[SOLVED] Overheating VRM's

Oct 9, 2020
68
9
45
4
Hi all. I have a Gigabyte Aorus Z370 Ultra Gaming 1.0 motherboard with an i5 8600k running @5ghz 1.38v. When running cinebench r23 my cpu draws around 130w-140w and my vrms hit around 90°c which is perfectly within spec. But if I try and run OCCT's cpu stress test with the setting set to:
Data Set - Small
Mode - Extreme
Load type - Steady

During this test my cpu pulls around 160w-170w and my vrms quickly reach 127°c of which is the temperature they throttle at. While there is no real load that I put my computer though that equals to what this benchmark does, I am planning on getting a i9 9900kf which when overclocked will pull around 160w-170w, which my vrms currently struggle with. I have a fan blowing fresh air over them at the moment but even at max speed that cannot keep them cool, what is there that I could do to help reduce temperatures?

If I cannot find a solution I will most likely get an x570 motherboard and a 3900x or 3950x.
 

CompuTronix

Intel Master
Moderator
... Gigabyte Aorus Z370 Ultra Gaming ... i5 8600k ... 5ghz 1.38v ... cinebench r23 ... 130w-140w ... vrms ... 90°c

OCCT's cpu stress test ... set to:
Data Set - Small
Mode - Extreme
Load type - Steady

... 160w-170w ... vrms ... 127°c ... planning on getting a i9 9900kf ... overclocked ... around 160w-170w ...
In OCCT settings you apparently overlooked "Instruction Set", which you no doubt left in "Auto", meaning that you inadvertently ran an AVX/2 workload.

As per Intel’s Datasheets, TDP and Thermal Specifications are validated “without AVX.

If you repeat the OCCT CPU test with Instruction Set configured for "SSE", then you'll see power consumption and temperatures which are nearly identical to CineBench R23.

Although CineBench R23 uses AVX code in a real-world rendering workload, the AVX code is much less intense than the AVX code used in OCCT and Prime95. When OCCT is configured for Small, Normal, Steady and SSE, or when Prime95 is configured for Small FFTs with all AVX test sections disabled, then power and temperatures will be nearly identical between these 3 test utilities. The principle difference is that when configured as specified, OCCT and Prime95 are steady-state workloads which are ideal for testing power consumption and thermal performance, whereas CineBench is a fluctuating workload that pauses between rendering cycles, and is more useful for testing stability.

https://siliconlottery.com/pages/statistics
According to silicon lottery the 8600k runs 5Ghz @1.4Vcore while the 9900k runs 5Ghz at 1.3Vcore so it should actually be running cooler.
Although our friends at Silicon Lottery furnish some fascinating statistics, respectfully, we can't compare a 6 Core 6 Thread CPU with an 8 Core 16 Thread CPU, then extrapolate how Core count, Hyper-Threading, voltage and package power consumption between 2 very different processors will interact to affect temperatures. This would be like attempting to compare the variables between apples and oranges. However, stanley88845 is correct where he said "the voltage may be lower with the 9900k but it will still draw more power, putting more stress on the vrms".

Also everything you never asked to know about heat in intel CPUs.
stanley88845, what he meant is that if you're interested in learning more about this topic, then you should read our temperature guide. Just click on the link he provided for you, or click on the link in my signature.

CT :sol:
 
Reactions: stanley88845
https://siliconlottery.com/pages/statistics
According to silicon lottery the 8600k runs 5Ghz @1.4Vcore while the 9900k runs 5Ghz at 1.3Vcore so it should actually be running cooler.

Also everything you never asked to know about heat in intel CPUs.
 

CompuTronix

Intel Master
Moderator
... Gigabyte Aorus Z370 Ultra Gaming ... i5 8600k ... 5ghz 1.38v ... cinebench r23 ... 130w-140w ... vrms ... 90°c

OCCT's cpu stress test ... set to:
Data Set - Small
Mode - Extreme
Load type - Steady

... 160w-170w ... vrms ... 127°c ... planning on getting a i9 9900kf ... overclocked ... around 160w-170w ...
In OCCT settings you apparently overlooked "Instruction Set", which you no doubt left in "Auto", meaning that you inadvertently ran an AVX/2 workload.

