[SPOILER="Prime 95 and customizing Turbo Boost on the 6700K"]
Prime 95 was used to confirm if the vCore undervolts were stable or not. Unfortunately the 6700K hit 97C instantly when all 8 threads were started on small FFT. It seems to have to do with the AVX instruction set which consume a huge amount of power. I did tests in Prime 95 but avoided 8 thread FFT because it would always load the core temps up way too high. I really doubt that a gaming (or any) application can ever generate this much heat.
I decided to focus on testing realistic loads mainly. Cinebench and Asus RealBench were used, particularly the builtin Handbrake encoding module in RealBench. What I found was that those massive AVX power draws from Prime 95 went away and the average was more like 68-75 watts during Cinebench or encoding.
Not wanting to blow up the 6700K, I customized the Turbo Boost setting to limit max TDP. That is in place for safety based on the current cooling and hasn't impacted any realistic performance much.
Here is a demonstration of the new boost setting in place when Prime95 is loaded:
During the AVX instruction set we get a peak of 81 watts for 8 seconds before being limited to under 65 watts.
You can see the temps immediately spike too in the lower right chart.
I'm not sure how long it takes for Turbo Boost to reset itself, but it seemed like the package would stay under the specified limit until it returned to idle. After idling, the boost would enable again. This kept temps pegged right around 70C which is perfect.
After some testing I settled on a Turbo Boost setting of 68 watts max with 78 watts boost. The limit is mainly for protection based on the current thermal solution. It can be raised later depending on how successful the intake fan turns out to be.
Note that this 6700K is already undervolted here, so 68 watts does not mean being crippled to ~71% performance based on the 95W TDP.
It is probably closer to 68w/82w = 83% of potential TDP available.
Also for 1-3 cores of usage, it should never need to throttle, only with 4 cores at max load will it potentially cause a performance loss. We will see if it causes a performance loss when actual gaming benchmarks are run.
Furthermore, I did some testing on overclocking the one and two core multipliers to 4.4 ghz while reducing the three and four core multiplier to 4.0 ghz. This worked for a higher score in the Cinebench single thread test.
So potentially, if we there is a single thread heavy game, we can actually increase performance while undervolting at the cost of reducing four core performance.
[/SPOILER]
