3DMark Variable Rate Shading Test Shows Big Performance Gains On NVIDIA, Intel GPUs, AMD Won't Run

UL Benchmarks (formerly known as Futuremark) has added a new Variable Rate Shading (VRS) test to its popular 3DMark benchmarking suite, and it is now available to the public. The new test taps into a feature of Microsoft's DirectX 12 API that allows developers to improve performance by selectively reducing the level of detail in parts of a frame.

The idea behind VRS is to lessen the workload on a GPU and push higher framerates, without adversely affecting the overall image quality of a scene as it is perceived by the viewer. In theory, the selective reduction of detail in any given frame will occur where it will not be noticed much, if at all.

"By using VRS to lower the shading rate for parts of the frame that are in deep shadow, far from the camera, or peripheral to the player's focus, for example, a game can run at a higher frame rate with little perceptible loss in visual quality," UL Benchmarks explains.

The new VRS feature in 3DMark consists of a series of tests designed to highlight specific techniques and allow users to compare differences in performance and image quality.  There is a caveat, however—it requires a graphics card and drivers that support Tier 1 VRS as Microsoft has defined it, which apparently rules out AMD's GPUs (we could not get it to work on a Radeon RX 5700 XT). Based on the information we currently have available, AMD's GPUs -- up to and including Navi -- do not have the hardware to support Tier 1 VRS. That said, there are other methods that could be used that could potentially achieve similar results, and this patent filing proves AMD has some form of the technology in their portfolio. However, as of today, VRS is a no-go on an AMD Radeon RX card of any kind.

Here's an overview of the test with comparisons showing VRS turned on and off...

It's difficult to discern a difference in image quality without looking extra closely at a scene or taking screenshots. This is especially true when moving around, versus a static screen. Even though the reduction in image quality is relatively minor, is VRS truly worth it? We explored this by running the VRS benchmark on multiple setups.

You may recall from our performance preview of Intel's 10-nanometer Ice Lake architecture that we ran the VRS benchmark on a Core i7-1065G7 configured at 15W and 25W. This is one of 11 Ice Lake CPUs recently announced, and the second most powerful of the bunch with 4 cores and 8 threads ticking along at 1.3GHz to 3.9GHz. It also features 8MB of SmartCache and Iris Plus graphics with 64 execution units (EUs).

Using a non-public, pre-release driver, at 25W we saw a nearly 41 percent jump in performance with VRS turned on, as it pertains to 3DMark's latest test. Similarly, at 15W the performance gain was just a hair over 41 percent. Those are not insignificant gains, folks.

Ryzen 9 3900X + GeForce RTX 2080 Ti

It's not just integrated graphics that stand to benefit from VRS, though. We also ran the test on a handful of discrete GPUs. Shown above is our benchmark run with a GeForce RTX 2080 Ti installed into a testbed alongside AMD's Ryzen 9 3900X CPU. As you can see, flipping the VRS toggle netted us a greater than 46 percent performance gain.

To rule out this being a possible outlier, we ran the same test on two other graphics cards, those being the GeForce RTX 2080 and GeForce RTX 2060 Super. Here are the numbers from each one (with the GeForce RTX 2080 Ti plotted as well)...

And no, it was not an outlier. Our GeForce RTX 2080 test run netted the biggest performance gain of all at 52.9 percent, while the GeForce RTX 2060 Super was not all that far behind at 49.5 percent.

This is interesting, and bodes well for Microsoft's VRS method as a whole. That's assuming the testing in 3DMark is an accurate reflection of the tech. The test runs in two passes, once without VRS to create a baseline of performance, and then a second time with VRS turned on. 3DMark then reports the average frame rate for each pass, and even does the courtesy of performing the math for the performance gain, as a percentage.

"With Variable Rate Shading, a single pixel shader operation can be applied to a block of pixels, for example shading a 4x4 block of pixels with one operation rather than 16 separate operations. In the 3DMark VRS feature test, the shading rate varies with camera distance when VRS is enabled: 4×4 for the furthest objects, (the blue areas in the image above), 2×2 for the middle-distance geometry (green areas), and 1×1 (red areas) for the closest foreground objects," UL Benchmarks explains.

There is also an interactive mode whereby users can change the VRS settings on the fly to see both how they affect performance (framerates) and image quality. For more details, check out the accompanying technical guide (PDF).

Regardless, as you can see, the performance gains are significant, so how this will play out for AMD in the future is anyone's guess. It's clear currently that both NVIDIA and Intel support the Microsoft spec in hardware and software, but AMD simply does not and won't in the current version of its Navi architecture.

The new VRS test is available as a free update to 3DMark Advanced Edition, which itself is currently on sale for $7.49 (down 75 percent from its regular $29.99 price). It's also available as a free update to the Professional Edition, though it's not in the freebie Basic version.