As per Intel’s Datasheets, TDP and Thermal Specifications are validated “without AVX.

If you repeat the OCCT CPU test with Instruction Set configured for "SSE", then you'll see power consumption and temperatures which are nearly identical to CineBench R23.

Although CineBench R23 uses AVX code in a real-world rendering workload, the AVX code is much less intense than the AVX code used in OCCT and Prime95. When OCCT is configured for Small, Normal, Steady and SSE, or when Prime95 is configured for Small FFTs with all AVX test sections disabled, then power and temperatures will be nearly identical between these 3 test utilities. The principle difference is that when configured as specified, OCCT and Prime95 are steady-state workloads which are ideal for testing power consumption and thermal performance, whereas CineBench is a fluctuating workload that pauses between rendering cycles, and is more useful for testing stability.

https://siliconlottery.com/pages/statistics
According to silicon lottery the 8600k runs 5Ghz @1.4Vcore while the 9900k runs 5Ghz at 1.3Vcore so it should actually be running cooler.
Although our friends at Silicon Lottery furnish some fascinating statistics, respectfully, we can't compare a 6 Core 6 Thread CPU with an 8 Core 16 Thread CPU, then extrapolate how Core count, Hyper-Threading, voltage and package power consumption between 2 very different processors will interact to affect temperatures. This would be like attempting to compare the variables between apples and oranges. However, stanley88845 is correct where he said "the voltage may be lower with the 9900k but it will still draw more power, putting more stress on the vrms".

Also everything you never asked to know about heat in intel CPUs.
stanley88845, what he meant is that if you're interested in learning more about this topic, then you should read our temperature guide. Just click on the link he provided for you, or click on the link in my signature.

CT :sol:
 
Reactions: stanley88845
Although our friends at Silicon Lottery furnish some fascinating statistics, respectfully, we can't compare a 6 Core 6 Thread CPU with an 8 Core 16 Thread CPU, then extrapolate how Core count, Hyper-Threading, voltage and package power consumption between 2 very different processors will interact to affect temperatures. This would be like attempting to compare the variables between apples and oranges. However, stanley88845 is correct where he said "the voltage may be lower with the 9900k but it will still draw more power, putting more stress on the vrms".
Well he said that he would expect the new CPU to draw the same watts for normal O/C as he is already drawing with the 8600k with the "extreme" stresstest with avx/2.
So with the same watts but less Vcore I assumed that it could only be less heat but sure, I'm no expert I'm not gonna ride around on that.
 

CompuTronix

Intel Master
Moderator
Good point, which brings up the importance of defining the conditions of our test procedures.

First, there's no discernable difference between the CPU test Mode in OCCT being set to "Normal" or "Extreme". Check it out for yourself.

Second, under the OCCT test conditions that stanley88845 used, which apparently included AVX/2, a 9900K overclocked to 5.0 GHz at 1.3 volts with an AVX Offset 0 in BIOS will not consume just 160 to 170 Watts, but instead may consume more than 240 Watts. Unfortunately, when under such high CPU power consumption, his Gigabyte Aorus Z370 Ultra Gaming motherboard VRMs would quickly throttle, which are not designed to properly support the power delivery required for a 9900K, and is why most Z390 motherboards have greatly improved VRMs.

CT :sol:
 
Reactions: stanley88845
I meant extreme as in the most demanding not as in the actual setting sorry, but will a 9900k overclocked only for SSE or a basic O/C however you want to call it draw 30% less power, as the temp guide tells us avx is 30% above normal TDP?! 240W - 30% would be around 170w so it would draw as much as his 8600k under avx/2?
 
Oct 9, 2020
68
9
45
4
In OCCT settings you apparently overlooked "Instruction Set", which you no doubt left in "Auto", meaning that you inadvertently ran an AVX/2 workload.

As per Intel’s Datasheets, TDP and Thermal Specifications are validated “without AVX.

If you repeat the OCCT CPU test with Instruction Set configured for "SSE", then you'll see power consumption and temperatures which are nearly identical to CineBench R23.

