High temperature difference across CPU cores i9 7940x
High temperature difference across CPU cores i9 7940x
Hello, this is my first message here. I'm currently testing a custom loop setup with an i9 7940x bare chip paired with the direct die frame for x299 processors, and a 2080ti graphics card. The system has 3 480 rads of heat spreader, with two in push-pull configuration, one in push-only, and a 360 in push. I'm running an ek dual d5 CPU with two Thermaltake D5 variants. I believe this configuration is more than sufficient for the workload.
Now regarding the cooling: the CPU performs adequately overall, but I'm consistently hitting the temperature limit on certain cores (110°C), while others stay in the 60s. Undervolting definitely helps, but it's really disappointing to see such high potential being wasted. I've remounted the block multiple times, applied liquid metal and thermal paste, adjusted mounting pressure, but nothing seems to resolve the issue. The temperature difference across cores remains well over 40 degrees.
Is anyone familiar with running a waterblock on a bare die? I was wondering if the base might be slightly curved or not flat enough. I don't have a spare block, but I'd like to confirm whether this setup is typical for bare chips. Thanks in advance for any advice!
The significant temperature difference between cores suggests either a faulty sensor reading or poor mounting. You might want to reinstall the IHS to check if the issue resolves. If it doesn't, proceeding with the fix would help confirm the problem type.
I used to run it using the ihs. Temperatures were higher but core-to-core differences were smaller. It supports more than one core. I can share screenshots once I'm at the computer. I’m curious if there’s a method to identify which core is on which die, though I doubt it given Intel’s core management on HCC CPUs. Mounting was quite a challenge—I disassembled the loop many times to verify it would be impossible. Still, I think I should give it another try.
I considered this for my 7920X and found two possible approaches, neither simple. One depends on the mesh cache, with latency differences between cores and memory controllers changing based on their positions. Others suggest this would be too noisy to rely on, given measurement errors outweighing the signal we seek. I questioned whether aggressive averaging could filter out the interference, but it remains unclear how to implement that effectively. The second method involves activating cores sequentially and watching temperature shifts, hoping neighboring cores' changes reveal a pattern. Regardless, understanding the chip layout would still be necessary before making assumptions about which cores are active or idle.
Looking at individual cores helps visualize how temperature distributes, though it might be somewhat uncertain. I don’t have the right tools for precise readings, so that’s not an option for me. Appreciate the details!
I haven’t found any straightforward or reliable way to do this. Eventually, I might try the second method on my 7920X with original IHS settings. I plan to run Prime95 in a single thread and adjust core movement manually using affinity data, checking individual core temperatures with hwinfo64. So far, I haven’t delidded it, but under heavy loads I’m seeing 10–20°C differences. Maybe those surrounding cores or uneven TIM are causing the issue, or it could be something else. I’m not ready to delid it just yet.