Radeon VII Long Term 110'c Junction Temp ?

[Dicehunter]

Thought I'd make a dedicated post rather than it getting lost in the Radeon VII thread.

As a few people are a ware the cooling plate on the R7's heat sink is very concave, Meaning quite a lot of it simply does not make contact with the GPU die itself hence the 110'c junction temp, I had to remedy mine by getting some 400, 800 and then 2000 grit sandpaper wrapped around a flat block of metal to make it flat, Temps went down a tremendous amount as the plate is now actually making contact with the die.

So that got me thinking, I'm pretty sure not everyone is going to break out the sand paper on a £700 card, How long will these cards last at such extreme temps on the junction ? 110'c prior to overclocking is hella high.

 

[Aretak]

One thing to consider is that it's extremely unlikely that the Radeon VII is the first card to ever have parts of the die running a lot warmer due to uneven contact or some other reason (even with CPUs you often have some cores that run much warmer than others). It's just that in the past you'd never know it, since there was only one thermal sensor to guage the entire GPU temperature from. It's only with all these sensors now scattered all over the place that you're able to get a somewhat accurate reading of exactly what's going on all over the chip, though even then you don't know the exact spot that's hottest - just the highest temperature from one of the sensors. With that in mind, it's pretty much impossible to guess at what sort of effect it might have long-term. There's no comparative data to reference, and even with long-term testing of just the Radeon VII it would be difficult to prove that any failure down the line was caused by a specific hot spot on the die without some sort of deep analysis of the silicon.

Ultimately, I think you just have to have faith that AMD have stress tested everything properly and know what they're doing.

 

[Dave2150]

I’d say you would have been better off RMAing your card rather than voiding the warranty like you did.

My R7 runs 81C, with 100JT, with fans capped st 2200rpm so it’s quiet, don’t think my card is a golden sample either as many achieve similar numbers.

3 year warranty on the sapphire models direct with ocuk inspires confidence also

 

[Zernath]

I’d say you would have been better off RMAing your card rather than voiding the warranty like you did.

I have to say, it takes some pretty big cajones to deliberately void the warranty on such an expensive card! I would rather RMA, and then see if the next one fared any better. Surely AMD wouldn't knowingly sell a card with a completely borked cooler?

Doing it yourself may work in the short term, but if anything goes wrong down the road, you're significantly out of pocket.

 

[vapor matt]

Thought I'd make a dedicated post rather than it getting lost in the Radeon VII thread.

As a few people are a ware the cooling plate on the R7's heat sink is very concave, Meaning quite a lot of it simply does not make contact with the GPU die itself hence the 110'c junction temp, I had to remedy mine by getting some 400, 800 and then 2000 grit sandpaper wrapped around a flat block of metal to make it flat, Temps went down a tremendous amount as the plate is now actually making contact with the die.

So that got me thinking, I'm pretty sure not everyone is going to break out the sand paper on a £700 card, How long will these cards last at such extreme temps on the junction ? 110'c prior to overclocking is hella high.

I say if it helps your card and reduces temps then why not do it. I will be fitting a water block as soon as it arrives. isn't that voiding the warranty.

 

[Zernath]

I say if it helps your card and reduces temps then why not do it. I will be fitting a water block as soon as it arrives. isn't that voiding the warranty.

Yes, but they'll only know if there's warranty seals on the card, otherwise you can stick the old cooler back on and they'd be none the wiser.

 

[Dicehunter]

Yes, but they'll only know if there's warranty seals on the card, otherwise you can stick the old cooler back on and they'd be none the wiser.

AFAIK there are warranty seals on all Radeon VII's in 2x places on screws that need to come off if you want to remove the block for cleaning or install a waterblock, Can't really get around it.

 

[Zernath]

AFAIK there are warranty seals on all Radeon VII's in 2x places on screws that need to come off if you want to remove the block for cleaning or install a waterblock, Can't really get around it.

Then it depends on the manufacturer, whether they accept stock cooler removal or not. Still not as risky as lapping in my opinion.

 

[nashathedog]

How are you measuring Junction temp please? I'm using GPU-Z 2.17.0 and unless they're using a different name for it it's not on here.

 

[Dicehunter]

Then it depends on the manufacturer, whether they accept stock cooler removal or not. Still not as risky as lapping in my opinion.

The card itself is rock solid and I was removing the cooler anyway to put a waterblock on it

How are you measuring Junction temp please? I'm using GPU-Z 2.17.0 and unless they're using a different name for it it's not on here.

Using Wattman.

 

[Jerry64]

How are you measuring Junction temp please? I'm using GPU-Z 2.17.0 and unless they're using a different name for it it's not on here.

In GPU-Z 2.18.0 its GPU Temp Hot spot

 

[nashathedog]

Using Wattman.

Cheers,
I'll see what temp mines hitting when I play some games next.

 

[nashathedog]

@Dicehunter I just did some daily event in dirt rally and Wattman tells me the junction peaked at 106, dragging along the line shows it spending most it's time at around the 100 mark. I'm currently running the core at stock which is 1802 @ 999mv and the memory at 1100. might try dropping the voltage a bit more, I also have the fan spped limited to 61% but it never hits that it tend to peak at 60%, I'm not sure why it doesn't use the extra 1%.

@Jerry64 Thank's I'll check it out.

 

[Dicehunter]

@Dicehunter I just did some daily event in dirt rally and Wattman tells me the junction peaked at 106, dragging along the line shows it spending most it's time at around the 100 mark. I'm currently running the core at stock which is 1802 @ 999mv and the memory at 1100. might try dropping the voltage a bit more, I also have the fan spped limited to 61% but it never hits that it tend to peak at 60%, I'm not sure why it doesn't use the extra 1%.

Whoever manufactured the heatsink for AMD really did a poor job as the cooling plate is very concave and hardly makes contact with the die itself hence why I ended up sanding it flat and dropping temps A LOT.

 

[nashathedog]

Whoever manufactured the heatsink for AMD really did a poor job as the cooling plate is very concave and hardly makes contact with the die itself hence why I ended up sanding it flat and dropping temps A LOT.

I wish I could afford to take the risk and do the same. I may still put it in a loop after Ryzen 3000 lands though. It's a shame how both AMD and Nvidia seem to mess up the simplest of jobs, They both moved to questionable dual/tri fan design reference cards, I find it baffling.

 

[fizzy]

Whoever manufactured the heatsink for AMD really did a poor job as the cooling plate is very concave and hardly makes contact with the die itself hence why I ended up sanding it flat and dropping temps A LOT.

What did you use for thermal paste/pad when you refitted the stock cooler?

 

[Dicehunter]

What did you use for thermal paste/pad when you refitted the stock cooler?

Thermal Grizzly Kryonaut.