High temps for 4790k in Prime95

[b33k3rz]

In Prime95 after the first part I get temps up to 97c, when the CPU begins throttling itself.

Idle temps are good, as are gaming temps.

What gives?

It's not overclocked, though I did enable XMP in Asus UEFI.



Specs:
4790k
r9 290
Asus z97-a
16gb 2400mhz ram
hyper 212 evo, Arctic silver 5

 

[VirtualLarry]

Is this enough evidence to say that the OP incorrectly applied the cooler mounting or the thermal paste? Or is a 4790K at stock, under AVX load, actually too much for even a 212EVO?

 

[coffeejunkee]

It most likely shows that by enabling xmp profile mct is also enabled. When mct is enabled vcore is increased too and that coupled with the extra AVX vcore bump will overwhelm a Hyper 212.

 

[b33k3rz]

Is this enough evidence to say that the OP incorrectly applied the cooler mounting or the thermal paste? Or is a 4790K at stock, under AVX load, actually too much for even a 212EVO?

I don't know a lot about voltages or XMP, but does XMP increase CPU heat? And if it does, Could I manually adjust voltage on my CPU?

 

[ShintaiDK]

I don't know a lot about voltages or XMP, but does XMP increase CPU heat? And if it does, Could I manually adjust voltage on my CPU?

What may happen is that your BIOS when you select XMP. Also enables an autooverclocking feature that overclocks the turbomodes.

So for example when you run Prime95. Instead of running at say 4.2Ghz and lets imagine 1.1V. Then it will run at 4.4Ghz and maybe 1.25V. Voltage is hypothetical since it depends on the CPUs VID table.

Try look in your BIOS if you can find something called Multicore Turbo or Multicore Enhancement and then disable it.

 

[PliotronX]

Is this enough evidence to say that the OP incorrectly applied the cooler mounting or the thermal paste? Or is a 4790K at stock, under AVX load, actually too much for even a 212EVO?

That is my vote, the 212 has a HDT base so proper TIM application is different. Even though I relidded my 4670k, my 212 does not hit those temps until 4.6GHz.

 

[coffeejunkee]

You do have to take some extra care when applying tim to hdt base coolers like the Hyper 212 (making sure the gaps are properly filled).

But my money is still on mct vcore increase + avx vcore bump. Improper tim application would increase temps by just a few degrees. MCT + AVX could increase vcore from regular 1.05 or so to prossibly 1.25. At 1.25 Haswell gets toasty, even with Intel's improved DC tim.

I don't know a lot about voltages or XMP, but does XMP increase CPU heat? And if it does, Could I manually adjust voltage on my CPU?

Yes, if you want mct without unnecessarily high vcore increase you have to set the multipliers and cpu voltage yourself. But the stress testing takes time. Or you can manually disable mct after enabling xmp, the extra 200MHz are not that much of a speed increase anyway.

Also, keep in mind it's only for (unrealistic) AVX loads like Prime that you will see high temps. For regular loads like gaming the vcore isn't automatically increased, thus temps stay lower.

 

[b33k3rz]

You do have to take some extra care when applying tim to hdt base coolers like the Hyper 212 (making sure the gaps are properly filled).

But my money is still on mct vcore increase + avx vcore bump. Improper tim application would increase temps by just a few degrees. MCT + AVX could increase vcore from regular 1.05 or so to prossibly 1.25. At 1.25 Haswell gets toasty, even with Intel's improved DC tim.



Yes, if you want mct without unnecessarily high vcore increase you have to set the multipliers and cpu voltage yourself. But the stress testing takes time. Or you can manually disable mct after enabling xmp, the extra 200MHz are not that much of a speed increase anyway.

Also, keep in mind it's only for (unrealistic) AVX loads like Prime that you will see high temps. For regular loads like gaming the vcore isn't automatically increased, thus temps stay lower.

HWMonitor during Prime95 test says that my CPU's VID is 1.293V

 

[Techhog]

That is my vote, the 212 has a HDT base so proper TIM application is different. Even though I relidded my 4670k, my 212 does not hit those temps until 4.6GHz.

