Unfortunately, the options for my CPU on XTU are rather limited. None on RAM, and on CPU I can only control the following:
-Core Voltage
-Core Voltage Offset
-Turbo Boost Short Power Max
-Turbo Boost Power Max
-Multipliers for all cores
-Processor Cache Ratio
-Cache Voltage
-Cache Voltage Offset
*so, I cannot adjust the current, as the option is not present
*also, can't control the reference clock (is locked/greyed)
_____
I replaced the power supply and now I have one that can push up to 330W - I guess this one is for Alienware, of more power hungry laptops. (previously I had the biggest one for Precision M6800, which is only 240W).
Temperature never raised above 86, but now, with current settings I rarely go over 81 degrees (tends to stay around 77-79). I replaced the thermal paste with Coollaboratory metal pads (much safer than liquid metal and totally okay if they touch aluminium, as the metallic foil is indium-based, not gallium).
For now, best position is all cores to 39x, anything higher will only push the frequency up on idle, but drops drastically under load.
Power I put both options to 61W, that covers the power needs of my CPU and leaves some room. The "cruise" drawn power seems to be 47W and for short period it can get up to 58W.
I've been playing with XTU both with Adaptive and Static voltage settings and I could observe that Static doesn't help at all - despite the fact that many overclockers insist on NOT using the dynamic offset and is best to adjust the voltage manually.
When on Static Voltage, overvolting drops performance a lot, for the smallest change, but when undervolting, the performance caps quickly and if I push it more, I only get BSOD
On the other hand, keeping Core Voltage on Default and reducing the offset to -0.100V seems to be the "sweet spot" ... well, I didn't get it right yet, as -0.095V -> -0.100V seems to be the zone that needs further attention.
Despite of many saying the Cache doesn't affect much, in my case, playing with Cache multiplier and voltage offset (i put them as close as VCore settings), seemed to bring some improvement.
____
I am a bit confused in regard to Benchmark / Stress Tests on XTU... on some settings Benchmarking ends well, with nice score, but then it fails on stress test, other times is quite the opposite... which I don't really understand why would it fail the benchmark, but then do well on longer stress test? Unless, the sudden change between load/idle is the moment when it fails (current drops too abruptly and cpu doesn't recover fast enough?).
Also the marks in XTU don't seem to reflect the actual performance of the machine (computing performance), but rather a combination between computing and power, from what i could see... which doesn't really help me. I need to evaluate the computational efficiency, regardless of how much power it takes.
Some people say that for Haswell Prime95 is not the right stressing tool (can bring accidental overvolting, due to dynamic offset?). What about OCCT, then? Or you would suggest a better alternative? I could just let it render a 3D file, in a CPU-based rendering software...that one keeps all cpus busy full-time.
!! While writing this (nothing else major was running on the machine), I just got a BSOD. I guess it's still a bit undervolted.