Is RX 9070 VRAM temperature regular value or hotspot?

[Knight47]

I haven't noticed anything like that happening. Does this ramping up last long enough to be seen in the performance overlay, or do I need to put my ear to the case?

It spins at 2000+rpm for a while until the gpu cools down a bit and then reduces the fan speed. So you can absolutely notice it and gpu-z can record it too.

 

[INSTG8R]

I’m seeing a lot of users reporting high VRAM temps. It seems to be across the board. And as @AusWolf posted very close to their limits.
Definitely concerning over the long term because there’s really nothing the user can do other maybe replace the thermal pads with something “premium“ even then you might knock 5C off if you’re lucky…it would take a design change to have any serious temperature improvement…

 

[rusty caterpillar]

It spins at 2000+rpm for a while until the gpu cools down a bit and then reduces the fan speed. So you can absolutely notice it and gpu-z can record it too.

Keep the fans fixed on 1300 or 1400 RPM, see how is going, when GPU enters the load is already cool and will take longer to reach high temps.

I keep my 1080 Ti 1300 RPM all the time so the load is starting at temps like GPU 23 C and VRAM 31 C.

 

[AusWolf]

And as @AusWolf posted very close to their limits.

I actually posted the opposite. 90 ˚C is quite far from the 108 ˚C BIOS limit.

If you mean Micron's 95 ˚C max operating temp, we can't compare to that, because that's edge temp, while these cards only report hotspot temp. With a 90 ˚C hotspot, your edge temp is way lower.

 

[tugrul_LordOfDrinks]

I actually posted the opposite. 90 ˚C is quite far from the 108 ˚C BIOS limit.

If you mean Micron's 95 ˚C max operating temp, we can't compare to that, because that's edge temp, while these cards only report hotspot temp. With a 90 ˚C hotspot, your edge temp is way lower.

Run memtest_vulkan for 5 minutes. If it doesnt fail, use your card for 5 years. If it fails in 1 minute, then sell it after 1 year.

 

[AusWolf]

Run memtest_vulkan for 5 minutes. If it doesnt fail, use your card for 5 years. If it fails in 1 minute, then sell it after 1 year.

I'm not sure running such an unrealistic load is a good idea. I'm not a fan of Furmark, either, for example.

 

[tugrul_LordOfDrinks]

I'm not sure running such an unrealistic load is a good idea. I'm not a fan of Furmark, either, for example.

If your gpu can not run software that just computes something, then it has defects or installation failures such as missing heat gel or thermal paste or something.

People multiply matrices or do convolutions in datacenters and these reach close to peak compute performance and maximum heat output. Same for gaming at 200% scaling & sampling with highest quality and all settings maxed such as the game aliens fire team elite or elex 2.

The point of furmark is testing the stability without waiting for years.

The point of memtest_vulkan is testing if your memory is producing errors or is running with expected performance under memory load.

I tested my 5070 for 5 minutes, it reached 90Celcius and did not produce errors even with +2000 mem overclock. Fan speed was 55%. No game will push the memory this far ever. So I can say 5 years is good for memory. Ofcourse running memtest everyday is bad. Its pointless.

 

[AusWolf]

If your gpu can not run software that just computes something, then it has defects or installation failures such as missing heat gel or thermal paste or something.

People multiply matrices or do convolutions in datacenters and these reach close to peak compute performance and maximum heat output. Same for gaming at 200% scaling & sampling with highest quality and all settings maxed such as the game aliens fire team elite or elex 2.

The point of furmark is testing the stability without waiting for years.

The point of memtest_vulkan is testing if your memory is producing errors or is running with expected performance under memory load.

Fair enough. I thought memtest_vulkan only stressed VRAM, which wouldn't make much sense considering that your GPU fans are tied to core temperature, not VRAM.

I just did a 5-minute run. Passed with a max VRAM hotspot of 92 ˚C.