The 212 EVO base is different from the 212+ though. The heatpipes touch on the EVO.

 

[Techhog]

HWMonitor during Prime95 test says that my CPU's VID is 1.293V

And we found the issue... probably. HWMonitor isn't totally accurate, but still, you most likely have a voltage issue. What's the VCore?

 

[b33k3rz]

And we found the issue... probably. HWMonitor isn't totally accurate, but still, you most likely have a voltage issue. What's the VCore?

Vcore under the motherboard voltages settings says 1.808V

 

[PliotronX]

The 212 EVO base is different from the 212+ though. The heatpipes touch on the EVO.

Shoot, I clearly made an error in saving a few bucks with the old version..

Vcore under the motherboard voltages settings says 1.808V

That is probably the input voltage which you could probably get away with lowering it at around 1.7v but it is safe on air up to 2v or so.

 

[Ken g6]

It most likely shows that by enabling xmp profile mct is also enabled. When mct is enabled vcore is increased too and that coupled with the extra AVX vcore bump will overwhelm a Hyper 212.

Actually, this has my vote. Prime95 likes fast RAM, so when you give it 1.65V 2400 RAM, not only does it push the on-die memory controller faster, it also pushes the rest of the CPU faster.

OP, see if you can find a stable 1.5V RAM speed. Or maybe start with 1600MHz and try to get the CPU cool enough first.

 

[ShintaiDK]

HWMonitor during Prime95 test says that my CPU's VID is 1.293V

And if you use CPU-Z, whats the clock, 4.4Ghz? Then you overclock the turbomodes.

 

[WhoBeDaPlaya]

I expect those kind of temps from a bad 4770K, not a 4790K!

 

[coffeejunkee]

HWMonitor during Prime95 test says that my CPU's VID is 1.293V

You should be looking at the actual vcore itself as displayed by (fex.) cpu-z. The VID is something different.

In cpu-z you can also see the multiplier. With a 4-core load like Prime it should be 42. If it's 44 it means mct is enabled.

 

[mallebabbe]

And don't forget that the prime95 28.1 and higher version will give you about 8 degrees more temp then the .27 version will do.

 

[mindbomb]

make sure the fan is on and spinning at the right rpm.

 

[crashtech]

I actually did not know that XMP could bump the core voltage or enable MCT. Seems like a tricky deal on the mobo makers part to try and keep competitive in tests.

 

[coffeejunkee]

I haven't checked my own board cause I'm a 'set everything manually anyway'-type of guy. But I've read about this happening several times.

It seems the idea is 'hey this guy runs fast memory, he probably won't mind if we overclock his cpu a bit for him'. Which is fine in principle, but some kind of warning message/dialog would be nice.

 

[GLeeM]

You should be looking at the actual vcore itself as displayed by (fex.) cpu-z. The VID is something different.

Naming snafu ... in HWinfo, VID is vcore (called cpu core voltage in Asus UEFI). So 1.293v is the "AVX2/FMA3 auto-bumped up" vcore.

 

[PliotronX]

I actually did not know that XMP could bump the core voltage or enable MCT. Seems like a tricky deal on the mobo makers part to try and keep competitive in tests.

Same here, thought it only dealt with memory and since I tweak the timings manually I have never used it so didn't know

 

[Compman55]

what kind of TIM is intel using under the IHS on the 4790k? You may be doing nothing wrong, the issue could be under the heat spreader.

 

[Techhog]

what kind of TIM is intel using under the IHS on the 4790k? You may be doing nothing wrong, the issue could be under the heat spreader.

Nobody else has this issue, though.

 

[coffeejunkee]

Naming snafu ... in HWinfo, VID is vcore (called cpu core voltage in Asus UEFI). So 1.293v is the "AVX2/FMA3 auto-bumped up" vcore.

Not sure...OP says he's using HWMonitor which is made by the same guys that make CPU-Z. But HWMonitor only displays cpu vcore.

But since he's talking about VID he might be using HWinfo which shows both VID and vcore (vcore is displayed in the sensor window under io-sensor readings).

If vcore is indeed 1.293V then the high temps are no surprise.