New drivers even for older Radeons: Performance increase test on RX 6950 XT

Years ago, it used to be common for cards to have optimizations throughout their lifecycle that would boost performance across a larger number of games. At the moment, both AMD and Nvidia offer drivers optimized for the currently released A titles, and performance increases for older games occur only exceptionally. When it comes down to it, they appear more often with Radeons than with GeForces. Maybe I’m being overly paranoid, but in recent years it seems to me that AMD has been timing improvements for older games for some major event where they can sell it to the public – it used to be a big driver update that usually came out sometime before the end of the year and brought new functions, nowadays this happens more often with the launch of new products.

The reason for saving performance optimizations for a similar occasion would be: when new cards are launched, reviewers often get the new drivers only a few days before the end of the embargo for testing, so there is not much time for a more extensive methodology to put in addition to the featured products old cards can also be resized with new drivers. Then when you put the results of the new cards on better optimized drivers next to the older results of the previous generation of cards, the new ones show a higher performance increase against the older ones and somehow look better next to them.
The last time something similar happened was the launch of fifties Radeons from the RX 6000 series. Because of this, I assumed that with the RX 7900 there would be some significant update of the AMD drivers with an increase in performance in older models. AMD took me over, not only did the drivers for the RX 7900 not bring performance improvements to the 6000s, they didn’t support them at all. At the launch of the RX 7900, I was probably not the only one who was surprised that it was not possible to compare older Radeons on the same driver branch as the RX 7900.

When it is no longer possible to measure older cards on the same drivers for the introduction of new products, it is usually possible at least after that on the public version of drivers for new cards. But this time it was extreme, new drivers with RX 6000 support didn’t appear even at the launch of RX 7900, not even a few weeks and later even months after it. We waited for them for almost a quarter of a year.

The latest drivers for the RX 6000 were released by AMD sometime last year at the end of November (AMD claims on the support website that they were released on December 8, but they still have the November marking 22.11.2 and they were available for download from the first of December directly on the AMD website). In December, with the launch of the RX 7900, the first drivers for the RX 7900 followed, followed by some hotfixes, but always with support exclusively for the Radeon RX 7900. AMD explained this by saying that before the launch it concentrated on tuning the drivers for the RX 7900, and the RX 6000 had to go aside But another month or two passed, and it wasn’t until mid-February that we finally got a new version of drivers with RX 6000 support. And you guessed it right – in addition to optimizations for the newly released Forspoken games, the Dead Space remake and new extensions for the Vulkan API, AMD is also proud by the increase in performance of Radeon RX 6000. Interesting enough for AMD to devote even blogpost on their community pages.

In it, he mentions performance gains over the drivers released for Windows 11 in November 2021 (a year after the launch of the Radeon RX 6800) as well as against the previous version from November 2022, and some of the performance gains look impressive. The charts show results from Cyberpunk 2077, F1 22, God of War, Marvell’s Guardians of the Galaxy, Hitman 3, Sniper Elite 5 and Wow: Shadowlands with performance increases ranging from 6 to 18%.

At first glance, it looks great, and it fits the hype – namely that “Radeons age like wine”. AMD’s marketing also plays a part in this, and it likes to feed this popular saying for a long time. The fact that at the beginning the performance is not much, but each new driver adds an average of two to three percent of performance, and when it accumulates over time, it is already a significant acceleration, I first heard from AMD marketers during the presentation of the Radeon HD 3870 sometimes Summers of the Lord 2007.

They successfully feed the idea that if you buy a Radeon, it will have worse performance at the beginning, but here and there drivers will appear that will increase the performance of the cards by a few percent, and if you put the percentage to the percentage, after two years you will be back by ten fifteen percent more powerful card than when purchased.

As the blog post on the website also says, „Over the course of just one year, we have seen increases in performance across numerous games with some achieving over 10% gains.“ were not in a single year, performance has increased in many games, with some exceeding ten percent. Great, isn’t it?

Well, how about it. It’s been some Friday since I noticed that AMD often puts relatively new titles in these comparisons. Around the time when the frequency of general performance updates dropped. And this is also the case with the mentioned blog post. If I sort out what was released and when, it looks something like this:

• WoW: Shadowlands – November 23, 2020
• Cyberpunk 2077 – December 10, 2020
• Hitman 3 – January 20, 2021
• Adrenaline 21.10.1 – 1 October 2021
• Marvel’s Guardians of the Galaxy – October 26, 2021
• God of War – 14 January 2022
• Sniper Elite 5 – May 26, 2022
• F1 June 22-28, 2022
• Adrenaline 22.11.2 – 1 December 2022
• Adrenaline 23.2.1 – February 15, 2023

So for four of the seven games in the first graph, AMD marketing compares the performance of non-optimized drivers against drivers that are already “game ready”. The graph makes it look like if you were playing God of War on 21.10.1 controllers, it was already running “up to 10% faster” this January. But it already ran “up to 10% faster” at launch, but the performance is compared with controllers released a few months before God of War itself. That new games run more stably and faster with drivers tuned for their launch than when you try them on six-month-old drivers is a fairly common phenomenon for GeForce as well. After all, this is why “Game Ready” controllers are created in the first place. Pretending that card performance has increased by ten percent in a year is like believing in retailers that the graphics you’re looking at in the store at launch are already eight percent off.