 

[INSTG8R]

If your gpu can not run software that just computes something, then it has defects or installation failures such as missing heat gel or thermal paste or something.

People multiply matrices or do convolutions in datacenters and these reach close to peak compute performance and maximum heat output. Same for gaming at 200% scaling & sampling with highest quality and all settings maxed such as the game aliens fire team elite or elex 2.

The point of furmark is testing the stability without waiting for years.

The point of memtest_vulkan is testing if your memory is producing errors or is running with expected performance under memory load.

I tested my 5070 for 5 minutes, it reached 90Celcius and did not produce errors even with +2000 mem overclock. Fan speed was 55%. No game will push the memory this far ever. So I can say 5 years is good for memory. Ofcourse running memtest everyday is bad. Its pointless.

Try VKpeak of you want something that will fully stress your card in a realistic way not a now long until it throttles because of temps

…

 

[tugrul_LordOfDrinks]

Try VKpeak of you want something that will fully stress your card in a realistic way not a now long until it throttles because of temps
View attachment 391447…

I will try this today. Thank you.

Try VKpeak of you want something that will fully stress your card in a realistic way not a now long until it throttles because of temps
View attachment 391447…

CUDA-Z can use 37 TFLOPS but this uses 23TFLOPS. And gpu power was low. Imo its not using FMA instructions on float32/int32.

 

[TSiAhmat]

Fair enough. I thought memtest_vulkan only stressed VRAM, which wouldn't make much sense considering that your GPU fans are tied to core temperature, not VRAM.

I just did a 5-minute run. Passed with a max VRAM hotspot of 92 ˚C.

How high was your fans rpm in the test? (For reference)

 

[AusWolf]

How high was your fans rpm in the test? (For reference)

Good question... I did it in Linux, where I can't see my fan rpm. But it was audible, a bit louder than it is in games.

 

[TSiAhmat]

Good question... I did it in Linux, where I can't see my fan rpm. But it was audible, a bit louder than it is in games.

Ah okay, then I assume it was a bit over 1500 rpm. Want to try it as well on my card ^^

 

[rusty caterpillar]

I'm not sure running such an unrealistic load is a good idea. I'm not a fan of Furmark, either, for example.

Furmark pushes very little the VRAM, Superpositon 8K optimized 20-25 min is pretty good check for VRAM.

I'll try Vulkan at some point

I will try this today. Thank you.





View attachment 391453CUDA-Z can use 37 TFLOPS but this uses 23TFLOPS. And gpu power was low. Imo its not using FMA instructions on float32/int32.

Can I set Vulkan test to 10 min instead of 5 min?

 

[pulcsi]

My 9070 non-xt is a 245W model and max temp for VRAM is 82°C. When GPU is almost idle, VRAM temp is already at 62°C while my GPU temp is 40°C and hotspot is 42°C

 

[tugrul_LordOfDrinks]

Furmark pushes very little the VRAM, Superpositon 8K optimized 20-25 min is pretty good check for VRAM.

I'll try Vulkan at some point


Can I set Vulkan test to 10 min instead of 5 min?

Imo vulkan test 5 minutes goes max temperature so it becomes more of a temperature test at 10 minutes. I think no needed if games dont heat vram that much.

 

[TSiAhmat]

VRAM heated to 90 C° in my case. (at rpm 1500)

Spoiler

 

[AusWolf]

VRAM heated to 90 C° in my case. (at rpm 1500)
View attachment 391606

Spoiler

View attachment 391605

I guess it's safe to assume that all is well with the 90 °C memory hotspot at this point?

 

[TSiAhmat]

I guess it's safe to assume that all is well with the 90 °C memory hotspot at this point?

Probably.

 

[JustBenching]

That is normal at absolute full load, playing a game is easy. But for Radeon I see the high temps as excessive, normal or not..

Anywhoo, I will continue to watch those repair guys on the toob. Always interesting stuff.

