I finally managed to buy the 6900xt reference model a few weeks ago, but after some disappointing/lower than average results i started troubleshooting.
The issue i seem to be having is that the junction temperature just pins to 110 degrees celsius within 10 seconds of starting a GPU intensive task.
Sometimes it peaks at 114 degrees celsius for a split second after which the core clock & power draw plummets.
I seem to be averaging out at 1976mhz core clock @ 215-230 watts.
Temperatures at average: 73 degrees celsius (with 110 degrees celsius junction)
Right now i have an open air case with additional airflow blowing directly through the fin stack.
I don't really care about the performance in general as long as it's close enough to advertised.
What i do worry about is the lifespan of the GPU running at these conditions.
What i've already tried:
If anyone has any tips or a potential solution to these issues (or simply that i have to RMA) i would appreciate it.
I'm gonna do some more tests today so i will be adding screenshots.
Basically your GPU card is overheating within 10 seconds.
The Hot Spot Maximum Operating Temperature (T.Junction) is 110c. Once it starts reaching or has reached and surpassed 110c the GPU card will automatically to throttle or slow down to keep the T.Junction (Hot Spot) at 110c or below.
If the GPU card is still under Warranty I suggest your open a Warranty Ticket to have it checked. Seems like either the Thermal paste is defective or the heat sink is not making good contact with the GPU chip or the GPU Fans is not working the way it should to keep the GPU card from overheating.
NOTE: If the GPU card is still under Warranty just RMA it rather then do some action that might void your Warranty.
EDIT: Does your PSU have good voltage outputs?
Also do you have two separate GPU power cables connected from the PSU to the GPU card or one Y cable from PSU to both GPU Power connectors?
Running the Power Color Demon RX 6900 XT.
On 5 minute or longer stress tests my junction temp peaks at around 70 to 74C with VPU clocks hovering from 2425 to 2485MHz.
Also CPU Ryzen 7 2700x at 4.2GHz MOBO MSI X370 M7 ACK RAM PC3200 4x8 Corsair Dominator Platinum PSU Thermaltake Tough Power 1200 watt SSD SATA 3 Samsung 870 EVO V-NANO 4TB.
I thought I would get higher junction temps running a 4.0 GPU in a 3.0 PCIE slot but I experience no bottlenecking that I can find through any test. The only thing I run into is some early access and DX9 games get small audio issues (just an annoyance has no impact on game performance).
I know the audio issue has nothing to do with AMD. I've tested the onboard Realtek Audio and an aftermarket Creative Labs PCIEx1 audio card and got the same audio issue so I'm leaning to the audio glitch being caused by windows 10 interference or the game platform Steam. All early access and older DX9 games I play from Steam have this issue.
I'm already in contact with AMD warranty and will do an RMA if possible.
This video here shows what happens with the AMD Radeon Software set to "Default" (excuse my terrible recording).
I do notice the fan speed is terrible but i'm guessing that it only looks at the Current/Package temperature?
It seems to finally ramp up towards the end.
Benchmark is Red Dead Redemption 2 @ Ultra (with unlocked settings, on Vulkan & everything Ultra). Resolution: 5120x1440.
I've done some more testing and the only way to get the junction temperature stable below 110 degrees celsius (hovering around 104-106) is to ramp up the fan control to basically 80-100%, undervolt to 1100mV & set a max frequency of 2100mhz.
This way the power draw stays below 210 watts and seems to be running stable with sadly, terrible performance.
The 1976mhz average I stated in my first message was with the AMD stress test for a longer period.
When it comes to the power supply I do not have the know-how to measure values.
But the 6900XT is replacing my custom watercooled Vega56 (with OC) which was drawing around 250 watts without issues.
I'm using 2 separate PCI cables from/to the power supply, so not the daisy-chain.
Im guessing that the thermal pad is just not making a great connection.
I read the 6900XT has no thermal paste, just a pad for the die.
Thanks for the reply!
Definitely, definitely RMA it. These are expensive cards, and you deserve the performance you paid for. Obviously, you have enough know-how to mitigate the problem with voltage, frequency, and fan curve--that's great and kudos to you--but you shouldn't have to.
I know it sucks to be without a card for a while, but the people I know who are most unhappy with their rigs are the ones who say, "I should have RMAed when I had the chance." Don't get stuck.
Best wishes to you.
Undervolt to 1000mv and lower the clock to 1600mhz
I'll just leave these pics here of my 6900xt Ref gpu as it came from factory after 2 months of use. I took it apart to install a waterblock on it (which btw it's great) when I realised the temps where ridiculous.
If i was you seeing these pics on some forum I would just take the **bleep** thing apart and repaste it!
PS: Shame on you AMD, some people are saving months to afford this Gpu