Power Consumption and Temperatures
Stock CPU, 3600 MHz Memory |
---|
CPU Voltage: | 1.129 V |
---|
DRAM Voltage: | 1.35 V |
---|
Idle Power: | 10 W |
---|
Load Power: | 129 W |
---|
VRM Temperature: | 70.1 °C |
---|
Chipset Temperature: | 38.7 °C |
---|
4.2 GHz CPU, 3600 MHz Memory |
---|
CPU Voltage: | 1.35 V |
---|
DRAM Voltage: | 1.35 V |
---|
Idle Power: | 20 W |
---|
Load Power: | 180 W |
---|
For temperature measurement, I use a Reed SD-947 4 channel Data Logging Thermometer paired with four Omega Engineering SA1 self-adhesive thermocouple probes. One probe directly touches the chipset and two are placed on select power stages. The last probe actively logs the ambient temperature.
For the BIOSTAR B550GTQ, each probe is spaced one third of the way along the Vcore bank of power stages. A probe is left out to log the ambient temperature. All temperatures are presented as Delta-T normalized to 20 °C, which is the measured temperature minus the ambient temperature plus 20 °C. The end result accounts for variation in ambient temperature, including changes over the course of a test, while presenting the data as if the ambient were a steady 20 °C for easy presentation. Additionally, there is no longer any direct airflow over the VRM with this new setup, placing extra strain on the VRM cooling.
For the numbers seen in the chart above, I am now using Prime95's Small FFT test for power consumption. For temperatures, I am using the maximum temperatures recorded over the course of my standard benchmark suite (almost always during either wPrime or Blender tests). However, relatively short tests do not put enough strain on the system to get a look at how the VRM performs at the limit, so I added an additional test to try to thermally abuse Vcore as much as possible.
This test typically involves a 30 minute Prime95 run at the maximum overclock the motherboard can maintain, again with no airflow over the VRM. For B550, I chose 4.2 GHz at 1.35 V as the most intensive load in long tests without thermal throttling the CPU. Temperatures are logged every second, and the two probes are then averaged for a cleaner presentation before subtracting the ambient to calculate the Delta-T. The results are charted below.
The BIOSTAR B550GTQ proved unable to maintain my standard overclock of 4.2 GHz at 1.35 V. While at first I saw extreme throttling with CPU speed dropping to well below 3 GHz every few seconds, I was able to resolve this by disabling the "A.I. TP Control" setting in BIOS. Unfortunately, once the throttling was disabled, the board was either unstable, crashing after a couple of minutes of testing, or would put too much power through the CPU, which had the 3900X hit unsafe temperatures.
Ultimately, I decided to run the Prime 95 test at completely stock settings, which resulted in a steady power draw of around 125 watts (measured socket draw using Ryzen Master). For reference, this is 50–70 watts below what most boards pull for the 4.2 GHz overclock. At this stock power draw, the B550GTQ VRM still saw temperatures above 95 °C. Given this thermal result, I would not recommend overclocking 125 W TDP Ryzen CPUs on this motherboard regardless of stability. 65 W parts such as the Ryzen 5 3600 should present no issue even when overclocked.
Biostar could improve this VRM design by dedicating more power stages to the Vcore, reducing the number of SOC stages from 4 to 2. It is also worth noting that power stages are not everything in VRM design. The Gigabyte B550I AORUS Pro AX uses six of the same Intersil ISL99390 power stages for it's VRM and displayed significantly lower temperatures.
The BIOSTAR B550GTQ does get credit for using high end power stages, a rare feature at this price point, which are both more efficient and feature protective features like overcurrent and over temperature protection.