Although CineBench R23 uses AVX code in a real-world rendering workload, the AVX code is much less intense than the AVX code used in OCCT and Prime95. When OCCT is configured for Small, Normal, Steady and SSE, or when Prime95 is configured for Small FFTs with all AVX test sections disabled, then power and temperatures will be nearly identical between these 3 test utilities. The principle difference is that when configured as specified, OCCT and Prime95 are steady-state workloads which are ideal for testing power consumption and thermal performance, whereas CineBench is a fluctuating workload that pauses between rendering cycles, and is more useful for testing stability.


Although our friends at Silicon Lottery furnish some fascinating statistics, respectfully, we can't compare a 6 Core 6 Thread CPU with an 8 Core 16 Thread CPU, then extrapolate how Core count, Hyper-Threading, voltage and package power consumption between 2 very different processors will interact to affect temperatures. This would be like attempting to compare the variables between apples and oranges. However, stanley88845 is correct where he said "the voltage may be lower with the 9900k but it will still draw more power, putting more stress on the vrms".


stanley88845, what he meant is that if you're interested in learning more about this topic, then you should read our temperature guide. Just click on the link he provided for you, or click on the link in my signature.

CT :sol:
Absolutely brilliant, thankyou so much.
 
Oct 9, 2020
68
9
45
4
Good point, which brings up the importance of defining the conditions of our test procedures.

First, there's no discernable difference between the CPU test Mode in OCCT being set to "Normal" or "Extreme". Check it out for yourself.

Second, under the OCCT test conditions that stanley88845 used, which apparently included AVX/2, a 9900K overclocked to 5.0 GHz at 1.3 volts with an AVX Offset 0 in BIOS will not consume just 160 to 170 Watts, but instead may consume more than 240 Watts. Unfortunately, when under such high CPU power consumption, his Gigabyte Aorus Z370 Ultra Gaming motherboard VRMs would quickly throttle, which are not designed to properly support the power delivery required for a 9900K, and is why most Z390 motherboards have greatly improved VRMs.

CT :sol:
The 8600k under the test conditions I put it under is decently similar to the sort of load that a 9900k would put on the vrms on Cinebench R23, right?
 

CompuTronix

Intel Master
Moderator
I meant extreme as in the most demanding not as in the actual setting sorry, but will a 9900k overclocked only for SSE or a basic O/C however you want to call it draw 30% less power, as the temp guide tells us avx is 30% above normal TDP?! 240W - 30% would be around 170w so it would draw as much as his 8600k under avx/2?
My exact words were "more than 240 Watts", which, depending upon Vcore, can be a bit conservative for an OC'd 9900K running an AVX workload. The temp guide states at Intel stock settings, P95 Small FFTs with AVX is about 130% workload, or 100% workload without AVX. stanley88845, instead ran OCCT, CPU, Small, Extreme, Steady, Instruction Set "Auto" (AVX/2), which is about 125% TDP workload. However, while the differences in these tests might seem trivial, when you factor in the differences for 2 additional Cores, 4 additional Threads, 9 MB Cache versus 16 MB Cache, no Hyper-Threading versus Hyper-Threading and differences in Core voltage, that's a lot of variables which all determine the power equation, so the answer isn't straight forward enough to generalize. I'll elaborate below.
The 8600k under the test conditions I put it under is decently similar to the sort of load that a 9900k would put on the vrms on Cinebench R23, right?
Not quite so; again, we need to be very specific when defining test conditions due to the number of variables involved.

When properly configured for equivalent features, P95 and OCCT have nearly identical workloads and power consumption with AVX / AVX2, or without AVX / AVX2. However, without AVX / AVX2, the same P95 and OCCT tests have power consumption nearly identical to CineBench R23 (as well as R20), although the CineBench workload characteristics differ somewhat from P95 Small and OCCT Small, as detailed in my first post.

If you mean comparing your 6 Core 6 Thread 8600K OC'd to 5.0 @ 1.38 Vcore tested on OCCT with a synthetic AVX/2 workload at 160 to 170 Watts, to an 8 Core 16 thread 9900K OC'd to 5.0 @ 1.3 Vcore tested on CineBench R23 with an AVX/2 real-world workload at a projected 160 to 170 Watts, then you're guesstimation is a bit low.

