Overclocking i7 7700k Need Help - High Temps in Prime95

[Somnax]

Overclocking the i7 7700k and in Prime95, every time the first test passes around the 5 minute mark, my temperatures spike from around 75 C to upwards of 100 C which causes Prime95 to fail shortly after.

I am using a Corsair H115i as front intake, one fan at the back as intake, and two fans at the top as exhaust. All fans are running at maximum speed.

 

Here is a test at 4.8Ghz, 3000 Mhz ram, 1.3 Volts. You can see the spike in temperatures slightly after halfway. The test fails right before cpu usage falls before 100 because I stopped prime95.

 

 

I thought maybe it fails because temps and so I started lowering voltage. At 1.26 Volts, I fail prime95 immediately. At 1.27 Volts, a similar thing happened.

 

This time it took less time to fail after the temperature spike, and the spike happened right after passing the first test.

I'm not sure what would cause the sudden increase in temperature if the cpu had been running at full load up to that point. It was only 5 minutes, so maybe it hadn't reached equilibrium, but the sudden spike still seems odd to me.

 

Has anyone experienced this before and have any advice? Would greatly appreciate it.

 

 

 

[App4that]

You can try 1.32v, that seems to me a magic number for stability. 

 

As for the temps are you sure the pump is plugged in? Is it properly seated? What thermal paste was used?

 

And if all that fails, you might have got unlucky and have one that needs to be delid. Up to you if you want to take the risk, or contact Intel on the problem. 

[Daveeede]

Prime95 is recommended anymore because it can effectively overstress and damage the cpu. I'd recommend running Aida64's stability test and maybe Asus's Realbench instead.

[Shadow_Storm56]

6 minutes ago, Somnax said:

Overclocking the i7 7700k and in Prime95, every time the first test passes around the 5 minute mark, my temperatures spike from around 75 C to upwards of 100 C which causes Prime95 to fail shortly after.

I am using a Corsair H115i as front intake, one fan at the back as intake, and two fans at the top as exhaust. All fans are running at maximum speed.

 

Here is a test at 4.8Ghz, 3000 Mhz ram, 1.3 Volts. You can see the spike in temperatures slightly after halfway. The test fails right before cpu usage falls before 100 because I stopped prime95.

 

 

I thought maybe it fails because temps and so I started lowering voltage. At 1.26 Volts, I fail prime95 immediately. At 1.27 Volts, a similar thing happened.

 

This time it took less time to fail after the temperature spike, and the spike happened right after passing the first test.

I'm not sure what would cause the sudden increase in temperature if the cpu had been running at full load up to that point. It was only 5 minutes, so maybe it hadn't reached equilibrium, but the sudden spike still seems odd to me.

 

Has anyone experienced this before and have any advice? Would greatly appreciate it.

 

 

 

yea use aida 64  and also it is a small die with a non soldered ihs it will get very hot when heavily overclocked but that does seem unusually high

[knightslugger]

Prime95 is pretty hard on CPU. I see no point in running a benchmark that simulates extreme number crunching. Aida64 is more real world simulation, and a little bit more which makes it a good stability stressor.

[porina]

If you run Prime95 in stress test "blend" mode, it will vary between small and large FFT tasks. The small ones fit inside the CPU so stresses the cores, but not the ram interface. The large ones don't fit, so will also stress the IMC and ram.

 

I just fired up 28.7 I happen to have lying around on my desktop. It starts at 320k FFT units, which will be too large to fit inside the CPU so will stress ram, but it wont be as heavily loading on the cores. I can't afford to run P95 for any amount of time at the moment as I'm in a compute challenge, but I suspect it might be switching to small FFT before the crash.

[Somnax]

So I reseated the cpu cooler. Using Antec Formula 7 Nano Diamond thermal compound for the paste. It is a few years old now. Not sure how much of a difference that makes. Installed Corsair Link because I hadn't done that yet, and verified the pump and the fans are working. Now going to run some tests on Aida64.

 

Based on Porina's comment, maybe it is because the second part actually stresses the cores more, but hopefully I will learn more after Aida64

[Somnax]

Here are the results after 10 minutes on Aida64. Temperatures are tame compared to the massive 100 C of prime95. 1.28V now.

 

Should I be concerned about the Prime95 failures even though Aida64 seems to run fine?

 

[App4that]

I wouldn't worry about the crashes in Prime if you can stay stable a few hours in Aida. I run at least 4 hours of Aida before moving onto games and normal tasks. 

[Somnax]

My motherboard is showing 80C. That seems extremely high for a motherboard. Motherboard is even showing a higher temp than the cpu

[App4that]

80c isn't that bad. Not great, but not alarming. Could be the sensor too. Try using Realtemp to see if it matches. 

 

https://www.techpowerup.com/realtemp/

[Somnax]

59 minutes ago, App4that said:

80c isn't that bad. Not great, but not alarming. Could be the sensor too. Try using Realtemp to see if it matches. 

 

https://www.techpowerup.com/realtemp/

Is there a way for realtemp to check motherboard temperatures? It is showing the same results for the cpu temps

[App4that]

Not that I know of. Why I'm wondering of its a false reading on the MB. Or there isn't enough airflow over the heatsink

[Somnax]

3 minutes ago, App4that said:

Not that I know of. Why I'm wondering of its a false reading on the MB. Or there isn't enough airflow over the heatsink

Well my back intake is up top, my front intake is the h115i near the bottom. And the top is exhaust. Maybe this set up isn't allowing for airflow over the actual heatsink?

