AMD Adrenalin 22.7.1 Driver Performance Test: Big Gains in DX11 & OpenGL
PC hardware companies work in funny ways sometimes. When AMD had a new driver ready that added extra features with arguably niche appeal, like its AMD Link in-home streaming tech, it was very excited to tell us about it, and have us tell you about it. However, when it released a new driver recently with real performance gains, the company didn't even publish its own blog post about it, let alone shout it from a mountain top.
That said, regardless, AMD's latest Radeon Adrenalin Software 22.7.1 release is indeed rather exciting for PC gamers, and worthy of shouting about.
Of course big tech companies make proclamations of "new and improved" every day, but it is not always just marketing hype. Back in May, AMD debuted what it claimed was a "rebuilt from the ground up" DirectX 11 driver, and now, with this latest release, the company has revised its OpenGL driver, too. Let's look at some of their claims...
All of the tests were done on a test bench with a Ryzen 7 5800X3D CPU, 32GB of low-latency DDR4-3600 memory, and AMD's own Radeon RX 6800 XT card. All driver settings were left at their defaults and we did not adjust any Windows settings, either. That means the Balanced power profile, Game Mode was left on, and AMD SmartAccess Memory was enabled.
So, without further ado, let's dive into the benchmarks, shall we?
3DMark Fire Strike Benchmark
Okay, this isn't a game. In fact, we expected no change in this benchmark because AMD has surely optimized it in the extreme, right? We tested both the original Fire Strike benchmark, which runs in 1920×1080, as well as the more punishing Fire Strike Ultra benchmark that runs in 3840×2160, four times the resolution.
The gains are small, to be sure—but they're real and consistent. Like with all of the rest of our benchmarks, we ran each test three times, and then took the middle of the three results. This prevents our results from being outliers, and ensures that they should be at least reasonably representative.
The 3DMark results hint at an interesting pattern. The new driver does improve GPU performance ever so slightly in this benchmark, but it also raises the normally-very-consistent Physics score. This score is basically Fire Strike's "CPU" score, and the new driver improving that seems to mean that the Radeon card is better able to make use of our Ryzen 7 5800X3D's horsepower.
Final Fantasy XIV Endwalker Benchmark
Now, FFXIV is an older game, and so our beastly Radeon card has no trouble whatsoever pushing it to phenomenal frame rates running in full 4K UHD with maxed-out settings. We do see a small and consistent gain in performance with the new driver, but it's very small indeed. We didn't check frame times in this benchmark, but we'd not be surprised to find that it's a smoothness gain.
However, the FFXIV benchmarks are actually a succession of smaller scenes played in sequence, and the benchmark also records how much time you spend staring at loading screens across the course of the tests. This is arguably the more interesting result from this testing. Now, a reduction of 3.5 seconds may not seem like that much, but remember that our test machine is ludicrously fast, and loading this content off of a PCIe 4.0 SSD. It's likely that a slower machine would see an even bigger benefit here.
Phantasy Star Online 2: New Genesis Benchmark
Don't read too much into the 1920×1080 results. The game has a hard framerate cap at 180 FPS, and scores over 30,000 mean that you're staying right near that cap for most of the benchmark. This test isn't completely deterministic either due to some randomness in the particle effects, and while it seems like a huge difference, these scores really aren't meaningful.
The more interesting gain is the ~5% uplift in 4K performance. Neither of these scores are particularly bad, but the new driver seems to have resolved some stuttering in the benchmark, and that appears to be what is improving performance here.
Counter-Strike: Global Offensive Benchmark
CS:GO runs quite well on integrated graphics, so why bother testing it? Well, it is one of the most popular games on Steam, and dedicated players want the absolute highest framerates possible. The test map that we used, known as "FPS Benchmark" and created by Mr. uLLeticaL, is not as demanding as some of the real scenarios in the game these days and so it produces hilariously high frame rates. The important part is that they are consistent and representative, which lets us see the difference between the drivers.
That is a 4.25% improvement, which is about what we've been seeing in DirectX 11 games. It may not be much, but an improvement is still an improvement, so we are pleased to see it.
ARMA III Benchmark
This is our biggest DirectX 11 gain yet, and the biggest one overall. ARMA III is a unique game in that it is extremely reliant on both single-threaded CPU performance and main memory bandwidth. While the engine it uses is multi-threaded to a degree, it really wants a super-fast CPU and system memory bus.
Of course, none of that changed between our tests; the only variable that moved is the graphics driver. All the in-game settings are the same; we checked! This result really points to some CPU usage optimizations in AMD's new DirectX 11 driver, just as we suspected with 3DMark and Final Fantasy XIV. 65.7 FPS in this benchmark is quite a high result; ARMA players, update your drivers immediately.
Even still, it runs better now. We set up a custom event using the Bloomfield Speedway and made sure to match weather conditions. There was still a surprising amount of non-determinism in this benchmark and so these results were a little inconsistent, but our strategy of running three times and picking the middle gave us a pretty solid improvement of just under 8% with Adrenalin 22.7.1.
Unigine Superposition Benchmark
Unigine's Superposition benchmark is a demanding test that uses millions upon millions of triangles to display a gorgeous laboratory scene. It is not ray-traced, or anything; it is strictly last-gen technology. Still, it looks absolutely fantastic, and even in 2022 it can put a hurt on graphics cards.
Superposition serves as a bit of a bridge for us between our DX11 tests and OpenGL titles. The benchmark supports both APIs, and naturally, the conventional wisdom on Radeon cards has been to run it in DirectX 11 mode owing to that mode's better performance on those cards.
