New GPU hotspot ridiculously higher than core

[Blueyoshi03]

Just got myself a new gpu. Got a 9070xt Nitro+, however with minimal usage i've immediately noticed the hotspot is ridiculous. Within minutes of loading up Icarus I noticed my core temperature barely pushing 50c, usually getting to 51 maybe 52, the fans were going insane and spotted the hotspot temp reaching as high as 90c. This seems unbelievably higher than the core temp and am concerned this is unusual.

Also i've undervolted it a little and noticed the board power is also a lot higher than the tdp of 330w. At 90-100% GPU usage i was seeing a sustained 350w and even spikes above 400w. I'm wondering if this is normal and also how dangerous this may be for the 12 pin connector as the last thing i want to see is that melt after all the 50 series stories.

 

[new boy]

A hot spot delta of 25-30ish is normal for these cards. I would not be happy with 40...

 

[hornetstinger]

FYI my pure is

max
GPU 43
GPU hotspot 67
Memory 76

12 degrees room temp
-100mv undervolt
Case 3x 12cm intake front, 1x12cm intake side panel blowing directly onto GPU, 2x 12cm exhaust.

 

[Rogueywon]

Just got myself a new gpu. Got a 9070xt Nitro+, however with minimal usage i've immediately noticed the hotspot is ridiculous. Within minutes of loading up Icarus I noticed my core temperature barely pushing 50c, usually getting to 51 maybe 52, the fans were going insane and spotted the hotspot temp reaching as high as 90c. This seems unbelievably higher than the core temp and am concerned this is unusual.

Also i've undervolted it a little and noticed the board power is also a lot higher than the tdp of 330w. At 90-100% GPU usage i was seeing a sustained 350w and even spikes above 400w. I'm wondering if this is normal and also how dangerous this may be for the 12 pin connector as the last thing i want to see is that melt after all the 50 series stories.

That smacks to me of improper thermal paste application. I'm not sure what things are like on Team Red, but on Nvidia cards, I'd raise an eyebrow at any delta above 15 degrees.

I wouldn't worry about transient power spikes to 400W or so. Looking around the 9070XT seems to have some fairly large transients, albeit nothing like to the extent of the old 3090 (which was notorious for them). I'd be shocked if you melted any 12 pin connectors at the power levels you're talking about. They're broadly similar to what we saw on the 4080, which didn't suffer the connector problems that its big brother the 4090 had.

 

[Tetras]

Also i've undervolted it a little and noticed the board power is also a lot higher than the tdp of 330w. At 90-100% GPU usage i was seeing a sustained 350w and even spikes above 400w. I'm wondering if this is normal and also how dangerous this may be for the 12 pin connector as the last thing i want to see is that melt after all the 50 series stories.

Sounds consistent with TPU's review. As for the connector.., who knows.

Just got myself a new gpu. Got a 9070xt Nitro+, however with minimal usage i've immediately noticed the hotspot is ridiculous. Within minutes of loading up Icarus I noticed my core temperature barely pushing 50c, usually getting to 51 maybe 52, the fans were going insane and spotted the hotspot temp reaching as high as 90c. This seems unbelievably higher than the core temp and am concerned this is unusual.

Some of the Sapphire cards apparently have foam in them that needs to be removed, not sure if this applies to the Nitro+ (maybe not?), but I assume you had a good look before you installed it?

 

[Blueyoshi03]

Sounds consistent with TPU's review. As for the connector.., who knows.


Some of the Sapphire cards apparently have foam in them that needs to be removed, not sure if this applies to the Nitro+ (maybe not?), but I assume you had a good look before you installed it?

I'm aware that there is a thin strip of foam on the pass through section of the card however it notes it's deliberately put there to protect the fins from the cables in that area.

Other than that there aren't any caveties available to remove foam from so hopefully there's nothing else.

 

[Firegod]

Doesn't sound right to me. I had something very similar with my 6900XT recently. Open a game and hotspot was maxing out. Cracked open the card and it had zero paste left on gpu. Applied new paste, all good. Depends if you are comfortable to check or if you just want to rma.

 

[Blueyoshi03]

If it is abnormal I might check it out. Never had to do anything manual with a brand new card but I modified my previous 1080ti with 2 chassis fans which made an enormous difference, so I'm familiar with taking gpus apart.

It does seem to be about 25-30c hotspot higher than core at any temp when under load so if that's abnormal for a card like this maybe I need to repaste. Tbh I never saw my hotspot temp on my 1080ti but the core regularly held at 90c so can't imagine the same ratio of hotspot could've been happening as 120c seems not too good for longevity.

