I9-9900k P95 Small FFTs 90c+ stock clocks

[sburd2006]

Freshly built system:
I9-9900k
Gigabyte Auros Ultra mobo
Kraken x72 front mounted
NZXT H700i case


The I9-9900k using stock clocks is hitting 90c+ when running P95 Small FFTs v29.4. I have a Kraken x72 in the front of my H700i case pulling are over the radiator. The pump is set to 100% as well as the fans and it instantly hits 88c+ when running the Small FFts test. I am using a Gigabyte Auros Ultra board and cannot figure out why the temps are so high. At idle I am seeing 27c-32c temps and in games I have not seen it go over 55c. I am going to get some thermal paste and reseat the AIO but would idle and game temps stay that low if it isn't seated well? In HWMonitor I am seeing a max of 1.296VCore that seems high when most sites I have seen recommend starting at 1.3 to see if 5GHz is achievable.

 

[chemmajorp53]

As far as I know, I'm not an expert, but running an AVX workload on prime95 at standard clockspeed produces unrealistic amounts of heat unless you specifically work with AVX. That's why most people recomment using programs other than prime95 to burn test. You can also set a negative AVX offset on most motherboards if you're interested in that. If it's only at 55C while gaming I wouldn't worry. Check and see what your temps are sitting at in something like cinebench.

 

[1LiquidPC]

I'd try running version 26.6 Prime95 if you are planning on testing with this particular software. It's the last version prior to using AVX instruction sets. Newer versions are not recommended for long-term testing/benching as the powerdraw and extreme strain on the motherboard VRMs and CPU could cause damage over time. The AVX loading used for these tests are above and beyond "real-world" scenarios for which AVX would be used in typical programs from my understanding.

 

[CompuTronix]

... i9-9900k ... Kraken x72 front mounted ... stock clocks is hitting 90c+ when running P95 Small FFTs v29.4 ... pump is set to 100% as well as the fans ... why the temps are so high ... I am going to get some thermal paste and reseat the AIO ... max of 1.296VCore ...

sburd2006,

On behalf of Tom's Moderator Team, welcome aboard!

In this instance, it's not likely that you'll gain anything by re-seating your AIO. "Throttle" temperature (Tj Max) for your 9900K is 100°C. However, since we know that cooler is better for ultimate stability, performance and longevity, it's prudent to observe a reasonable thermal margin below Tj Max. Accordingly, here's the nominal operating range for Core temperature:

Core temperatures above 85°C are not recommended.

Core temperatures below 80°C are ideal.

Core temperatures increase and decrease with Ambient temperature.

Workload, Power and Core temperatures are all directly related. To put CPU thermal testing into perspective, we need to take a closer look at Power. Let's use an example with round numbers to keep it simple:

If a CPU is designed to run at 100 Watts TDP (Thermal Design Power), then it should draw 100 Watts of Power using Intel's stock settings in BIOS. Intel uses a proprietary test software that applies a steady-state workload (with steady Core temperatures) to drive the CPU to a steady 100 Watts. This means their software workload is a "100% TDP workload". So ...

CPU TDP = 100 Watts
Workload = 100% TDP
Power draw = 100 Watts

Make sense so far?

Let's examine some aspects of TDP that are never discussed or clearly explained:

There's a difference between "100% TDP workload" and "100% CPU utilization".

CPU utilization means processor resource activity, not % TDP workload. There's a wide assortment of workloads (apps, games, etc.) that run at 100% CPU utilization, which is the top of the "utilization scale".

There's no "top" of the scale for % TDP workload. There's an assortment of popular test utilities that will push a CPU well above 100% TDP. The worst is AVX versions of Prime95 (27.7 to 29.4) which can reach nearly 130% TDP. This is the most extreme workload of any other utility. Let's plug this into our example:

CPU TDP = 100 Watts
Workload = 130% TDP
Power draw = 130 Watts

CPU TDP ... 30% Exceeded

Still make sense?

How would AVX affect Core temperatures?

