet.

This is an EVGA 1080Ti ICX?

 

[freeagent]

When you look at something expensive that you bought, you are supposed to take pleasure in the little details, they are supposed to give warm and fuzzies

 

[TSiAhmat]

If it really is 108 C° on Vram than we are far off any thermal throttling/damage... Wish there was an official statement from aibs or AMD that specified what temps are appropriate for the card.

 

[JustBenching]

Who talked about a cable here? The topic is about safe VRAM temperatures. (So far), there's no indication of anyone's graphics card catching fire, or killing itself in any other way due to high VRAM temps, unlike the cable example, of which there's plenty of.

High vram temps doesn't kill the vram over night. Not saying the vram temps of the 9070 is a problem, im just saying that if it is, it won't die in a couple of days. It slowly degrades (youll be hitting lower and lower memory speeds as time goes on) and at the end it will artifact even when you underclock it. My 3090 had a similar issue, I had to drop my memory OC every couple of weeks cause memory was constantly running at 90+. At some point I just swapped the thermal pads and all was fine.

 

[AusWolf]

High vram temps doesn't kill the vram over night. Not saying the vram temps of the 9070 is a problem, im just saying that if it is, it won't die in a couple of days. It slowly degrades (youll be hitting lower and lower memory speeds as time goes on) and at the end it will artifact even when you underclock it. My 3090 had a similar issue, I had to drop my memory OC every couple of weeks cause memory was constantly running at 90+. At some point I just swapped the thermal pads and all was fine.

Ah, I get 'ya. With that, we'll see, I guess.

 

[rusty caterpillar]

This is an EVGA 1080Ti ICX?

Yes FTW 3, 3 sensors on each memory group.
I bought it second hand and in just 5 min FurM test one group Mem3 went to 71C and some power to 76C.

Superposition uses way more VRAM than FurM, yet my MEM3 doesn't go not even on 60 C, same mem group that went 71 C with old leaky pads in Furmark 5 min test
Thermal putty + fan on the back + removed grill for rear case fan etc all done by me, not Lian Li or EVGA.
Still have one issue to resolve card sits 3cm away from glass side panel, thanks to Lian Li design for XL cases, recirculating hot air. 18CM width on main case chamber is just a SAD joke, is like a prebuilt cheap case

 

[Zach_01]

Yes FTW 3, 3 sensors on each memory group.
I bought it second hand and in just 5 min FurM test one group Mem3 went to 71C and some power to 76C.

Superposition uses way more VRAM than FurM, yet my MEM3 doesn't go not even on 60 C, same mem group that went 71 C with old leaky pads in Furmark 5 min test
Thermal putty + fan on the back + removed grill for rear case fan etc all done by me, not Lian Li or EVGA.
Still have one issue to resolve card sits 3cm away from glass side panel, thanks to Lian Li design for XL cases, recirculating hot air. 18CM width on main case chamber is just a SAD joke, is like a prebuilt cheap case

The profound difference between EVGA iCX cards and RDNA3/4 cards is that EVGA reports PCB temperatures near(on the side of) some VRAM modules while RDNA3/4 cards report hotspot of junction temperature and those sensors are located exactly underneath the VRAM modules, which is the closest point to the VRAM silicon that gets hot. Any closer than this you are inside the chip.

It’s apples and oranges that are not debatable for comparison.
The real difference between those 2 cases can be more than 20 degrees C.

 

[dfg555]

What's with people's obsession with VRAM Temps or Temps in general unless it's LN2 OC'ing Competition. Ever since monitoring programs started implementing, I keep seeing just pointless posts about it. Go play your games and send your GPU into warranty if it dies. 100c is hot for you/humans, not for semiconductors lol.

 

[rusty caterpillar]

The profound difference between EVGA iCX cards and RDNA3/4 cards is that EVGA reports PCB temperatures near(on the side of) some VRAM modules while RDNA3/4 cards report hotspot of junction temperature and those sensors are located exactly underneath the VRAM modules, which is the closest point to the VRAM silicon that gets hot. Any closer than this you are inside the chip.

It’s apples and oranges that are not debatable for comparison.
The real difference between those 2 cases can be more than 20 degrees C.

View attachment 390845

I don't know about "oranges" much yet, I don't have it and if the sensors are under VRAM, I'm happy because are more accurate, however I'm confident AIB cut more corners compared to the past when comes to cooling.

But back to the "apples" 1080 Ti
How come a difference of less 10 C to GPU 8C to Mem3, 6c to Mem2 and 8C to PWR4 sensors?
If AIB does closest to perfection cooling solutions, how come with just different TIC and a silly fan on the back you can improve that much the temps. Still don't think the gap is to wide?

Another remark, if 9000 series has sensors under VRAM than the heat traversing PCB, cooled from the other side of the GPU might have a greater impact to sensors AVG temps IMO. But will see.




Initial test bellow 5 min Furmark , and I will make another test in about 1 week with same length.

What's with people's obsession with VRAM Temps or Temps in general unless it's LN2 OC'ing Competition. Ever since monitoring programs started implementing, I keep seeing just pointless posts about it. Go play your games and send your GPU into warranty if it dies. 100c is hot for you/humans, not for semiconductors lol.

Because VRAM tends to fry and not throttle down as the GPU core, also VRAM temps can be influenced by other components. 1 single chip goes your card is bricked. Play the lottery if that is your wish some of us don't.

 

[AusWolf]

What's with people's obsession with VRAM Temps or Temps in general unless it's LN2 OC'ing Competition. Ever since monitoring programs started implementing, I keep seeing just pointless posts about it. Go play your games and send your GPU into warranty if it dies. 100c is hot for you/humans, not for semiconductors lol.

I guess they're stuck in 2012 when your 32 nm CPU had a 65 °C tJmax.

 

[r.h.p]

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.

i have to agree i run GPU fans at a ..lol pretty high rpm in adrenalin even at desktop. My CPU is set to turbo in Bios.