AMD's 23.2.1 Windows 10 & 11 driver is no threat to Nvidia's ray tracing crown anytime soon

AMD's previous graphics driver, version 23.2.1, preceding the one that was released today, was the first display driver of 2023 for its last gen RDNA 2-based Radeon RX 6000 series GPUs. That's because the company had skipped drivers for Navi 2x cards with a couple of updates as it focused on the new RX 7000 series only. Aside from that, AMD also promised big performance improvements with the 23.2.1 driver version. According to the company, users were told to expect up to 27% performance jump in performance compared to the launch driver of Windows 11.

And although AMD did not claim anything specific about ray tracing, some of the users online reported that they were saying massive improvements by up to 40% in ray tracing specific tests like the DirectX Ray Tracing feature test in 3DMark's suite. However, it was also remarked by such users that they weren't really seeing gains anywhere close in actual games; and it is likely they were right as AMD would surely have had brought up the point in its own performance charts.

To test the claims for ourselves, I used my personal Asus TUF RX 6800 XT OC graphics card and ran a few freely available ray tracing benchmarks. These tests aren't exactly feature tests that load up the rendering test scene with just rays, but are meant to simulate somewhat realistic gaming workloads. Aside from two hardware ray tracing tests which are DirectX 12 only, I also used CryEngine's Neon Noir benchmark which is based on DirectX 11 and does not require decicated hardware for ray tracing, instead relying on compute shaders to do the job. All three benchmarks are listed below:

• Bright Memory: Infinite Ray Tracing Benchmark

• Pure RayTracing benchmark by MARvizer

• CryEngine Neon Noir Ray Tracing Benchmark

Our test rig is given below:

• AMD Ryzen 5 PRO 4650G
• Gigabyte A520M DS3H motherboard
• 2x 8GB Kingston HyperX Predator DDR4-3600 CL17 dual channel memory
• Asus TUF AMD Radeon RX 6800 XT OC
• Kingston A2000 PCIe Gen3 NVMe SSD
• Tested drivers: Adrenalin 22.11.2 and Adrenalin 23.2.1
• OS: Windows 10 22H2 (Build 19045.2486)

In terms of the methodology, we chose to run all the benchmarks at two settings, one at 2160p(4K) at the highest settings, to be totally GPU bound, and also 1080p (FHD) at the lowest available settings, to simulate a somewhat CPU dependent scenario. The latter was done to gauge the overhead of the drivers. We ran each test three times and took the median results.

We also did an 8K run on the Bright Memory benchmark though it was hardly a successful test since the GPU core was barely even stressed as the available 16GB VRAM buffer on the 6800 XT was completely saturated and nearly exhausted (indicated by dedicated memory in the screenshot below). This is also why the GPU power draw, indicated by TGP (total graphics power) and PPT (package power total) are only around 93 Watts. For the same reason, the GPU Junction and Hotspot temperatures are also very low (55 and 65 Celsius respectively).

Moving on, we get into the results of the DX12-based hardware ray tracing numbers first. In Bright Memory, the 4K "Very High" performance of the two drivers is near identical, as in both cases, we achieved ~13 fps. The 1080p "High" preset data though was ever so slightly in favor of the newer driver as the 6800 XT managed 56 fps on the 23.2.1 driver, a lead of 1.82% on top of the 55fps achieved on 22.11.2. However, this is clearly within margin of error and the verdict is essentially a tie.

Things took a turn for the worse in the Pure RayTracing benchmark suite as the newer driver proved to be worse. At 4K using the "Pro Settings" preset, the 23.2.1 driver was 3.3% slower than the 22.11.2. And in the 1080p "Casual Settings" as well, there was a 1.7% fall in performance with the newer driver as it went from 62.64 down to 61.57 fps.

Things are not all bad with the new driver, though. Although technically the CryEngine Neon Noir benchmark is not a hardware ray tracing test, it was interesting to see that newer driver actually delivered some measurable gains here.

In this test, in both instances, ie, at 4K "Ultra" and 1080p "Very High", the 6800 XT was faster on the newer 23.21 driver. At 4K ultra, the difference was 2.04% in favor of the newer driver (4,650 vs 4,706), while in the 1080p Very High test, the 6800 XT was 2.1% ahead of the 22.11.2 (14,645 vs 14,344).

Ultimately, the numbers seen here are pretty uneventful and boring if you are an AMD fan and would have liked to see big gains in ray tracing, perhaps to better compete with Nvidia's RTX 3000 series cards. However, it is hit or miss depending on the different intensity of the workloads.

For example, in more GPU heavy situations that involve higher ray tracing effects like on the Pure RayTracing benchmark Pro settings, there is a performance loss. Meanwhile, typically, in the less GPU intense tests, there is a significant though imperciptible raise, perhaps suggesting that driver overhead has been slightly lowered with the newer 23.2.1 driver.

AMD's 23.2.1 Windows 10 & 11 driver is no threat to Nvidia's ray tracing crown anytime soon