The performance settings in BIOS seems to default so that in "normal" state you get a quiet NUC and in "turbo" state you get to hear the ventilation a little more with the increased performance while still keeping a temperature belove < 70 degrees, and Intel don't recommend changing any of this. Intel Tuning Plan is probably meant for rigs with extra cooling applied, and the NUC casing doesn't leave room for that kind of cooling.
The BIOS setting Sustained Mode Time just seems odd, because if the turbo mode is supposed to start as the workload increase, why should it stop because of a time limit instead of lesser CPU use??
If you're passively cooled forget about any overclock. With really good active cooling I was able to raise the TBPM from 15W to 30W which raised the TDP of the graphics to ~26W according to the Intel Extreme Tuning Utility and my CPU temps went ~80C. With the regular cooler/kit your temp will go to about ~85C. This is when running graphics stress test in IETU. However with sustained 30W and good active cooling your graphics chip will continuously run at 1GHz instead of being throttled down by power controller. Also the raise in power raised 3d Mark 11 (perf preset) from P1207 to P1723. This is with DDR3 1833.
Im using an Akasa Newton X chassis which seems to perform better than the default fanbased cooling of D54250WYK.
I previously tried:
Turbo Boost Short Power Max: 30W
Turbo Boost Power Max: 30W
Processor Current Limit: 40A
and as you say it gets hot after a while.
Im now running with:
Turbo Boost Short Power Max: 25W
Turbo Boost Power Max: 25W
Processor Current Limit: 34A
which seems to fixed the stuttering problem I got previously when lighting effects were applied by the game (like when shooting).
To keep things cooler I also apply:
r_dynamic 0 (dunno if this have any effect)
I think the adjustable TDP is a great differentiator between the Core i3 and Core i5 NUCs. Even though the HD 5000 graphics chip in the i5 has 40 cores (execution units) vs. the 20 cores in the HD 4400 graphics in the i3, its performance is not 2x. There are already benchmarks on the net comparing the two. The reason is simple - you cannot break the laws of physics and the 20 cores in the i3 are already throttled to fit in the total TDP of 15W shared with the CPU so throwing more cores at the problem will not accomplish much. However by allowing us to raise the TDP of the i5 chip, Intel has allowed us to unleash the potential of the HD 5000 graphics. The results speak for themselves - 50% improvement in 3d mark 11 by raising the TDP to 30W.
But unfortunately too few results on safe overclocking limits.
How is the math/relation between regarding current limit vs turbo boost short power vs turbo boost power vs gpu performance?
Power = Current * Voltage. You can set limits to both power and current and the voltage is variable and changes at runtime. In general power = heat and you use the power limit to limit heat. You leave the current limit high enough so that you run into power throttling. Think of the current limit as an o-sh*t circuit breaker.
The short power boost allows you to set a short duration during which you can go beyond the standard power limit since in that short period of time the CPU will not ramp up heat that quickly - the heat sink will soak the extra heat in the beginning allowing your to have that short power boost.
You tune all this in the Intel Extreme Tuning Utility. Running GPU stress you can fine-tune both - short power boost duration and limit and steady state power limit based on the maximum temperatures you want to tolerate. Looking at the datasheet for the i5-4250U CPU the max voltage is 1.84V and max default current is 40A for 28W and 32A for 15W (page 90 of the spec sheet that you can download here: https://www-ssl.intel.com/content/dam/www/public/us/en/documents/datasheets/4th-gen-core-family-mobile-u-y-processor-lin…
Just a friendly reminder: Overclocking may have an impact in system stability; it can result in increased temperatures and erratic system behavior. It will shorten the life of your product and voids the warranty of your processor.
Here are my findings for those who like to tinker and realize the risks. With the stock cooling setup you can do two things:
1) You can buy fast RAM (such as F3-2133C11D-16GRSL) and then manually set it up to run at 1866 with the following timings: tAA: 10, tRCD: 10, tRP: 10, tRAS: 29, tCK: 1, tRRD: 5, tRC: 38, tRFC: 249, tWR: 14, tWTR: 7, tRTP: 7, tFAW: 25. This will give you over 17GB/sec bandwidth. Notice that as soon as you put 2133 RAM into the NUC, the BIOS will set it to run at 2133 and the system will be unstable. You must immediately go into BIOS and change the RAM from auto to manual 1866 (x14 multiplier), reboot, and then start setting the rest.
