Hey folks, first off apologies if this belongs in another section. I have been running my RTX 3090 FE card generally under 100% utilization overnight for training machine learning models and rendering 3D content with octane render. While my GPU core temperatures have been under control (I have a much larger case now corsair 5000X with lots of direct cooling) I just realized my GPU memory temperatures have been going past 110 C which seems like a bad idea for the extended periods I run the card.
 
I watched gamer nexus teardowns and a bunch of videos on thermal pad replacements people have been doing to these FE cards to lower the memory temp (crypto miners have shared a bunch of resources on this). I bought 3 Thermalright 85x45x1.5mm thermal pads and mustered up the courage to take the card apart and begin surgery. I found that the nvidia stock thermal pads were easily compressed and had a bit of fibrous material in the pad design. The stock thermal compound on the GPU die was very dried out as well and looked thin in some areas. I have cleaned and prep the chip surfaces and GPU die and applied MX4 using an X and 4 dot pattern since I hear this works well for convex dies. I didn't do too much research into compound selection as MX4 was what I had around. Should I try something better? or am I overthinking it. Overall pleased as the memory temps dropped by roughly 20C and the fans are running about 50% quieter even with a 114% power target.
 
Stock Stress Test Results MAX values (furmark 2 hour):
GPU Temperature: 71.0 C
GPU Memory Junction Temperature: 106.0 C
GPU Fan1: 2431 RPM (93%)
GPU Fan2: 2431 RPM (93%)
 
GPU Clock: 1950.0 MHz
GPU Memory Clock: 2438.0 MHz
 
Mod Stress Test Results MAX Values (furmark 2 hour) + 114% Power Maximum (Nvidia Performance Panel):
GPU Temperature: 71.3 C
GPU Memory Junction Temperature: 88.0 C
GPU Fan1: 1372 RPM (52%)
GPU Fan2: 1372 RPM (52%)
 
GPU Clock: 1950.0 MHz
GPU Memory Clock: 2438.0 MHz