Taming my 9900k

[Gorgon]

In an effort to expand my Compute capabilities I recently picked up a i9-9900k and a 3900x.

 

The 3900x has proven relatively plug-and-play installing nicely in my Aorus x570 Pro WiFi and the performance showing a nice uplift after replacing the "adequate" stock cooler with a NH-D15.

 

The 9900k, however, has lived up to its reputation as a thermal monster. Making matters worse the GigaByte z370 Gaming 5 it has been installed in suffers from a rather weak VRM (4-phase) that is itself prone to overheating. A first attempt at cooling this beast using a Hyper 212 Black with a Noctua NF-A12iPPC2000 PWM strapped onto it showed poor results. While perhaps adequate for light gaming my requirements are for it to remain stable under 24x7 compute and even with an under-volt and a -300MHz AVX off-set and 4.6GHz all-core limit could not pass Prime95 torture test (mprime -t) of longer duration.

 

Replacing the Hyper 212 with a 120mm CLC with the NF-A12iPPC2000 showed similar results.

 

Next up was an upgrade to an EVGA 280mm CLC and a Fractal Meshify S2 to house the system in.

 

A first attempt at a high all-core overclock with an under-volt showed that 4.6GHz all-core at 1.160V Vcore appeared to be the best attainable while passing Prime95 with AVX enabled. Futher testing showed, however, that LLC at Turbo and Vcore at 1.250V were required for stability. Again there were some stability issues and after dropping the all-core down to 4.5GHz and Vcore to 1.130V the system remained stable for Folding with the Cores at 64-72C, VRM at 81C (17C Ambient).

 

Stability was still elusive though with the system randomly locking up after 8 or 9 hours with nothing in the logs (hard lock) likely indicating a hardware issue.

 

After clearing the CMOS settings and running Prime95 it was observed that the VRM was hitting 115C with no end in sight during small FFT tests with AVX enabled.

 

Removing the plastic I/O shroud on the motherboard the plastic spacers on the VRM heat-sinks were removed to allow better contact of the heat-sinks to the MOSFETs and a 80mm fan installed to provide airflow over the VRMs. This improved things but only slightly.

 

Prime95 Torture tests were then run while observing the VRM thermals. An AVX off-set of -300MHz was found to be required at stock settings to keep the VRM thermals during small FFTs under 100C. A Prime95 Torture Test was run overnight (17 hours) and no errors or warnings were recorded.

 

The system has returned to folding driving 2 RTX 2070 Super Hybrids and with 14 threads CPU Folding (an 8 and a 6-thread slot) with no stability issues in almost 48-hours. The UPS shows the system having an idle load of 55W and 225W when just the CPU slots are enabled so the CPU draws about an additional 170W folding.

 

Hindsight being 20/20 the better approach would have been to just retire the z370 motherboard and get a new x570 board and another 3900x. Realistically this system would need a z390 motherboard with a beefier VRM to work well and the x570 with a 3900x would still crush a 9900k performance wise.

 

At some point, when I can take the system off-line for an extended period, I intend to see how much I can lower Vcore as that could both alleviate the strain on the VRM and may potentially allow a slight increase in the maximum all-core clock.

[LFreeman]

TBH This is the reason I upgraded my Maximus Hero X to the XI. I had done a lot of reading on the VRMs working, but would cause issue if I tried to push my 9900k to 5.0Ghz. In the end I did get the Hero XI. I am at 5.0GHz at 1.170V. I have a Corsair H150i keeping it cool. Max temp in P95 for 12 hrs. was 81c. I mainly game and when I F@H I Just use my GPU.

[j1philli]

2 hours ago, Gorgon said:

In an effort to expand my Compute capabilities I recently picked up a i9-9900k and a 3900x.

I really enjoy your process posts @Gorgon. Thanks for the insight!