Massive overheating problems with RTX 3090!

[Harlequin]

wow , is this Thermi part 3?

 

[Wrinkly]

I think it was the Titan X that had memory on the back that ran over 100C, so nothing new here. Never buy the stock cooler.

 

[Jay85]

Glad mines under a waterblock, it sits around 76c while mining and sub 60's when gaming.

 

[LtMatt]

Well it draws a lot of power comes with the territory.

I used to let my MBA 6900 XT reach 90c+ on the GDDR6 to keep a whisper silent fan speed with an overclock.

If it’s within spec it’s okay IMO.

 

[ToOo]

completely agree with @LtMatt

The GDDR6x on a 3090 draws about 55W just for active read write, with losses on the VRM that's basically 60W of power dissapation just running the DRAM. Just think for a minute about how hot an old 60W incandecent bulb would get

Micron says 110C is fine and there will probably be a modest engineering margin on that.

 

[sparkymark75]

completely agree with @LtMatt

The GDDR6x on a 3090 draws about 55W just for active read write, with losses on the VRM that's basically 60W of power dissapation just running the DRAM. Just think for a minute about how hot an old 60W incandecent bulb would get

Micron says 110C is fine and there will probably be a modest engineering margin on that.

Have Micron officially said that? All I've seen is summation that the safe temp limit is 110c.

 

[chrcoluk]

I wouldnt treat a temp safe to avoid hardware failure as its ok to run like that from day to day.

 

[ToOo]

The maximum rated "Case" temperature per the Micron datasheet is 95C

The thermal probe is measuring the internal "Junction" temperature, this will be many degrees hotter.

I could make a calculation if the themal resistance of the package was in the datasheet but it isnt, might try and find some specs for similar 180-Ball FBGA packages

EDIT:
Unfortunately i can't find the package thermal resistance spec, i did find an application note on the Micron website in which the best example they give is 5°C/W for Rjc on a 1.5W multi die NAND package but that's not really useful for drawing conclusions about GDDR6.

 

[Disco_P]

The card doesn't throttle until 110C and cards have 3-5 year warranties so nvidia and partners must be happy with that temp otherwise they're gonna end up with a bunch of fried £1400+ cards to replace.

 

[heartburnron]

Well it draws a lot of power comes with the territory.

I used to let my MBA 6900 XT reach 90c+ on the GDDR6 to keep a whisper silent fan speed with an overclock.

If it’s within spec it’s okay IMO.

How are you measuring the RAM temp Matt? I didn't think the 6900 XT has a sensor for it?

 

[LtMatt]

How are you measuring the RAM temp Matt? I didn't think the 6900 XT has a sensor for it?

Hwinfo64 mate measures the memory Junction temperature. Throttling temp in GDDR6 for 6000 series is 100c. So anything below that is within spec.

 

[heartburnron]

Hwinfo64 mate measures the memory Junction temperature. Throttling temp in GDDR6 for 6000 series is 100c. So anything below that is within spec.

Oh right, thanks. Is it a recent addition as I've looked before and can't see it now either? What version are you on and what is it labelled?

 

[LtMatt]

Oh right, thanks. Is it a recent addition as I've looked before and can't see it now either? What version are you on and what is it labelled?

One of the recent versions but not the latest, will confirm tomorrow.

 

[heartburnron]

One of the recent versions but not the latest, will confirm tomorrow.

Thanks mate.

 

[Grim5]

Have Micron officially said that? All I've seen is summation that the safe temp limit is 110c.

GDDR6x starts throttling at 110c, so 110c is the limit - if 110c was not the limit, it would throttle earlier

 

[HRL]

Oh right, thanks. Is it a recent addition as I've looked before and can't see it now either? What version are you on and what is it labelled?

v6.42 onwards. VRAM Memory Junction Temp.

 

[fs123]

The maximum rated "Case" temperature per the Micron datasheet is 95C

The thermal probe is measuring the internal "Junction" temperature, this will be many degrees hotter.

I could make a calculation if the themal resistance of the package was in the datasheet but it isnt, might try and find some specs for similar 180-Ball FBGA packages

EDIT:
Unfortunately i can't find the package thermal resistance spec, i did find an application note on the Micron website in which the best example they give is 5°C/W for Rjc on a 1.5W multi die NAND package but that's not really useful for drawing conclusions about GDDR6.

If you watch the video LTT got 100C directly on top of the memory chip using a thermal camera and they got around 102C memory junction temp using GPU-Z/HWinfo. Essentially the case temp is within a few degrees of the internal junction temp which means Nvidia is letting the memory temp run over spec.

 

[pp111]

The temperatures of the 3080 and especially the 3090 ram has actually put me off buying one. I must confess I was 80% the way there just by the issues associated with buying one, but the temperatures just tip me over the edge. I will get a 3070 and make do with that.

 

[chroniclard]

The temperatures of the 3080 and especially the 3090 ram has actually put me off buying one. I must confess I was 80% the way there just by the issues associated with buying one, but the temperatures just tip me over the edge. I will get a 3070 and make do with that.

Why....they are completely fine in normal usage.

 

[Joxeon]

The temperatures of the 3080 and especially the 3090 ram has actually put me off buying one. I must confess I was 80% the way there just by the issues associated with buying one, but the temperatures just tip me over the edge. I will get a 3070 and make do with that.

The majority won't run much hotter than 90c in normal gaming situations, it's only mining that pushes temps to the max and probably something nvidia didn't account for as most of this issue can be solved by just using decent thermal pads which are only a couple of $ extra.