How We Test Graphics Cards

Power Consumption And Noise Levels

Measuring Power Consumption

The time we've invested into measuring graphics card power consumption correctly is immense...but worth it. If you've already read The Math Behind GPU Power Consumption And PSUs, you already know why. If not, we suggest you check it out. 

Let's start small. Not every graphics card has an auxiliary power connector. Some get all of the power they needed from the motherboard using 3.3V and 12V rails. We measure those rails between the graphics card and motherboard through loops on the riser card.

The voltage readings are also measured at the motherboard's 24-pin ATX connector, since that's where the rails begin their journey to the graphics card slot.

Values corresponding to auxiliary six- and eight-pin connectors are important as well. They can only be measured through their cables, right where they feed into the card. Since some cards with multiple connectors split them up to drive different power phases, it's sometimes necessary to use two amp clamps and probes, adding up to eight analog channels we monitor and log simultaneously.

All of this is implemented through two triggered oscilloscopes (master/slave) that act like a single eight-channel scope. Measurements are recorded through an active low-pass filter at the highest sampling rate possible in order to minimize aliasing effects and interfering noise as much as possible.

How Much Detail Do We Actually Need?

This is a good question that, unfortunately, is complicated to answer. We are sometimes criticized in our reviews for the way results are presented. Rightly so, perhaps. If you don't already have an in-depth understanding of the material, it's easy to overlook certain details or outright misinterpret them.

Recently, to make the data we're collecting more accessible, we increased the measurement intervals and introduced new software that's able to analyze short load spikes (and drops), checking them for plausibility. The resulting curve is a lot flatter, but also easier to interpret. Knowing that spikes can and do occur is important; they can be omitted, though, if it makes everything else we present more understandable.

Power Consumption
Test Method
Contact-free DC Measurement at PCIe Slot (Using a Riser Card)
Contact-free DC Measurement at External Auxiliary Power Supply Cable
Direct Voltage Measurement at Power Supply
Test Equipment
2 x Rohde & Schwarz HMO 3054, 500MHz Digital Multi-Channel Oscilloscope with Storage Function
4 x Rohde & Schwarz HZO50 Current Probe (1mA-30A, 100kHz, DC)
4 x Rohde & Schwarz HZ355 (10:1 Probes, 500MHz)
1 x Rohde & Schwarz HMC 8012 Digital Multimeter with Storage Function

Measuring Noise

We have another room we use for measuring noise, outside of our German lab. It's actually a room within a room, which helps us isolate against vibration and structure-borne noise (low sounds transmitted across the floor) to a large extent.

Especially when it comes to measuring the noise from quiet graphics cards, conventional sound level meters offer little more than optimistic estimates. They're definitely not accurate. Depending on the time of day and what's going on outside, our chamber makes it possible to record values as low as 20 dB(A) in a reproducible way.

Our previous-gen reference system serves as the foundation of our audio measurements. It lives on an open-air test bench that we made even quieter: without a graphics card and after insulating the power supply, it generates a base noise level of less than 23 dB(A) from 20 inches (50cm) away.

Each test run can be monitored from a separate control room. Just like everything else on this floor, the devices used in this room are exclusively fed via the DC power network.

Time and again we are asked why we present our results in dB(A) instead of Sone. The reason is simple: The definition of the so-called loudness in Sone is based on the definition of the volume level. A sine wave of 1kHz pitch at 40 dB (decibels) corresponds to a volume level of 40 Phon, which is, in turn, the base value for one Sone. If the sound feels twice as loud, the result is called two Sone. This seems practical, logical, and convenient enough. In practice, though, it can be difficult to apply.

Sone is only indicative of how loud an event can be perceived subjectively by an average human being. In that way, it's just another example of problematic psychoacoustics, and thus no better than its dB(A) counterpart. After all, nothing is more inaccurate than subjective impressions. Why? Test for yourself.

  • Take a radio and remember the value on its volume control scale.
  • Then, without looking at the scale, change the loudness to a value that feels twice as loud.
  • Next, (again without looking at the scale) return the loudness to what feels like half the current loudness.
  • Finally, compare the new position of the knob with its initial position. Unless you have a musician's trained ear, the result will probably surprise you. 

The only reasonably reliable method to determine loudness in Sone is the one described by E. Zwicker (implemented in DIN 45631:1991-03). However, it's somewhat complicated, and for quieter sounds below a value of one Phon (approximately 40 dB), it is still quite inaccurate. An unweighted dB measurement represents an actual level of sound pressure. It is thus a real measured value, rather than some event weighted using formulas. This is also why industry likes to use dB as a base unit. Unfortunately, dB isn't of much use to us either.

By instead using A-weighted dB values, we get filtered decibel results that are supposed to represent subjective physiological audio perception. Ironically, this brings us full circle. Sone and dB(A) are interpretations that do not stand for real measured values, but are based on assumptions. A result in dB(A) is an A-weighted sound pressure level based on a hearing threshold of 0 dB = 20 micro-Pascal. It allows for other factors, such as time weighting (fast or slow), to make the result mimic reality a little more closely. So where do we go from here?

The question remains: what exactly should be measured in practice? A-weighted sound pressure level or loudness? At least for now, we're using the former primarily for its better comparability and reproducibility. It's also worth mentioning that the equipment we're using has to be regularly and professionally calibrated to prevent deviations from casting doubt on our findings.

Simple Control Measurements

We can perform these control measurements in the lab at different distances to the object we're measuring. Of course, this is no substitute for an exact measurement, in spite of the calibrated microphone, good shielding, and similar analysis software as what we use in our measuring room. Still, it's good enough to give you an idea of what we're doing.

Setup For Measuring Noise Levels
MicrophoneNTI Audio M2211 (with Calibration File, Low Cut at 50Hz)
AmplifierSteinberg UR12 (with Phantom Power for Microphones)
Creative X7
SoftwareSmaart v.7
Measuring Chamber
Custom-Made Proprietary Measurement Chamber, 3.5 x 1.8 x 2.2m (LxDxH)
Measurement Position
Perpendicular to Center of Noise Source(s), Measurement Distance of 50cm
Measurement Data
Noise Level in dB(A) (Slow), Real-time Frequency Analyzer (RTA)
Graphical Frequency Spectrum of Noise

MORE: Gaming At 3840x2160: Is Your PC Ready For A 4K Display

MORE: How Well Do Workstation Graphics Cards Play Games?

Create a new thread in the UK Article comments forum about this subject
This thread is closed for comments
No comments yet
Comment from the forums
    Your comment