2) You can increase the TDP from 25W short burst / 15W steady to 35W short burst / 31W steady along with an increase in max current to 42A. This will increase your 3d Mark 11 score from ~1200 to ~1700 or about 40% increase due to 50% increase in graphics capabilities. This chip (HD 5000) has 40 cores and with 15W total to be shared with the CPU it has not enough power to fully utilize those 40 cores. By raising the limit you allow those cores to run at full speed rather than be throttled, while still allowing safe temperatures (<80C) at the expense of a much louder fan. Double check your temperatures by running graphic stress test in the IETU.
Nice! I'm in a Streacom NC2 case, running parallel math computations. Elsewhere on the web one reads that raising these power limits will only help with graphics. I was seeing erratic parallel performance with two cores until I reset my BIOS as proposed here: 34/30 Burst/Sustained and Current Limit 42. Now my two core parallel performance is nearly twice one core performance, comparable to a high end Mac mini on two cores, and somewhat faster than a two core Linode.com instance.
In BIOS 0033, they've tinkered with access to these parameters. I couldn't click through "Cores" as described various places, but I could find these settings by search, then add them to favorites. I hope Intel isn't trying to phase out these controls, as for my use they'd be chopping my NUC in half. Perhaps they're just making them less accessible so newbies don't mess with them by accident in the stock case. (Edit: discussed here: Re: Please restore NUC BIOS navigation of Performance Pages to that of BIOS Version 0027 or earlier.)
Overpowered as described, in a passive case, my cores run nearly 70 C. Running pretty much any 120mm fan at 5 volts off a USB port provides the faintest air current, enough to drop these core temps to around 55 C in near silence. That is my working configuration, with the Streacom case oriented vertically. Much better than the stock case, which I used briefly to confirm my unit worked.
Dear Syzygies, Dear LeszekM, Dear Apachez, et al,
Imagine my surprise, then , when , in the middle of a long interview session with Dr. Ing.Pierugo Gobbato in 1988 he let drop this morsel:
<< I remember that my father visited Henry Ford in Dearborn in 1939. That was when Mr Ford said to him: " When I see an Alfa Romeo , I lift my hat">>
"I lift my hat" to you Gents. I am clearly in the presence of true overclockers.
Cc LeszekM, Apachez, et al,
A couple of ideas....
1. Try tilting the fan so it is not impinging at 90 degrees onto the heat-sink.
2. Kind of an old-school idea... try making a chimney e.g.
I would think that a chimney could be made from smooth walled PVC piping of the correct diameter
with a half cylinder slot cut out of the base so that it clips onto the face of your heat-sink,
leaving a suitable gap at the bottom for air to go in (the "fireplace" end).
It may be advantageous to use some step-down (in diameter) reducer(s) to a smaller pipe bore
(so you possibly get greater airspeed, but most importantly make the whole structure less top heavy)
for the tall vertical end of the chimney.
Cc Syzygies, Apachez, et al,
Cc "BIOS Engineering"
There is something inconsistent with the way the BIOS is handling the XMP settings going on here...
Your NUC should have set your RAM to
"Profile 1 ... the Enthusiast / Certified setting"
"Profile 2 ... the Extreme or Fastest possible settings"
When I put my RAM in, the BIOS set it to 1866 (I expect this is "Profile 1 ... the Enthusiast / Certified setting"),
and then I had to go and manually select 2133 (I expect this is "Profile 2 ... the Extreme or Fastest possible settings").
I do not know if you are having a BIOS version issue ( a "glitch" _not_ a bug ;-) ) or a RAM SPD XMP Profile issue,
but my BIOS version is
and my RAM is
G.Skill F3-2133C11D-8GRSL DDR3L-2133 PC3L-17000 4096MB X 2 CL11-11-11-34 SO-DIMM
It looks like PC parts without casing. Maybe if you drill holes in a freezer, stuck wires through and placed the NUC in the freezer then overclocking it would be cool ...