Hardware monitoring tool HWInfo64 just received an update (V6.42) that adds support for “GDDR6X Memory Junction Temperature” monitoring on the RTX 30-series cards. (This only works with GDDR6X equipped Ampere cards like the RTX 3080 and RTX 3090, and not the RTX 3070.)
With this feature enabling users to check their VRAM temperature, the former findings of absurdly higher temps have now been validated through the software. Tomshardware ran some checks wherein looping Metro Exodus at 4K Ultra settings for several minutes on an RTX 3080 Founders Edition allowed the GDDR6X modules to hit a peak temp of 102C. This is above the TJmax for GDDR6X which is 95C (the temperature at which the components will try and shut down to protect themselves).
Curiously, enabling DLSS and Ray Tracing brings it down to a high of 94C instead. The RTX 3090 Founders Edition in Cyberpunk 2077 with DLSS and Ray Tracing enabled allowed the GDDR6X temps to peak at 100C.
It gets worse when it comes to mining though, specifically in Ethereum mining that was tested. Both the RTX 3080 and RTX 3090 had their GDDR6X modules’ temperatures peaking at a blistering 110C. Naturally, the GPU would lower its clock speed, severely, to compensate for these ridiculously high VRAM temps.
The same was verified to occur on multiple different boards from various vendors and seems to be a general issue. Also worth noting is that these are the temps BEFORE applying any fancy overclocking, usually employed in mining to squeeze out every bit of hashing performance.
There are certain ways to reduce these temps. Tomshardware used MSI Afterburner to drop the GPU and GDDR6X clocks as low as possible (-450 MHz on the core and -502 MHz on the RAM), but it didn’t really do much. The cards would still drop to much lower GPU clocks (900-950 MHz) after a few minutes, though the mining performance stood relatively unaffected at a decent 95MH/s. Decreasing the power limit to just 60 percent finally got the VRAM temps down to 90C, but then so did the mining performance, falling through to a 60-65MH/s.
The takeaway from this is that mining for longer durations, 24/7 as is usually done, with the memories running at 100-110C, seems like a recipe for a premature card failure. If you are someone utilizing these cards for mining, it’s advisable to turn down the power limit, although the mining speeds could be affected. For gaming, however, the GDDR6X runs at around 100C or lower, just breaching the safe limits, but Nvidia deems it acceptable from their perspective and hence shouldn’t be a major concern as it’s usually not 24/7. Try not to overclock it if you notice these temps with HWInfo64.
There are no precise answers for now as to why Nvidia is allowing these temps on the GDDR6X chips and the cards don’t appear to start throttling GPU clocks until the temps hit 110C. The “GDDR6X Memory Junction Temperature” also isn’t clear in its meaning, but it’s worth assuming it’s the hottest part of any of the GDDR6X chips measured internally, for now.