Well, as you can see for yourself, that's not necessarily the case anymore. Okay, the DirectX 11 mode still performs a little better. The difference is vanishingly small, though, and it is not far from being run-to-run variance. The OpenGL mode of the benchmark performs excellently on our RX 6800 XT now, and the DirectX 11 mode even sees a very small uplift with the new driver. Excellent work, AMD.
Testing Xonotic is a lot like testing the old Quake games was back in the day: you run a timedemo. This means that you play back a demo recording at the maximum speed possible for the machine and then compare that performance against the original speed. At the end of the demo, Xonotic spits out the minimum, average, and maximum one-second frame rates.
We tested Xonotic using the built-in "big key benchmark" demo, and the performance gains are plain to see. It would be easy to argue that 271 average FPS is plenty fast, even for a game like this, but you could really feel that 39 FPS low when playing, as performance would hitch and stutter at times. The new driver feels drastically smoother and makes this classic-style FPS a joy to play.
Before the new driver, RAGE was a stuttery, awful mess on Radeon cards. Testing using Adrenalin 22.3.1, the game hitched and stuttered non-stop. Despite the framerate counter reading values as high as 95 FPS, it felt more like 45. Let's see what happens when AMD throws some optimizations at it.
Well, would you look at that. RAGE actually has a 60 FPS cap in place normally, and testing in 3840×2160, despite the stuttering, was still giving us average FPS values around 58. That simply wouldn't do, so we cranked the resolution all the way up to 7680×4320 and used a mod to uncap the framerate.
That gave us more useful values for testing, and the play experience bears out these results. With the new driver, RAGE is smooth as silk, and an absolute blast to play at 8K120. It loads in noticeably faster, too.
DOOM (2016) Benchmark
If you have read this far, you already know the answer by this point: no. Doom '16 runs very well in OpenGL on the Radeon RX 6800 XT with the 22.7.1 driver. This is such a drastic improvement over the previous driver's performance that we are going to show you a smoothness graph to demonstrate the difference.
The green line represents the frametime with the 22.3.1 driver, and the orange line represents the frametime with the 22.7.1 driver. That huge spike at the beginning with the 22.7.1 driver is an artifact of the tool that we use to capture frametimes, CapFrameX; you can ignore that. Overall, the new driver is both faster and drastically smoother than the old driver.
ZDoom Eviternity Benchmark
As with Doom '16, we tested ZDoom in both OpenGL and Vulkan. Surprisingly, despite AMD saying nothing about Vulkan gains in this driver update, the new driver runs considerably better in both OpenGL and Vulkan in this particular benchmark. We're ready to chalk that up to peculiarities of ZDoom, as its Vulkan renderer is probably closely related to its OpenGL renderer.
Cyberpunk 2077 Benchmark
Using the built-in benchmark, we actually see a small performance improvement in Cyberpunk 2077 between the old and new drivers. This benchmark is not deterministic, but it is reasonably reliable from run-to-run and this result was fairly consistent. Clearly AMD's been trying to shore up its slightly sub-par ray-tracing performance. With these settings, Cyberpunk 2077 is quite playable and still looks pretty nice.
OpenGL Emulator TestingOne of the main uses of the OpenGL API for gamers on Windows is to play video games in console hardware emulators. Most emulator applications are open-source software, and that means that the developers often prefer to use cross-platform APIs. As a result, you do see a preference for OpenGL among emulator authors.
Recently, that preference seems to have shifted toward Vulkan. The newer and lower-level API has all of the cross-platform advantages of OpenGL, yet none of the legacy cruft of that aged API. Vulkan works well on GPUs from every vendor, albeit with some inconsistencies in feature and extension support.
Still, there are a lot of emulators out there that perform or function best with OpenGL. One of the most notable is CEMU, the Wii U emulator. CEMU famously got Vulkan support at the very end of 2019, but certain mods and "graphics packs" for the emulator still require OpenGL. That essentially meant that Radeon players did not have access to those because OpenGL performance on those cards was so poor.
Testing today, we scaled The Legend of Zelda: Breath of the Wild all the way to 8K resolution (that's 7680×4320) with an uncapped framerate, and it ran at a fairly-consistent framerate in the high-30s outdoors. That's stunning performance compared to earlier this year, when this same card struggled to run Super Smash Bros at a playable framerate regardless of settings. You don't even need to apply the AMD-specific graphic packs to fix up broken features in this game, and you can obviously improve performance by lowering the resolution.
We also tested other emulators, including PCSX2 for the PlayStation 2, Yuzu emulator for the Nintendo Switch, and Mupen64Plus on Retroarch. All of them worked flawlessly with the new driver. This is great news for emulation fans using Radeon cards as well as integrated Radeon graphics.
Final ConclusionsRegular readers will recognize that our test set this time around was pretty strange. We did not run our usual batch of games because we generally focus on testing newer and more up-to-date titles, and those games do not typically employ DirectX 11 or OpenGL as their API.
However, every single application we tested saw an improvement, even if it was very minor. The biggest improvements were in OpenGL titles, and primarily in terms of minimum framerate and frame time consistency. All of these areas have traditionally been weak spots in certain places for Radeon cards.
Indeed, a major frustration of Radeon cards historically has been that, while recent and popular games will generally work well, you could have some struggles with older, niche, or less popular titles. Obviously, we have not had time to test that many games yet, but so far, we are very impressed with the work that AMD has done here, particularly for the OpenGL API. This driver sees huge gains in performance and smoothness for OpenGL, and it does some fine work for DirectX 11, too.
Kudos to AMD for stepping up and smoothing out games on these older APIs. We hope the company can continue making this kind of progress.