“Stress” tests vary widely and can be characterized into two categories; stability tests which are fluctuating workloads, and thermal tests which are steady workloads. Utilities that don't overload or underload your processor will give you a valid thermal baseline. Here’s a comparison of utilities grouped as thermal and stability tests according to % TDP, averaged across six processor Generations at stock settings rounded to the nearest 5%:

All tests will show 100% CPU Utilization in Windows Task Manager, which is not % TDP workload. Core temperatures respond directly to Power dissipation (Watts), which is driven by workload.

Prime95 v26.6 Small FFT's (pre-AVX) is a true and steady 100% workload, which is the best test you can possibly run for checking thermal performance.

As per Intel's Datasheets, TDP and Thermal Specifications are validated "without AVX". That's why BIOS has an "AVX offset" adjustment. If you don't use AVX apps such as for rendering or transcoding, then don't run utilities that use AVX just to test stability or thermal performance.

• Prime95 v26.6 - http://www.mersenneforum.org/showthread.php?t=15504

When testing, in addition to watching Core temperatures, also watch Package Power to better understand how they correlate to various workloads.

If you want to get yourself up to speed on this topic, then you want to read this: Intel Temperature Guide - http://www.tomshardware.com/forum/id-1800828/intel-temperature-guide.html

It's a "Sticky" at the top of the CPUs Forum.

Once again, welcome aboard!

CT

 

[-Dirt-]

Hi, first post here!

CompuTronix, your recommendation of keeping the 9900k under 80c has me worried. I ran my 9900k with Prime 95 v27.9 small FFTs for 2 hours ONCE! It peaked at 93c with all cores 5Ghz, 1.31v, no AVX offset, and never crashed. It actually never thermal throttled either due to my using a 1000mm rad with nine 120mm fans. Do you think I damaged my chip by it being as hot as it was for about 2 hours...?

The system has never crashed so I bet it's undamaged but just thought I'd check.

 

[mdd1963]

You would likely be at least 5C cooler with Prime95 26.6, which is the defacto standard...

 

[-Dirt-]

You would likely be at least 5C cooler with Prime95 26.6, which is the defacto standard...

Thanks! Temps are way better with 26.6. I've now gotten 5.1Ghz all-core stable at 1.31v with 82c max temp.

 

[rigg42]

Thanks! Temps are way better with 26.6. I've now gotten 5.1Ghz all-core stable at 1.31v with 82c max temp.

That's a great overclocker if that's the case. I doubt you're actually stable though. That's a 5.2ghz quality bin if you are. P95 26.6 small FFT is really only good for checking temp limits and isn't a great stress test. I'd run an extended (4-8hr) realbench stress test with half of your memory. That's about as good of a real world stress test as you can run.

 

[mdd1963]

Agree that Prime95 alone is not the 'end all' of stress/stability testers...(What good is any system that passes Prime95, but then crashes in CInebench, Blender, Furmark, CPU-Z, or in a game? Lots of folks are happy if they can set some sort of goofy record in whatever benchmark, and need a system just stable enough to pass the 2 minute benchmark, but I find that approach an exercise only in fringe 'hypothetical maximums')

But, Prime95/blended and small FFTs is at least a good start, and if greeted with crashing app/blue screens/reboots 1-2 minutes into Prime95, one quickly knows extended further testing at existing settings is unnecessary. (i wanted my 7700K at 4.8 GHz at default core voltage, but, it was not to be, as a blue reboot 'there was a problem' screen quickly let me know. An extra .05V bump made it stable at 4.8 GHz, but boosted temps to by 8C to ~84C, which I find undesirable, especially in my GPU-limited state)

 

[-Dirt-]

That's a great overclocker if that's the case. I doubt you're actually stable though. That's a 5.2ghz quality bin if you are. P95 26.6 small FFT is really only good for checking temp limits and isn't a great stress test. I'd run an extended (4-8hr) realbench stress test with half of your memory. That's about as good of a real world stress test as you can run.

Doesn’t Realbench use AVX?

 

[rigg42]

Some. It's not anything like using a newer version of prime. Temps will peak lower than P95 26.6 small fft. It's generally regarded as an excellent stress test for overclocks.