In other graphs, however, you can already see the performance increases between the November and the current version. It’s already clear from them that there have been some, albeit mostly minor, performance gains over and above what the launch drivers bring.

And the last one is the increase in performance in the TOP 5 tested games, there it is probably a clean selection of the best of the most optimistic results that could be measured. In the notes, you will also learn that the different settings and resolutions were different for different cards, so it’s more like comparing apples to pears, but it’s quite understandable; the performance increase of the RX 6400 cannot be tested well enough in 4K with details on ultra.

It is worth paying extra attention to which games the performance has increased. Both AMD and Nvidia are primarily focused on optimizing “overcooked” titles. They argue, for example, that there are too many games coming out and they can’t optimize for everyone, so they focus on optimizing at least the most popular ones. This is quite logical and makes sense for controllers being prepared for currently released blockbuster games. But if you look at the list of titles listed in the latest “release notes” for Radeons, there aren’t that many of the current top-notch games, and you see more of the old ones:

• Up to 4% increase in performance for Marvel’s Spider-Man Remastered @ 4k, using AMD Software: Adrenalin Edition™ 23.2.1 on the Radeon™️ RX 6950XT GPU, versus the previous software driver version 22.11.2 RS-518
• Up to 3% increase in performance for Sniper Elite 5 @ 4k, using AMD Software: Adrenalin Edition™ 23.2.1 on the Radeon™️ RX 6950XT GPU, versus the previous software driver version 22.11.2 RS-519
• Up to 6% increase in performance for Shadow of the Tomb Raider @ 4k, using AMD Software: Adrenalin Edition™ 23.2.1 on the Radeon™️ RX 6950XT GPU, versus the previous software driver version 22.11.2 RS-520
• Up to 7% increase in performance for Quake II RTX @ 4k, using AMD Software: Adrenalin Edition™ 23.2.1 on the Radeon™️ RX 6950XT GPU, versus the previous software driver version 22.11.2 RS-521
• Up to 4% increase in performance for Hitman 3 @ 4k, using AMD Software: Adrenalin Edition™ 23.2.1 on the Radeon™️ RX 6950XT GPU, versus the previous software driver version 22.11.2 RS-522
• Up to 6% increase in performance for Marvel’s Guardians of the Galaxy @ 4k, using AMD Software: Adrenalin Edition™ 23.2.1 on the Radeon™️ RX 6950XT GPU, versus the previous software driver version 22.11.2 RS-523
• Up to 19% increase in performance for F1 2022 @ 4k, using AMD Software: Adrenalin Edition™ 23.2.1 on the Radeon™️ RX 6950XT GPU, versus the previous software driver version 22.11.2 RS-525
• Up to 9% increase in performance for DOOM Eternal @ 4k, using AMD Software: Adrenalin Edition™ 23.2.1 on the Radeon™️ RX 6950XT GPU, versus the previous software driver version 22.11.2 RS-526
• Up to 4% increase in performance for Borderlands 3 @ 4k, using AMD Software: Adrenalin Edition™ 23.2.1 on the Radeon™️ RX 6950XT GPU, versus the previous software driver version 22.11.2 RS-527
• Up to 4% increase in performance for Hogwarts Legacy @ 4K using AMD Software: Adrenalin Edition™ 23.2.1 on the Radeon™️ RX 6950 XT and Radeon™ 7900 XTX, versus the previous software driver version (22.11.2 for 6950 XT, 23.1.2 for 7900 XTX) RS-530

In the list you will also find things like Shadows of the Tomb Raider (2018), Quake II RTX (2019), Doom Eternal (2020) or Borderlands 3 (2019), for which today it can hardly be said that performance optimization will be appreciated by hundreds of thousands of players, who are just throwing themselves into these games. Most of them, on the other hand, have already finished it long ago. But you will still find the mentioned titles abundantly represented in graphics card testing methodologies.

And so you have hundreds of new games, in which the performance did not have to increase even a finch in two years, and a few titles that are not played as much anymore, but they are tested a lot. And when the performance of a quarter of the games increases by six to ten percent in most tests, the average performance goes up by a few percent, and suddenly you have “on average four percent more powerful” graphics at home. Quite easy and effective. I’m not surprised that manufacturers do this, but that’s why you should take such claims about how the performance of cards increases over time with a grain of salt.

Tuning the controllers for the games that are used in the tests, so that the averages come out better, has been bothering me for quite some time. However, it is practiced by both AMD and Nvidia, so in the methodology I try to stick to games that are not tested by anyone, and on the contrary, I do not get too involved in things that are tested by everyone. If you’ve ever wondered why I have any of the games in the methodology when so many people don’t, this is the main reason.