95c ain't normal either. I mean there is nothing special about amd cpus that make them different to any other CPU, it's just amd marketed smartly to stop people from complaining. Those CPUs work exactly like any other CPU, they boost until they either hit max clocks or they throttle from temperature. Since they are harder to cool they hit the 95c throttle point before they hit the max clocks. So acting like it's normal or 95c is the target doesn't really check out.

Regarding the vram, 30xxx gpus (especially the 3090 with vram on the back) could hit up to 110c. Nvidia was saying it's normal, but it isn't. Use them enough and they will 100% degrade. I know, I've been there.

 

[Vayra86]

95c ain't normal either. I mean there is nothing special about amd cpus that make them different to any other CPU, it's just amd marketed smartly to stop people from complaining. Those CPUs work exactly like any other CPU, they boost until they either hit max clocks or they throttle from temperature. Since they are harder to cool they hit the 95c throttle point before they hit the max clocks. So acting like it's normal or 95c is the target doesn't really check out.

Agreed. I'm not a fan of a CPU running 95C either. AMD can say whatevert he fck they want about it. The warranty still doesn't get elevated to say 7 years. If they did, they can market 400C for all I care. Until then, I'll be that stubborn asshole running his parts under 90C if I ever can.

Still I won't be saying that a CPU that runs 95C and AMD saying its as intended is a bad product. Its just not for me.

 

[JustBenching]

Agreed. I'm not a fan of a CPU running 95C either. AMD can say whatevert he fck they want about it. The warranty still doesn't get elevated to say 7 years. If they did, they can market 400C for all I care. Until then, I'll be that stubborn asshole running his parts under 90C if I ever can.

Still I won't be saying that a CPU that runs 95C and AMD saying its as intended is a bad product. Its just not for me.

Not like you have a choice though, my 9800x 3d was burning like it had fever. But to be fair, it was probably my cooler that just didn't like the amd die, the guy I sold it too has much better temps with his aircooler.

 

[AusWolf]

Agreed. I'm not a fan of a CPU running 95C either. AMD can say whatevert he fck they want about it. The warranty still doesn't get elevated to say 7 years. If they did, they can market 400C for all I care. Until then, I'll be that stubborn asshole running his parts under 90C if I ever can.

Still I won't be saying that a CPU that runs 95C and AMD saying its as intended is a bad product. Its just not for me.

AMD says 95/89 ˚C is fine, which I agree with. I just don't want to run it that hot because I don't like the idea of my clocks being limited by temperature. It's like I paid for a product that I can't fully utilise because my cooling is shit. Maybe good enough for some, not good enough for me. I like having at least a little bit of thermal headroom.

Not like you have a choice though, my 9800x 3d was burning like it had fever. But to be fair, it was probably my cooler that just didn't like the amd die, the guy I sold it too has much better temps with his aircooler.

I've had some bad luck with X3D and liquid. I couldn't for the love of god cool my 7800X3D with a 280 mm AIO. Then I swapped for a be quiet Dark Rock 4, and everything was fine. I still can't figure out why.

 

[Vayra86]

Not like you have a choice though, my 9800x 3d was burning like it had fever. But to be fair, it was probably my cooler that just didn't like the amd die, the guy I sold it too has much better temps with his aircooler.

I didn't buy a 9000 series for that reason. The 7800X3D is tweaked such that it never reaches >85C.

Hot CPUs are a no no. I make my systems last. >10 years is standard. 15 happens. Still running a 3570k rig downstairs as HTPC, its pretty cost effective at this point Haven't changed anything in it except a PSU. 25 dollar cooler on it.

 

[AusWolf]

The 7800X3D is tweaked such that it never reaches >85C.

Mine is bone stock and still doesn't reach 85 ˚C (83-84 max under full load).

I make my systems last. >10 years is standard. 15 happens. Still running a 3570k rig downstairs as HTPC, its pretty cost effective at this point Haven't changed anything in it except a PSU. 25 dollar cooler on it.

I like that sentiment very much.