 

I monitored the cpu cooler temp with corsair link and it never went above 34C. Makes me think maybe there is a  bad connection between the heatsink and the cpu so I'm ordering new thermal paste

[Spudbilly]

7 minutes ago, Somnax said:

Well my back intake is up top, my front intake is the h115i near the bottom. And the top is exhaust. Maybe this set up isn't allowing for airflow over the actual heatsink?

 

I monitored the cpu cooler temp with corsair link and it never went above 34C. Makes me think maybe there is a  bad connection between the heatsink and the cpu so I'm ordering new thermal paste

Aida64- run it for 2-4 hours. If it passes then you are good.

 

MB temps - check your BIOS temps. If they match software monitoring apps then that will indicate the actual temp you are dealing with. 

 

Airflow setup - your fan setup sounds fine. Can you plz use HWMonitor app and show a screenshot of the whole window displayed after running 10 minutes of aida64. I can tell what's up once I see that. 

[Somnax]

Here's the result after 10 minutes

[done12many2]

16 minutes ago, Somnax said:

Here's the result after 10 minutes

 

Remember that you don't need nearly the same amount of VCore to pass AIDA64 or most daily tasks up to a fairly moderate load.  If you aren't using anything heavy in AVX, drop your VCore a lot more and retest with AIDA64.  You should also give strong consideration in running RealBench in stress mode as it throws in a mild amount to AVX to ensure your safe for moderate work loads.

 

I'm in the middle of binning a handful of 7700 CPUs and I can tell you that out of the 3 I've tested so far, none of them have been bad enough to require as much voltage as you're running at 4.8 GHz for an AIDA64 / RealBench type load.

 

I happen to be binning these chips with the exact same cooler (h115i) so I'll run a 4.8 for comparison sake for you.

[Somnax]

20 minutes ago, done12many2 said:

 

Remember that you don't need nearly the same amount of VCore to pass AIDA64 or most daily tasks up to a fairly moderate load.  If you aren't using anything heavy in AVX, drop your VCore a lot more and retest with AIDA64.  You should also give strong consideration in running RealBench in stress mode as it throws in a mild amount to AVX to ensure your safe for moderate work loads.

 

I'm in the middle of binning a handful of 7700 CPUs and I can tell you that out of the 3 I've tested so far, none of them have been bad enough to require as much voltage as you're running at 4.8 GHz for an AIDA64 / RealBench type load.

 

I happen to be binning these chips with the exact same cooler (h115i) so I'll run a 4.8 for comparison sake for you.

Now I'm running aida64 with 1.25 volts. Seems to be running fine. It still fails immediately in Prime95 which is why I upped the voltage.

 

Is it normal for your computer to move to a snails pace if you do both cpu and gpu stress at the same time. As in my mouse barely moves when I select that option.

 

Here's a screenshot after 10 minutes at 1.25 volts

 

[done12many2]

2 minutes ago, Somnax said:

Now I'm running aida64 with 1.25 volts. Seems to be running fine. It still fails immediately in Prime95 which is why I upped the voltage.

 

Is it normal for your computer to move to a snails pace if you do both cpu and gpu stress at the same time. As in my mouse barely moves when I select that option.

 

Here's a screenshot after 10 minutes at 1.25 volts

 

Don't do both at the same time as the tests were never designed to be run that way.  You are actually cheating stability by doing that in some cases.

 

If you need to test full system load (CPU + GPU) run RealBench as I stated earlier.  It will utilized GPUs and CPU at the same time.  Pretty great test.  Harder to pass then AIDA64, but what I consider to be a better prep for most daily situations from gaming to productivity.

 

I'll have a 15 minute AIDA64 run up for you in about 10 minutes using the same CPU and cooler as you.  I'm running a different clock speed, but similar voltage, which is the bulk of what actually matters temperature wise.

[done12many2]

Here's a run at a slightly higher voltage than that of your own.  We may have different ambient temps, TIM and other things that contribute to difference in temps.

 

 

[Somnax]

That is definitely more of what I was expecting. I ran aida64 at 4.9ghz 1.25 volts for 25 minutes. Got a high of 90 C rip

 

 

[done12many2]

15 minutes ago, Somnax said:

That is definitely more of what I was expecting. I ran aida64 at 4.9ghz 1.25 volts for 25 minutes. Got a high of 90 C rip

 

 

 

Are you running CPU cache at 1:1 or something?  That will raise your temps a bit, but not much.

[Omie]

I'm pretty sure those temps are normal. Kabylake has way higher temps than Skylake. From what I've seen, everyone who has a 7700k has temps like that. Someone who delidded this CPU resulted in getting 20-30 degrees drops. 

[Somnax]

Now I am at 4.8 Ghz and 1.22 volts. I passed 4 hours of Aida64 and an hour of realbench. Still have the issue where Prime95 fails instantly, which is why I increased voltage initially

 

[done12many2]

9 minutes ago, Somnax said:

Now I am at 4.8 Ghz and 1.22 volts. I passed 4 hours of Aida64 and an hour of realbench. Still have the issue where Prime95 fails instantly, which is why I increased voltage initially

 

 

I wouldn't worry about Prime95 at all unless you know you need that level of stability, which 95% of people don't.  The load generated by Prime95 has its purpose, but is usually overkill for most people.

 

In addition to AIDA64 and RealBench, run one or two other stress tests and you can call yourself stable.  You've already done more than most on LTT by running more than AIDA64 itself, which is pretty much a joke.

 

I'm having pretty good luck.  I'm still in the process of raising the overclock slowly, but surely.  This is where I'm at right now.  I might go a little higher on the CPU speed once I delid, but for now, I'll work on increasing CPU Cache and memory.

 

Good luck.