4790K overheating with NH L12

[Bonebreaker777]

Nice to hear you have a working CPU, but still it runs very hot.

[Sylvester]

Yes, worth thinking about, replicating the SCAN test temps shows my cooler is set up OK. Intel ARK 4790K spec states max running temp Tcase is 72.72°C.

With Noctua and RAM at XMP (2400 @ 1.64v), CPU temps max out at 82°C in AIDAx64 at 4x 4.4GHZ turbo. This has to be balanced against MrRockliffe's point that temps are artificially high in a stress test and real use temps will be closer to spec. Also with volt tweaks the max temp drops 7°C so far.

I observed the transient power draws on the faulty CPU were spiking over 200W whereas the replacement keeps the power steadier. But power and temps are chicken and egg. Its hard to know at this point if this replacement is representative of most 4790K or whether the entire batch has a problem and this is the best of a bad bunch.

Should Intel have used better thermal interface or is this a good thermal interface with QC issues? I think getting to the bottom of this is confused by questions about the boundary between a fault, extreme usage and normal operating conditions. Too many variables, does not compute...

[Sylvester]

OK, never say die, I have tested again with Asus RealBench 2.2 using the same set up. This is billed as an open source real world stress test which was recommended by SCAN.



This is with Noctua NH-L12 & GTX770 VGA. BIOS volt tweaks (vCore offset -0.065v, power limiter 115W) reduced max temps by 6°C in a previous bench AIDAx64, RAM at XMP (which added 1°C in AIDAx64 tests, so net reduction of 5°C) produced 73°C max CPU temps in RealBench (ambient 25.1°C). This slightly exceeds Tcase which is 72.72°C. Delta T here is 47.9°C. Judging from Noctua NH-L12 reviews this is the kind of temp you would normally get with an overclocked i7 at about these clocks.

eg the delta T in this NH-L12 review on Tom's Hardware shows delta T 47°C in Prime95 v27.6 with an undervolted 1.05v 4770K overclocked @4.0GHz. Replicating these Prime95 tests, temps are in the same ball park eg delta T 47°C with a 105W power limiter in place which throttles the CPU down to 4.0 GHz and zero offset vcore at 1.0421v. So I think that clears up the question about whether this CPU is any worse than others of its kind. Its not but the first one was.

[MrRockliffe]

Glad to hear you've had it sorted. Just out of interest why are you using a low profile cooler? Haven't looked through all of the posts so unsure as to the system you're using

[Sylvester]

A picture is worth a thousand words.



This build has been in planning for a while, slowly gathering parts. Portability was a factor so the first choice was a lightweight aluminium Lian Li PC-V354 mATX case (as a combined birthday / Christmas present from members of my family, so it absolutely has to be used now!) There is limited space for cooling due to the PSU location. A hydro would fit the lower front 120mm intake but would add half a kilo and blow warm air across the VGA and only take about 10°C off the temps. So air looked worth trying first and the Noctua was the best low profile air cooler, if it is unusable then a hydro is an option but it might work out now the CPU has been fixed.

[Sylvester]

These are my best attempt at a balanced summary and conclusion, which I posted on the Intel forum.

For those who are having similar problems...

• my conclusion is that the first CPU was overheating as the replacement CPU is running significantly cooler.
  
  • So this means there are faulty overheating CPUs out there and it is worth exchanging them if you have one, but this is happening in the context of a CPU type which does run hot in bench tests and requires some BIOS adjustment and a better cooler than the Intel retail cooler to bring the temperature closer to spec (Tcase 72.72°C) under heavy load but it still tends to exceed this even under coolers with an appropriate thermal wattage.
• FYI The two key BIOS adjustments I made to enable this cooler were
  
  • to set the power limiter somewhere between 88 -125 W, currently 110W
  • and to set a negative vcore offset of -0.050v.
  • These restrict turbo to 40x in Prime95 v28.5 but allow it to run steadily and keep the temps close to Tcase.

For Intel these facts indicate a quality control issue. Further to this, with a non faulty CPU, using the retail cooler results in thermal throttling under load and the CPUs are not able to operate to spec either in terms of operating temperature or clock frequency under loads like Prime95 v28.5 with third party cooling like the NH-L12.

This suggests to me IMHO that the package thermal interface relies on thermal throttling to prevent CPU damage rather than providing a sufficient pathway for peak heat transfer, by design. As a result the CPU cannot achieve its specified capability in bench tests like Prime95 without third party cooling capacity significantly exceeding the rated thermal output of the CPU.

[Bonebreaker777]

About the RAM running at 1.64V at 2400Mhz.

How much the temperatures would change if you would limit it to 1.5V at the nearest stable frequency (or even to 1.35V at the nearest stable frequency again)?

[PowerPie5000]

I don't usually rely too much on the voltages displayed using HWMonitor. I bought my 4790k not long after release and it seems ok apart from the high VID when the cores boost to 4.4GHz (up to 1.263v). I'm using an Alpenföhn Matterhorn Pure CPU cooler (single fan) with Arctic MX-4 paste and the temps usually max out around 58c when gaming. My Corsair Vengeance RAM is 1600 @ 1.5v.

