
Sidney discusses more about performance in 12.4.4, and upcoming improvements in 12.4.5
X-Plane 12.4.4 is finally out of beta, and we’ve seen some complaints about crashes and performance, so I’d like to address what has happened and what we are going to do going forward.
First things first, we don’t want to ship an update that wrecks performance and/or causes crashes. The engine team has worked quite hard to squash bugs and improve performance, and we aren’t even close to the full totality of performance improvements coming in the future.
The second thing I want to highlight real quick is, if you have an AMD RDNA 4 card (that’s the RX90xx series of cards) and see much worse performance with 12.4.4, please downgrade to AMD Adrenalin 26.9.1. If you are on Nvidia and are using Nvidia’s 6xx series of drivers and are seeing an increase in GPU crashes, please downgrade to driver 59x.xx. I’ll edit this post with updated driver versions and/or advice when it becomes available.
What’s going on with AMD
X-Plane 12.4.4 ships with a new version of Zink, an open-source OpenGL driver that I wrote about 3 years ago here. Zink accepts OpenGL calls, the thing that third parties use to render instruments and user interfaces, and turns them into Vulkan calls for the GPU. On AMD this has significantly improved performance over using native OpenGL/Vulkan interop through the AMD drivers. However, the version we shipped wasn’t perfect and in particular, had issues with emulating the line rendering modes that OpenGL provided. This caused some issues with some plugins that relied on it, in particular the FlightFactor 777v2 which didn’t draw any flight paths on the ND with Zink enabled. For 12.4.4 we have finally shipped a new version of Zink (and, by the way, also open sourced our changes to it) that has significantly improved line rendering support and makes the 777v2 usable by AMD users.
Unfortunately, we have now learned that there are issues with the new version. Two issues in particular: on one hand, we have users with Polaris GPUs (that’s the RX5xx series of cards, so RX590. Not the RX50xx generation), which are no longer able to enable Zink at all. This new version of Zink requires the VK_KHR_maintenance5 extension which these older GPUs don’t support. If you have a Polaris GPU, not running with Zink is not an option because it significantly degrades performance. We are working on removing this requirement from Zink and getting you back on board.
On the other hand we have also learned that RDNA 4 cards running AMD Adrenalin 26.10.1 or later see massive performance degradation with Zink enabled. This is new with X-Plane’s 12.4.4 version of Zink and this new driver. If you downgrade to 26.9.1 or earlier, the performance will come back. From user data I can tell you that with this configuration (RDNA 4, AMD Adrenalin 26.10.1+ and 12.4.4) the GPUs will run at a significantly reduced power profile and also much hotter while also not kicking in the fans as much as before. I don’t believe this is cause for concern for the longevity of the GPU itself, but it explains why GPU times have become much worse because the GPU doesn’t run at its full performance power profile.
TLDR
- Older AMD RX 500-series (Polaris) GPUs: Zink can no longer be enabled due to an unsupported Vulkan extension.
- AMD RDNA 4 GPUs: Users running Adrenalin 26.10.1 or newer may experience significant performance degradation with Zink enabled. Rolling back to driver 26.9.1 or earlier restores performance.
So, how did this make it into release?
Despite our testing efforts, some issues only become apparent when an update reaches the wider range of hardware configurations used by the X-Plane community. Unfortunately, this was one of those cases.
We’ve since identified the problem and are taking steps to expand our testing coverage to help prevent similar issues in future releases.
This is why beta periods are important and why even very obvious issues should always be reported. If you run an X-Plane beta and run into an issue but don’t report it, there is a good chance we will not know about it. Please report any and all beta bugs and performance regressions to us.
For performance-related issues, please include an ETW trace with your bug report whenever possible, as this provides valuable diagnostic data that helps us identify bottlenecks and pinpoint the underlying cause.
Sometimes having a lot of reports is really helpful; for example, it took quite a bit for us to make the connection between decreased performance reports and RDNA 4, and it was only after having enough data from user bug reports that we saw the common thread.
Regardless though, this is a stupid fumble on our part and improving our hardware test coverage is definitely an important step as both Laminar Research and X-Plane is growing.
TLDR
- Some issues only become apparent across the wider range of hardware used by our community, which is why beta testing and bug reports are so important. Please report any issues you encounter, including ETW traces for performance-related problems whenever possible. Your reports help us identify patterns, diagnose problems and improve X-Plane. We’re also expanding our own testing coverage to help catch more of these issues before release.
Explaining Nvidia Issues
One other thing that’s going on is that Nvidia users are seeing an increase in GPU crashes. This isn’t actually related to 12.4.4, but 12.4.4 can make this worse than 12.4.3 was. The underlying problem is that there is an issue with how the Nvidia 6xx driver series handles VRAM pressure when doing OpenGL/Vulkan interop. In those cases, the driver will automatically move memory from the GPU to system memory and somehow, somewhere along the line, explode in the copy engine. The copy engine is what Nvidia calls their DMA queue, the piece of tech that is responsible for moving memory between RAM <-> VRAM without stalling in-flight rendering and compute operations.
