Intel Core i9-10900K Stress Tested: Comet Lake Flagship Hits 93°C

News
By Zhiye Liu
last updated

Power draw peaked at 235W.

Intel Core i9-10900K

(Image credit: Intel)

A Chinese Weibo user has pushed a flagship Intel Comet Lake-S CPU to its limits. The tester cooked an Intel COre i9-10900K for a little over 47 minutes with its  power draw and maximum temperatures hitting 235W and 93 degrees Celsius (199.4 degrees Fahrenheit), respectively.

The i9-10900K features a 10-core, 20-thread design with a 3.7 GHz base clock and a 5.3 GHz Thermal Velocity Boost (TVB) clock. Intel sells the i9-10900K as a 125W CPU. However, the deca-core processor has configurable TDP (cTDP) of 95W that essentially drops the base clock to 3.3 GHz. The PL1 (power level 1) is only one side of the coin, and the side that Intel openly markets.

On the flip side, the i9-10900K has a PL2 (power level 2) of 250W, which is the maximum power draw when turbo is engaged. Even at 235W, the i9-10900K is operating within Intel's specifications; but if there were any doubts on whether the chip can pull as much power, the Weibo user may have just laid them to rest.

(Image credit: WolStame/Weibo)

When you have a processor pulling over 200W of power, the amount of generated heat is going to be off the charts. The tester ran the i9-10900K inside Lenovo's latest Savior Blade 9000K, which is a pre-built machine exclusive to the Chinese market. A Legion-branded 240mm AIO liquid cooler was responsible for keeping the i9-10900K's temperatures in check.

For the duration of the stress testing, HWiNFO64 registered average and maximum temperatures of 87 degrees (188.6 degrees Fahrenheit) and 93 degrees Celsius, respectively. The reviewer mentioned that the room temperature was 24 degrees, however, he didn't specify if it was Fahrenheit or Celsius. We suspect it's the latter since China uses the metric system. The poor Legion cooler was able to maintain the i9-10900K at 4.8 GHz, which is the chip's all-core boost clock. However, you can probably imagine the level of noise coming out of the PC case.

The biggest drawback with operating temperatures this high is that you won't get to enjoy the i9-10900K's full potential. The i9-10900K leverages Intel's Thermal Velocity Boost and can hit 4.9 GHz on all cores, as long as the operating temperature is below 70 degrees Celsius (158 degrees Fahrenheit). Once you cross that threshold, you're losing 100 MHz of performance on every core. 

If cooling was an important factor before, it's even more crucial with Intel's Comet Lake-S processors.

Zhiye Liu
Zhiye Liu
RAM Reviewer and News Editor

Zhiye Liu is a Freelance News Writer at Tom’s Hardware US. Although he loves everything that’s hardware, he has a soft spot for CPUs, GPUs, and RAM.

See more CPUs News
41 Comments Comment from the forums
  • AlistairAB
    93 degrees at stock settings with a liquid cooler. No thanks. This CPU is just a mistake.
    Reply
  • Metal Messiah.
    Yup. I've seen this info before. Also posted an article on this. I was just wondering how other CPUs are going to perform in the Comet Lake S lineup.

    The all-core and single core boost figures seems to be just for marketing purposes, IMO. Intel claims up to 5.3 / 4.0 GHz Thermal Velocity Boost Singe / All-core Turbo, though they fail to mention these speeds are only going to be valid in "short bursts", for interval of time.