 

[Tetras]

I'm aware that there is a thin strip of foam on the pass through section of the card however it notes it's deliberately put there to protect the fins from the cables in that area.

Other than that there aren't any caveties available to remove foam from so hopefully there's nothing else.

Yeah, that's fine, you're not supposed to remove that.

It does seem to be about 25-30c hotspot higher than core at any temp when under load so if that's abnormal for a card like this maybe I need to repaste. Tbh I never saw my hotspot temp on my 1080ti but the core regularly held at 90c so can't imagine the same ratio of hotspot could've been happening as 120c seems not too good for longevity.

Afaik, AMD cards do typically have a higher hotspot (delta between core/hotspot) than nvidia cards.

 

[Parr]

If it is abnormal I might check it out. Never had to do anything manual with a brand new card but I modified my previous 1080ti with 2 chassis fans which made an enormous difference, so I'm familiar with taking gpus apart.

It does seem to be about 25-30c hotspot higher than core at any temp when under load so if that's abnormal for a card like this maybe I need to repaste. Tbh I never saw my hotspot temp on my 1080ti but the core regularly held at 90c so can't imagine the same ratio of hotspot could've been happening as 120c seems not too good for longevity.

Your choice, but I'd be wary of taking a new card apart to investigate the paste (or TIM on the case of the nitro) as it will void the warranty. I've heard reports that a high delta is usual for these cards, and temperatures up to 95 on the memory is not that uncommon, but if you're not happy RMA it (though I appreciate it is a giant pain in the arse with the current shortages of replacement cards)

 

[Major774]

Does sound like a thermal paste issue.

I had a reference RX 6950 XT with high hotspot temps, which made the fans spin faster than they needed to. The replacement card had the same issue, so I ended up removing the factory graphite pad and slapping some thermal paste in there.

 

[Alt0153]

every review ive seen about the new amd card has a 25c+ hotspot difference

 

[Firegod]

I vote @Blueyoshi03 does a tear down and shares findings.

 

[Kirby Wurm]

Bloody hell, my brother has one and just advised his delta is 22-25°c! If the ~40°c is consistent then that's unusually high.

You may have luck trying to sort it yourself, and with the current scarcity I can understand doing so, though a return may be a safer bet: I'm not sure if Sapphire cover cooler replacement with their warranty.

I may just have PTSD from the experience of repasting my 6800XT Nitro+ and seeing good results at load but far worse ideal temps, so feel free to ignore my thousand yard stare.

 

[Blueyoshi03]

At the moment I've undervolted, lowered the power limit to 0%, and underclocked slightly and it's performing possibly 5% slower than stock but the hotspot and core are dramatically cooler and the board power is maybe 100w lower on average.

The delta has calmed down too. Still about 20-25c. I never like my gpu screaming at it's absolute limits but I'll see how it performs going forward and possibly consider later on if I'll try to repaste. Does look like a particularly awkward card to breakdown.

 

[Foliedelanuit]

It took me ages to realise the fan tuning/custom curve in Adrenalin uses the hotspot rather than core for the temperature value. Can see why it would have had your fans firing away

 

[teejay73]

My red devil 6950 xt also suffered from a high hotspot delta of 40c, and given that I'd already put a water block on it, I stripped it down and applied some fresh paste on it, and the delta dropped to 15c-20c. Obviously, as has already been said you'll probably void your warranty by stripping it down, but it may be your current option. I guess a RMA might be a fruitless task if they can't supply a replacement straight away.

 

[g67575]

For now, it’s probably not worth messing about with thermal paste.

Cards run hot these days, and both AMD and Nvidia amp up the GPU core clock speed beyond what is sensible or efficient, requiring high voltages.

Try lowering the max GPU core clock to 2870Mhz or 2800 Mhz. Max out the fans under load. Undervolting is always a good idea, but I personally think going much below the auto undervolt config, yields limited results.

Edit - I looked at the boost clock for your card, and it’s over 3ghz!

Sapphire NITRO+ RX 9070 XT Specs

AMD Navi 48, 3060 MHz, 4096 Cores, 256 TMUs, 128 ROPs, 16384 MB GDDR6, 2518 MHz, 256 bit

www.techpowerup.com

That may be the source of the problem. Try 2900Mhz max, then 2800Mhz.

I can’t find any precise figures, but keeping all components on your graphics card at or below 90 Celsius under sustained usage, seems sensible.

At least that keeps you 20 degrees C below the typical thermal throttle point.

Thoughts - I wish AMD’s software would allow custom throttling hotspot temperatures (lower, rather than higher). Being able to set this at 90 or 95 degrees C would help to protect the lifespan of graphics cards.