We’ve seen the increased GPU crash rate for quite a while now, and I reached out to Nvidia at the end of September, together with the issue of broken SSAO rendering at the time. SSAO has been fixed by recent Nvidia drivers, but the GPU crashes remained. The confusing part about this is that all of our forensics for device losses are very clear in that X-Plane wasn’t running any commands on the GPU at the time of the crashes. So I also spent a lot of time looking at our own diagnostics and trying to figure out why they were broken, being under the belief that this was potentially an issue on our end.
Nvidia is actively working on this, and I was able to build a repro case for them that can reliably generate a GPU crash in about 10 minutes. Their driver team is actively looking at this, and I’m very hopeful that this will be fixed with a driver update soon. In the meantime, downgrading to Nvidia 59x or earlier drivers will also solve this. Mind you, though, this is one specific class of GPU crashes, not all GPU crashes; however, this is currently making up about 75% of all GPU crashes, so it’s the number one crash reason by a big margin.
We have also heard reports from Nvidia users that the new version of Zink reduces performance. We are actively investigating this and working on it; by the looks of it this is down to some configuration changes that we made to Zink. I have nothing concrete here yet; I just wanted to mention that we are aware of this and are working on it.
TLDR
- NVIDIA users may experience increased GPU crashes due to an issue with the 6xx driver series, particularly under high VRAM usage. NVIDIA is actively investigating the problem, which currently accounts for approximately 75% of reported GPU crashes. In the meantime, rolling back to 59x or earlier drivers should resolve this specific issue.
- We’re also investigating reports of reduced Zink performance on NVIDIA GPUs following the 12.4.4 update.
What’s going on with VRAM
One reason I was excited for 12.4.4 was that it ships with sparse memory. Sparse memory is a technique that allows us to not commit certain resources to VRAM and instead leave them unbound. We use this, for example, for emissive LIT textures, which are mostly a single colour except for the few emissive pieces of the texture. The hardware will fill in the missing blanks with a fallback colour, and we get to save VRAM. Additionally, there was a lot of work on breaking up large meshes like UHD terrain up into smaller chunks and streaming those individually to VRAM so that we don’t have to make the whole 200mb+ worth of vertices resident on the GPU just because you happened to look at a single triangle. Long story short, the goal with 12.4.4 was to reduce VRAM usage. Yes, we are aware that memory prices are high and the engine team is working not only on optimizing performance but also VRAM usage.
While X-Plane 12.4.4 introduces improvements to VRAM management, the results can vary depending on hardware configurations and rendering conditions, with some scenarios showing higher VRAM usage. For one, sparse textures are no longer defragmented by our VRAM defragmenter. This can lead to increased memory fragmentation where there is plenty of free VRAM available, but only in tiny chunks instead of the single big chunk of memory needed for a resource. Additionally, some meshes are no longer eligible to be demoted from VRAM to system memory like they used to in 12.4.3. These are regressions in 12.4.4; that is to say, they aren’t intentional, and we will fix them.
The side effect of all of this is that this was enough to push some Nvidia users over the edge to trigger the 6xx GPU crash from VRAM pressure. This is why, despite this driver also affecting 12.4.3, more users are affected with 12.4.4 now. Additionally, I have also discovered a bug in the loader code where X-Plane could hang indefinitely on the loading screen if VRAM pressure was high. I have seen reports about indefinite loading hangs in the past but was never able to reproduce this until now and didn’t make the connection to VRAM usage. That is to say, this bug is also not new and has been with X-Plane since at least 12.2.0, but 12.4.4’s VRAM change was enough to trigger this as well for a higher number of users.
TLDR
- X-Plane 12.4.4 introduced new VRAM optimisations designed to reduce memory usage, but some unintended regressions have resulted in higher VRAM consumption in certain scenarios. This has contributed to increased GPU crashes on NVIDIA 6xx drivers and more frequent loading screen hangs under high VRAM pressure.
- We’ve identified these issues and are working on fixes for an upcoming update.
Where we are at now
We’re currently working on X-Plane 12.4.5, which aims to address several issues, including Zink performance regressions, increased VRAM usage, and loading problems under high VRAM pressure. We’re also tackling a number of smaller issues identified through our own testing and your feedback.
Which brings me to this: thank you to everyone who has submitted bug reports! Our intention is not to make the X-Plane experience worse for people, but keep in mind that X-Plane has an incredibly large testing surface; Add-ons, hardware, and drivers all come together at the end of the day and it’s essentially impossible to test every possible configuration.
Please give our betas a try and if you run into issues, let us know about them before reverting back to the previous version.