    https://forums.tomshardware.com/threads/intel-i9-10900k-10-core-flagship-cpu-tested-runs-very-hot-consumes-235w-at-4-8-ghz-with-over-90c-temps.3605203/
    Reply
  • PCWarrior
    •It is 87C average with 88C current. 93C was just the maximum spike and likely occurred when the cpu tried to briefly boost to its TV boost of 4.9GHz before downclocking to 4.8GHz.
    •It was cooled by a 240mm water cooler with a thin rad. This 240mm watercooler is no better than a decent air-cooler and inferior to many good air-coolers like the Noctua NH U14S (and certainly inferior to the best air-coolers like the Noctua NH D15 that trade blows with the best 280mm watercoolers).
    •It was tested with a load equivalent to Prime 95 small FFTs with AVX. That is like testing a building for an earthquake of 12degrees in the Richter scale, an earthquake that will never ever happen (well not until the sun goes nova or comets fell on Earth or some other such cataclysmic event occurs)
    •It is with power-limits removed. Had power limits been enabled the CPU would be forced to downclock to a lower frequency where power consumption would stay at 125W. For this workload that frequency would be 4.0GHz-4.3GHz.
    •As shown, with power limits removed, the CPU has no problem to turbo indefinitely at its all-core turbo 2.0 speed of 4.8GHz. And had temps stayed below 70C it would be turboing indefinitely to 4.9GHz.
    •The 5.3GHz is the turbo with only 1 or 2 cores active. Turbo depends on the workload’s needs. The nature of most workloads only requires brief frequency boosts. If the workload is 1-2core and bursty the turbo will be bursty. If the workload is 1-2core and sustained then the 5.3GHz turbo will be sustained too (i.e. for as long as the workload is active). If for example you run Cinebench R15/R20 single- threaded (or dual-threaded) this CPU will have no problem maintaining that 5.3GHz speed for the entire duration of the test, even without removing power limits and even just having a simple 212EVO to cool it. In the worst-case scenario, temps will be above 70C in which case you will get a sustained 5.2GHz (which is the turbo boost 3.0 frequency which is not depended on temperature – other than TJmax that is).
    Reply
  • Zizo007
    PCWarrior said:
    •It is 87C average with 88C current. 93C was just the maximum spike and likely occurred when the cpu tried to briefly boost to its TV boost of 4.9GHz before downclocking to 4.8GHz.
    •It was cooled by a 240mm water cooler with a thin rad. This 240mm watercooler is no better than a decent air-cooler and inferior to many good air-coolers like the Noctua NH U14S (and certainly inferior to the best air-coolers like the Noctua NH D15 that trade blows with the best 280mm watercoolers).
    •It was tested with a load equivalent to Prime 95 small FFTs with AVX. That is like testing a building for an earthquake of 12degrees in the Richter scale, an earthquake that will never ever happen (well not until the sun goes nova or comets fell on Earth or some other such cataclysmic event occurs)
    •It is with power-limits removed. Had power limits been enabled the CPU would be forced to downclock to a lower frequency where power consumption would stay at 125W. For this workload that frequency would be 4.0GHz-4.3GHz.
    •As shown, with power limits removed, the CPU has no problem to turbo indefinitely at its all-core turbo 2.0 speed of 4.8GHz. And had temps stayed below 70C it would be turboing indefinitely to 4.9GHz.
    •The 5.3GHz is the turbo with only 1 or 2 cores active. Turbo depends on the workload’s needs. The nature of most workloads only requires brief frequency boosts. If the workload is 1-2core and bursty the turbo will be bursty. If the workload is 1-2core and sustained then the 5.3GHz turbo will be sustained too (i.e. for as long as the workload is active). If for example you run Cinebench R15/R20 single- threaded (or dual-threaded) this CPU will have no problem maintaining that 5.3GHz speed for the entire duration of the test, even without removing power limits and even just having a simple 212EVO to cool it. In the worst-case scenario, temps will be above 70C in which case you will get a sustained 5.2GHz (which is the turbo boost 3.0 frequency which is not depended on temperature – other than TJmax that is).
    Its still a volcano requiring a nuclear power plant 🔥🔥🔥
    Reply
  • NightHawkRMX
    A Legion branded 240mm aio is a little vague. Really will have to wait until we get a name brand AIO tested on these CPUs not in an OEM case so that we can compare it to the same cooler on existing CPUs.

    PCWarrior said:
    It was cooled by a 240mm water cooler with a thin rad. This 240mm watercooler is no better than a decent air-cooler and inferior to many good air-coolers like the Noctua NH U14S (and certainly inferior to the best air-coolers like the Noctua NH D15 that trade blows with the best 280mm watercoolers).
    You cannot say this cooler is any worse nor any better than a specific air cooler since there is no test of those air coolers against this particular Legion branded AIO on this CPU. Also, case airflow would have played a part. A good 240mm AIO can tie expensive air coolers, or it can lose to them. It really depends on the exact AIO and exact air cooler tested. Not all 240mm aios are the same.
    PCWarrior said:
    It is with power-limits removed. Had power limits been enabled the CPU would be forced to downclock to a lower frequency where power consumption would stay at 125W. For this workload, that frequency would be 4.0GHz-4.3GHz.
    Most motherboards disabled these limits by default (Even this OEM board has these limits disabled), so this is more or less out of the box performance. Any CPU can run cool if you purposely downclock it below out of the box operation which lowers performance greatly.
    You basically are guessing what frequency that CPU would run at with those limits enabled since the CPU is not released and you cannot actually test it.
    Reply
  • AlistairAB
    PCWarrior said:
    •It is 87C average with 88C current. 93C was just the maximum spike and likely occurred when the cpu tried to briefly boost to its TV boost of 4.9GHz before downclocking to 4.8GHz.
    •It was cooled by a 240mm water cooler with a thin rad. This 240mm watercooler is no better than a decent air-cooler and inferior to many good air-coolers like the Noctua NH U14S (and certainly inferior to the best air-coolers like the Noctua NH D15 that trade blows with the best 280mm watercoolers).
    •It was tested with a load equivalent to Prime 95 small FFTs with AVX. That is like testing a building for an earthquake of 12degrees in the Richter scale, an earthquake that will never ever happen (well not until the sun goes nova or comets fell on Earth or some other such cataclysmic event occurs)
    •It is with power-limits removed. Had power limits been enabled the CPU would be forced to downclock to a lower frequency where power consumption would stay at 125W. For this workload that frequency would be 4.0GHz-4.3GHz.
    •As shown, with power limits removed, the CPU has no problem to turbo indefinitely at its all-core turbo 2.0 speed of 4.8GHz. And had temps stayed below 70C it would be turboing indefinitely to 4.9GHz.
    •The 5.3GHz is the turbo with only 1 or 2 cores active. Turbo depends on the workload’s needs. The nature of most workloads only requires brief frequency boosts. If the workload is 1-2core and bursty the turbo will be bursty. If the workload is 1-2core and sustained then the 5.3GHz turbo will be sustained too (i.e. for as long as the workload is active). If for example you run Cinebench R15/R20 single- threaded (or dual-threaded) this CPU will have no problem maintaining that 5.3GHz speed for the entire duration of the test, even without removing power limits and even just having a simple 212EVO to cool it. In the worst-case scenario, temps will be above 70C in which case you will get a sustained 5.2GHz (which is the turbo boost 3.0 frequency which is not depended on temperature – other than TJmax that is).

