How We Test Graphics Cards

Testing on an open bench requires careful planning, else it leads to misguided results. Most gamers have closed cases, affecting thermal, acoustic, and even performance results compared to exposed hardware. We're working to represent that more accurately.

It is impossible to deny the truth our readers have provided as feedback on multiple occasions. And we have long been aware of the shortcomings tied to testing graphics cards in open-air test beds, where the results we generate almost always differ from those collected in a closed case. Is there a solution that makes everyone happy?

Over the past six months, I developed a concept for our laboratory that allows us to combine the pleasant advantages of an open test setup (quick and easy access to hardware) with the changes our readers requested. The devil is in the details though, as a number of additional components are needed to accommodate our infrared and power consumption testing methodologies.

The next step was to get in touch with manufacturers at Computex 2016, where I found a partner in Lian Li. The company truly listened to what reviewers actually need.

The original PC-T70, shown above, is a classic open-air test bench afflicted by the same issues we were trying to avoid. Therefore, I drew up a modification that'd match our requirements pretty well.

Assembly and Reconstruction of Lian Li's PC-T70

In order to make this a closed case, there will be an optional set of side plates, a back plate designed to closely match standard PC cases (including an I/O shield), and an acrylic cover.

Each of the side plates is made of solid, black-colored steel and can carry up to two 120mm or 140mm fans. They're also equipped with a dust filter that can be attached using a magnetic lock. Although Lian Li typically favors aluminum construction, we specifically asked for steel because it's less expensive and more utilitarian.

To fully enclose the internals we also needed a cover. And since we wanted to look inside, the solution was naturally acrylic. This cover is perfectly suited to pop on and off via magnetic strips. I consciously decided against a hinge because it'd take up too much space with the case open.

From here on out, you won't be reading about components that we expect will be available to everyone. A lot still has to happen before we're able to fully utilize our equipment, necessitating a bit more customization.

For starters, there's our high-res Optris PI640 that can't easily take measurements inside of a closed case. Since we don't need the whole armada of fans, we have two additional side plates that can be attached instead to cover the side openings. Both parts are positioned on top of each other with a layer of special foil in between, which is permeable for infrared waves and whose transmittance we know.

This kills two birds with one stone because the IR radiation can escape, while warm air stays inside. After all, that was the whole point of our airtight exercise.

Since the new readings needed to be kept compatible with previous measurements, this setup employs similar hardware as our current system. Unfortunately, our old motherboard was damaged by electrostatic discharge, giving us a good reason to switch from a non-conductive test bench made from acrylic to a grounded case. And so MSI's X99S XPower AC is replaced by an MSI X99S XPower Gaming Titanium. The RAM also had to be switched out for four Corsair Vengeance DDR4-3200 modules. They operate at the same 2400 MT/s as the old ones, though. So, as far as performance is concerned, nothing changes.

We changed RAM modules because Crucial's Ballistix kit sports reflective heat spreaders. In some tests, they distort (if only marginally) the IR results due to their reflective surface.

Water Cooling

Air cooling isn't desirable for a number of reasons. First, a large tower-style heat sink would block too much of the IR camera's view. Second, in a very small case like this, the cooler would dissipate so much waste heat that mainstream graphics cards with lower TDPs would be thermally overpowered, affecting the measurement results. Water cooling helps us work around both challenges.

Our solution employs a classic open loop composed of individual components. The pump (an Alphacool Eispumpe VPP755) is decoupled and mounted in the lower part of the system, where it rests on foam. To the pump, we attach an Alphacool NexXxoS UT60 Full Copper 360mm radiator measuring 6cm wide and loaded up with three PWM-controlled be quiet! Silent Wings 2 fans in a push configuration. 

The reservoir is positioned on the side, where it's easily accessible. Through a tee fitting on the top, we're able to refill liquid at any time through a screw-on funnel.

Here's where the quick-release magnetic strips come into play, making it easy to get inside and install a water block on the graphics card, for example. You can even see in the images below where Lian Li drilled tubing openings for this exact purpose.

Fans, Sensors, and Other Components

Separately, I put together a mid-tower system inside of be quiet!'s Silent Base 800 using a fairly similar platform, logging the temperature over time of Asus' hot Radeon R9 290X StriX. The air, GPU, and motherboard temperature readings I collected served as a baseline to help configure the new machine's fan speeds.

The airflow of a standard PC case is simulated by a fan in the back for exhaust, plus a controllable fan on the opposite side of the case pulling in fresh air. That back fan spins at a fixed 700 RPM, while other one is dialed in to 800 RPM. The fan on the side can also be removed, its vacancy  filled with a cover. 

At the front right, you can see a small, aluminum heat sink. It is placed at the warmest spot inside of the closed case, and it houses a temperature sensor. More on that later. I also attached two small USB speakers on the floor behind the front, since nothing goes better with benchmarking than some music.

A modified 850W be quiet! Dark Power Pro 11 is hidden under the case's floor. Its modular cables had to be altered somewhat so they'd work with our oscilloscope's clamps, giving us the most stable voltage readings possible. The chassis itself serves as a path to ground, to which we made several connections.

Before we move on to our methodologies, let's review the speeds and feeds:

Test System
Hardware:Intel Core i7-5930K @ 4.2 GHz
MSI X99S XPower Gaming Titanium
Corsair Vengeance DDR4-3200 @ 2400 MT/s
1x Crucial MX200, 500GB SSD (system)
1x Corsair Force LS, 960GB SSD (applications, data)
be quiet! Dark Power Pro 11, 850W power supply
Windows 10 Pro (with all updates)
Water CoolingAlphacool Eispumpe VPP755
Alphacool NexXxoS UT60 Full Copper 360mm
Alphacool Cape Corp. Coolplex Pro 10 LT
Case:Lian Li PC-T70 with extension kit and modifications

MORE: Best Graphics Cards

MORE: Desktop GPU Performance Hierarchy Table

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