A few hours ago I tested an 8600K OC'd to 5.0 @ 1.364 Vcore with OCCT, CPU, Small, Extreme, Steady, Instruction Set Auto (AVX/2), which closely replicates the settings you used. All tests were run for 10 minutes. OCCT power consumption stabilized at 185 Watts, which differs from the 160 to 170 Watts you posted, possibly due to your test duration being shorter. I then ran your CineBench test, where power consumption stabilized at 128 Watts, which is a bit below the 130 to 140 Watts you posted, possibly due to our test sample running at a slightly lower Vcore. Also, ambient temperature and cooling haven't been mentioned, so it's quite likely that our results differ in those aspects as well. All tests were performed at 21°C ambient.

Many users are unaware that Core temperatures affect power consumption. As temperature increases, leakage current increases, which in turn increases power consumption. In electronics, although this runs contrary to Ohms Law, where resistance increases with temperature, leakage current more than cancels out the effect. This means hotter CPUs consume more power at the same Vcore than cooler CPUs. Moreover, at Intel stock settings, Hyper-Threading increases power consumption by about 12% and Core temperatures by 4 to 6°C. The difference can be considerably higher depending on how much a CPU is overclocked. Additionally, delidding not only reduces Core temperatures significantly, but in turn, it also reduces power consumption.

We don't happen to have a 9900K at hand just this moment to repeat the tests, nor can I accurately recall previous 9900K test results, however, I tested an 8700K OC'd to 5.0 @ 1.332 Vcore. I repeated your OCCT test, but instead with SSE (no AVX). Power consumption stabilized at 156 Watts. CineBench R23 stabilized at 160 Watts, while Prime95 Small FFTs, no AVX stabilized at 161 Watts. Based on these findings, if we add a proportionate increase in power consumption for the additional Cores, Threads and Cache, then my guesstimate would put your hypothetical 9900K OC'd to 5.0 @ 1.3 Vcore at about 210 Watts, which is a familiar value.

This means that when running a valid 100% workload, your motherboard VRMs will not be capable of supporting the power delivery required to sustain an OC'd 9900K. Nonetheless, I again tested the 8700K, but with CPU-Z > Bench > Stress CPU, which is a steady-state workload at about 80% that's more indicative of peak workloads during games that are highly CPU intensive. (Keep in mind that % CPU Utilization is processor resource activity; NOT actual % TDP workload.) Power consumption stabilized at 123 Watts, which would calculate to about 164 Watts for your hypothetical OC'd 9900K. This suggests that if you limit thermal and stability stress tests to utilities that don't exceed 80% workloads, and you only game without using rendering or transcoding software, then you might be able to squeeze by.

If you read Section 11 - Thermal Test Basics in our Intel CPU Temperature Guide 2021, as TerryLaze and I both previously suggested, and look at figure 11-2, which is the %TDP scale, you'll get a better perspective of how much different steady-state and fluctuating workloads in stress test utilities can actually vary, which directly impacts power consumption and temperature. For example, while utilities such as CineBench and Asus RealBench run real-world AVX code, just as does rendering or transcoding software, the synthetic AVX code in OCCT and Prime95 is much more intense. This is why real-world AVX workloads typically may approach, but not exceed, OCCT or P95 without AVX, and is again why Intel's Datasheets state that TDP and Thermal Specifications are validated without AVX. So even if you run software that uses AVX, it's misleading to enable AVX worloads in test utilities if you expect to see real-world temperature and power consumption values.

The bottom line is that providing you only use your rig for gaming, your present motherboard is marginal, at best, for you to consider upgrading from your 8600K to a 9900K. However, if your case allows enough access to the back of your motherboard, check the screws which fasten the heat sinks to the VRMs to ensure that they're at least snug. Even if you can gain a quarter turn, it may help to drop your VRM temperature several degrees. If you decide to purchase the 9900K, then I suggest that you run CPU-Z > Bench > Stress CPU, which is steady-state workload at about 80% to test thermal performance, and Intel XTU, which is a fluctuating workload that's also about 80% to test stability. By using these particular 80% workload stress tests, your CPU will require less Vcore and consume less power, while you'll still be able to achieve a stable overclock without pushing your VRMs to their thermal throttle limit.

CT :sol:
 
Reactions: stanley88845

ASK THE COMMUNITY

TRENDING THREADS