Low voltage, high temperature
Low voltage, high temperature
Hi there,
CPU model is i7 4790k
Device: Gigabyte GA-Z97X-UD5H-BK
Cooler: Hyper 212X (will be swapped for Noctua NH-D15 in two days)
I'm just starting out, so please understand if I'm explaining things slowly.
I adjusted the BIOS to the default optimized settings and saw the Gigabyte board increasing to 4.4 across all cores at 1.267 volts. I performed an OCCT v6.1.1 linpack test using these settings, but the screen would go black and the PC would restart after about 4 seconds. This occurs roughly four seconds into the test.
Aida64, Intel burn test, and Cinebench worked fine, though they became very hot—around 83°C in about ten minutes.
I manually set the clock to 4.4 and voltage to 1.2, then ran Intel burn test, Aida64, Cinebench, and OCCT, but not linpack. I used large data sets, auto thread count, and auto instruction set. Tests stopped after ten minutes as temperatures remained at 83°C even at lower voltage.
I suspect the cooler might be struggling so I ordered the Noctua NH-D15. Still unsure why the stock settings cause a black screen; I worry something else could be at play and my new cooler might not solve it.
Note: Package temperature during idle is between 22°C and 42°C
Hey there, for basic tests such as AIDA, Prime95 or OCCT they push your CPU to its full potential. In everyday use like gaming or apps that use a lot of resources, you won't reach that level. If you're really serious about overclocking, the results from stock coolers—especially Intel ones—won't be impressive. Thirdly, a CPU isn't just a CPU; it could be a specific model with high heat output or voltage needs. For instance, my i7-8700k can run at 4.7ghz with 1,280v, but pushing it further would require 1,360v and cause a big temperature rise. People have managed to hit 5ghz at around 1,225v. Just wait for Noctua to come out, if you're aiming for that level, start testing from the very beginning—at least 20 minutes is a good minimum, ideally 30 minutes or longer until the temp stabilizes.
It seems the test should have been conducted on the Hyper 212X rather than the stock intel cooler. Should I not proceed with the OCCT linpack test? Can I assess stability using just the Aida64 and the OCCT large data sets? I managed to run those for 10 minutes before they stopped, while the OCCT linpack black screened in about 4 seconds.
That would be quite concerning, what's your airflow rating for both cases and general ambient temperature? From my own experience, I'd rely on the OCCT default test, and the linkpack test includes AVX which is essentially the heavy load you mentioned in your previous post.
I have excellent airflow; I removed the front panel earlier to check, but it didn’t help. I’m also puzzled about why it’s heating up so much at 1.2 volts.
It seems like the issue might be related to the CPU itself, needing improved cooling and higher voltage. Other potential causes could be the cooler or airflow. The motherboard is another possibility, but Gigabyte hasn’t failed me yet, and Z chipsets are known for good performance. Keep in mind that the CPU maker doesn’t guarantee anything beyond the stock version, and it’s likely they don’t cover turbo features.
I've updated the setup and noticed great performance. I've adjusted the voltage to 1.225 overclocked at 4.6, and after 36 minutes of AIDA64, temperatures have fluctuated between 54 and 63°C. The peak reached 70°C. Can I think this is stable after a few hours?
Another victory for Noctua to wrap it up quickly
😉
As promised, hold off on the replacement cooler and ignore stock temperatures. Your temps seem pretty good to me. Typically, thermal throttling begins above 90°C, so reaching 80°C under stress without errors is a huge improvement. I’d even suggest pushing the OC further.
Yeah, the Noctua is really good, the temps are down a bit. I've been testing for a few days now and had to increase the voltage to 1.25 at 4.6. The temps are still between 54 and 71c.
I'm starting to question whether I should have increased the voltage from the start. I haven't stayed stable during any of the stress tests, even under heavy load. I wasn't stable on any of them, not even at stock values, so it's unclear what's going on.
I've managed to run my simulators for hours without crashing, just like at stock values. I'll keep monitoring and might lower the voltage. I'm beginning to think the most stressful test is the one I use with the PC.
Note: I set the voltage to 1.25 but the LLC/Ring offset adds 0.153, so even though I see 1.25 at full load on CPU ID, the monitor shows 1.260 and 1.272 on the hardware side.
I assess stability using only the Aida64 and OCCT large datasets. I managed to execute them for 10 minutes before stopping them, as the OCCT Linpack encountered a black screen in approximately 4 seconds.