Tuesday, March 4th 2025

Leakers Record 90°C+ VRAM Temperatures on Unnamed Radeon RX 9070 XT Custom Cards
The Chiphell forum has provided a steady flow of AMD RDNA 4-related leaks—going back to early December; members believed that "Radeon RX 8800 XT" GPUs were imminently entering into a mass production phase. Since then, Team Red and board partners have officially revealed a full deck of Radeon RX 9070 XT and RX 9070 models—complete with a "modernized" naming scheme. By Christmas (2024), insiders appeared to have working units in their clutches—denizens of Chiphell have continued to dole out pre-release info; even deep into launch week. UNIKO's Hardware picked up on the latest signals; with owners of unnamed custom Radeon RX 9070 XT cards: "calling out bad cooling on GDDR6 VRAM."
As highlighted by Wccftech, NDA-busting disclosures have alluded to commendable GPU thermal measurements—when driven at full load—but several leakers have noted less than stellar results from VRAM temperature readings. Chiphell-sourced GPU-Z screenshots indicate a maximum recorded VRAM temperature of 94℃, with the involved GPU's hotspot hitting a top temp of 79°C. This leaked candidate seems to be a 329 W TBP-rated model. Wccftech observed inconsistencies with the other evaluated sample: "under full load, the GPU temperature at the hot spot reaches a max of 63°C but the memory temperature touches 88°C. The surprising thing to note in the first case is that the TBP is only 237 W, which seems weird considering the Radeon RX 9070 XT (reference spec) starts at a TBP of 260 W. This might be a bug...The user has confirmed that the first one isn't the RX 9070 as one would think, but it is the RX 9070 XT as well." The "guided" upper limit for newer VRAM standards is 95°C, so one of the anonymous custom cards is dancing dangerously in close proximity to the proverbial flame. Certain hardware news outlets reckon that GDDR6X memory will succumb to damage once a 120°C ceiling is hit. Hopefully, these issues are limited to a handful of review samples—a couple of AMD's trusted board partners have opted for Honeywell PTM7950 thermal pads and robust heatpipe formations. Stay tuned to TechPowerUp for W1zzard's incoming RDNA 4-related verdicts.According to Uniko's translation, the founder of Chiphell commented (in full) on the VRAM cooling situation(s): "previously, I only paid attention to the GPU temperature and ignored the video memory. The result is good... The video memory of several brands I have is extremely hot... One is 88℃ and the other is 94℃... It's really like going back to the mining era overnight...I suggest that you don't rush to pay attention to the evaluation of No. 5 to see if it is a common phenomenon or an individual case. If it is an individual case, you can avoid it when choosing a card."
Sources:
Uniko's Hardware, Wccftech
As highlighted by Wccftech, NDA-busting disclosures have alluded to commendable GPU thermal measurements—when driven at full load—but several leakers have noted less than stellar results from VRAM temperature readings. Chiphell-sourced GPU-Z screenshots indicate a maximum recorded VRAM temperature of 94℃, with the involved GPU's hotspot hitting a top temp of 79°C. This leaked candidate seems to be a 329 W TBP-rated model. Wccftech observed inconsistencies with the other evaluated sample: "under full load, the GPU temperature at the hot spot reaches a max of 63°C but the memory temperature touches 88°C. The surprising thing to note in the first case is that the TBP is only 237 W, which seems weird considering the Radeon RX 9070 XT (reference spec) starts at a TBP of 260 W. This might be a bug...The user has confirmed that the first one isn't the RX 9070 as one would think, but it is the RX 9070 XT as well." The "guided" upper limit for newer VRAM standards is 95°C, so one of the anonymous custom cards is dancing dangerously in close proximity to the proverbial flame. Certain hardware news outlets reckon that GDDR6X memory will succumb to damage once a 120°C ceiling is hit. Hopefully, these issues are limited to a handful of review samples—a couple of AMD's trusted board partners have opted for Honeywell PTM7950 thermal pads and robust heatpipe formations. Stay tuned to TechPowerUp for W1zzard's incoming RDNA 4-related verdicts.According to Uniko's translation, the founder of Chiphell commented (in full) on the VRAM cooling situation(s): "previously, I only paid attention to the GPU temperature and ignored the video memory. The result is good... The video memory of several brands I have is extremely hot... One is 88℃ and the other is 94℃... It's really like going back to the mining era overnight...I suggest that you don't rush to pay attention to the evaluation of No. 5 to see if it is a common phenomenon or an individual case. If it is an individual case, you can avoid it when choosing a card."
36 Comments on Leakers Record 90°C+ VRAM Temperatures on Unnamed Radeon RX 9070 XT Custom Cards
Lazy.
I don't see the point in commenting on this just yet, also, I think vram temps on 5090FE can get relatively high too, 94°c in TPU reviews, I don't see people crying over it.
:kookoo:
AFAIK 90-100C under load for VRM is actually pretty normal.
Look at the fan speeds.
I'll bet they're running a memory-intensive benchmark which doesn't challenge the cores much so the fan speed stays low and the memory heats up more than under a typical load.
Besides those temps being within the safe range anyway.
To go sir :rolleyes::laugh:
237W used at only 0.71V with 1693 MHz clock speed? Yet with a 45% memory controller load. Please explain the conditions under which a card does that. Never in a game, maybe when running other loads?
And the same for 329W at a mere 0.73V and 2028 core clock, but somehow with a 60% memory controller load. Again, in no game that exists.
These screenshots are irrelevant until someone can explain how to generate this type of load.
The highest I've seen on my stock MBA 7900XTX (355 W) was 71 - 89 - 94 C for edge, hotspot and VRAM respectively. According to HWinfo, the limit is 110 C for hotspot and memory. The fans were going up to 68% on the default curve. With a more aggressive profile the temps could have been lower.
This was pretty much a worst case scenario, tested in a 28 C room with the card pumping 350 W all the time, with a reference cooler. Judging by the low memory used I suspect Furmark.