Here's a screenshot for comparison using HWMonitor (not a very accurate tool imo). It's currently idling and the room temps is about 20c:


[IMG][/IMG]

[Sylvester]

About the RAM running at 1.64V at 2400Mhz.

How much the temperatures would change if you would limit it to 1.5V at the nearest stable frequency (or even to 1.35V at the nearest stable frequency again)?

I checked that when replicating SCANs replacement test with AIDAx64 and 1.5v 1333 vs 1.64v 2400 (XMP with manual volts) made 1°C difference raising the max temp of a >20min test to 82°C from 81°C using the Noctua cooler and no volts or limiter tweaks. So not nearly as much difference to this replacement CPU as it did on the faulty CPU, which I also tested here, where the frequency change had a bigger effect than voltage.

The volts and limiter tweaks had a much bigger -6°C effect in the right direction, so worth having the XMP and lose a little turbo which is the trade off. XMP 2400 delivers 29Gb/s memory bandwidth whereas 1600 1.5v was 19Gb/s, according to Sisoft Sandra.

[MrRockliffe]

Awesome build! I think I'll be looking into a small micro atx build in the future. Full tower is too big for my needs.

[PowerPie5000]

Awesome build! I think I'll be looking into a small micro atx build in the future. Full tower is too big for my needs.

My last build was micro-ATX as I wanted a "space saving" setup (I still have it), but I ended up going for a standard ATX tower with my new build... It just feels more airy and looks tidier inside too as everything is not so cramped .

Anyway, what exactly is the "System Agent Offset" voltage thing about? Mine is set to +0.199v by default in the bios (shows as +0.204v using HWMonitor in my previous post). My 4790k is running stock with everything set to auto in the bios. Also, are the memory controller and DRAM voltages supposed to be the same? Speccy shows the memory controller as 1.696v and the Asrock OC utility shows the DRAM as 1.500v (ram voltage is correct).

I never really looked in to this stuff too much with my Sandy & Ivy setups.

[1stCowGirl]

A picture is worth a thousand words.



This build has been in planning for a while, slowly gathering parts. Portability was a factor so the first choice was a lightweight aluminium Lian Li PC-V354 mATX case (as a combined birthday / Christmas present from members of my family, so it absolutely has to be used now!) There is limited space for cooling due to the PSU location. A hydro would fit the lower front 120mm intake but would add half a kilo and blow warm air across the VGA and only take about 10°C off the temps. So air looked worth trying first and the Noctua was the best low profile air cooler, if it is unusable then a hydro is an option but it might work out now the CPU has been fixed.

only one thing is hurting.... the NH-fans!
those are regular NH-1600rpm.
what you need is the NF-A14 industrialPPC-3000 PWM 12cm 3000RPM _ (comes in 14cm also).

http://www.noctua.at/main.php?show=p...0&lng=en&set=1

here is a nice mod _ (was needed a 12cm Fan but tested the 14CM_ resaults were Awsome on the 4790K _ (HS: C12P SE14)

[Sylvester]

Thanks 1stCowGirl, interesting mod. So what does that fan sound like? Also what is the heatsink?

Since my last update I tried closing up the case after getting a couple of better outflow fans, I chose Akasa Viper S-Flow 140mm and 120mm. I removed the dust grill on the 140mm top outflow, not much point in trapping dust inside the case. These are pretty quiet at 75% but still make a fair breeze.

With case closed though temps still went to mid 70's gaming with 44x4 turbo so I throttled back turbo to 41x4 42x3 43x2 44x1 which gets me gaming temps of low 50's in Kerbal Space Program to low 60's with Elite Dangerous Beta which is below Tcase. The main thing is the memory is at 2400 XMP, which is the only reason I have got to move on from the QX9650 X38 build I have used for 6 years, which has a perfectly good quad processor but the platform memory bandwidth is 8Gb/s which will bottleneck anything over a GTX760 or R9 280, where this Haswell with GSkill is 29Gb/s which allows a much better graphics card. I have put a GTX770 in there for now as the best bang for buck, but will replace that later when its worth doing.

[Sylvester]

Just wanted to add that as a result of the ongoing discussions on the Intel communities board and hardware testing by several contributors Intel have issued the following advice document to assist people experiencing overheating with the I7 4790K processor.


Troubleshooting Intel® Core™ i7-4790K / i5-4690K overheating.

[krazy_olie]

Just wanted to add that as a result of the ongoing discussions on the Intel communities board and hardware testing by several contributors Intel have issued the following advice document to assist people experiencing overheating with the I7 4790K processor.


Troubleshooting Intel® Core™ i7-4790K / i5-4690K overheating.

Thanks,
I may ditch Asus' confusing ai suite for intel's stuff.

Turning on XMP seems to cause high voltages and temps for me.