    Yeah not so fast. Loading in Frostbite games uses AVX and gets to 100 percent and gets nearly as hot as Prime95, I don't want my fans hitting max temp every time I load into a level.
    Reply
  • NightHawkRMX
    PCWarrior said:
    If for example you run Cinebench R15/R20 single- threaded (or dual-threaded) this CPU will have no problem maintaining that 5.3GHz speed for the entire duration of the test, even without removing power limits and even just having a simple 212EVO to cool it
    Sure, if you load the 1 or 2 of the 20 threads and stress the CPU, you don't need a fancy cooler and the CPU doesn't draw as much power since you are only loading one-twentieth or one-tenth of the CPU.

    Most modern software can utilize far more than 1 or 2 threads, so whatever point you were trying to make is irrelevant for 99% of the buyers of this CPU.
    Reply
  • Gurg
    PCWarrior said:
    •It is 87C average with 88C current. 93C was just the maximum spike and likely occurred when the cpu tried to briefly boost to its TV boost of 4.9GHz before downclocking to 4.8GHz.
    •It was cooled by a 240mm water cooler with a thin rad. This 240mm watercooler is no better than a decent air-cooler and inferior to many good air-coolers like the Noctua NH U14S (and certainly inferior to the best air-coolers like the Noctua NH D15 that trade blows with the best 280mm watercoolers).
    •It was tested with a load equivalent to Prime 95 small FFTs with AVX. That is like testing a building for an earthquake of 12degrees in the Richter scale, an earthquake that will never ever happen (well not until the sun goes nova or comets fell on Earth or some other such cataclysmic event occurs)
    •It is with power-limits removed. Had power limits been enabled the CPU would be forced to downclock to a lower frequency where power consumption would stay at 125W. For this workload that frequency would be 4.0GHz-4.3GHz.
    •As shown, with power limits removed, the CPU has no problem to turbo indefinitely at its all-core turbo 2.0 speed of 4.8GHz. And had temps stayed below 70C it would be turboing indefinitely to 4.9GHz.
    •The 5.3GHz is the turbo with only 1 or 2 cores active. Turbo depends on the workload’s needs. The nature of most workloads only requires brief frequency boosts. If the workload is 1-2core and bursty the turbo will be bursty. If the workload is 1-2core and sustained then the 5.3GHz turbo will be sustained too (i.e. for as long as the workload is active). If for example you run Cinebench R15/R20 single- threaded (or dual-threaded) this CPU will have no problem maintaining that 5.3GHz speed for the entire duration of the test, even without removing power limits and even just having a simple 212EVO to cool it. In the worst-case scenario, temps will be above 70C in which case you will get a sustained 5.2GHz (which is the turbo boost 3.0 frequency which is not depended on temperature – other than TJmax that is).
    Years ago I called out Toms'Hardware for claiming a new AMD product was a fire breather in review tests. TH had run the test in the hot apartment (Arizona?Nevada?) of the tester where the air conditioning was having trouble maintaining 76º F. At that time I indicated that testing should occur at temperature conditions of 70-72º which approximate the ideal average office temp for maximum productivity in the US.

    Tom's Hardware has since adopted that standard for its reviews.

    There is absolutely no indication of the ambient air temperatures in which this run was conducted.
    Reply
  • Soaptrail
    AlistairAB said:
    Yeah not so fast. Loading in Frostbite games uses AVX and gets to 100 percent and gets nearly as hot as Prime95, I don't want my fans hitting max temp every time I load into a level.

    Don't worry it will be a very short wait for the next next gen Intel CPU's to be released. :P
    Reply
  • mdd1963
    So, 87-88C even at only 4.8 GHz? Not promising...

    We will hopefully see what it does on competent (and apparently needed) larger AIOs within a month or so....
    Reply