OpenXR Toolkit (upscaling, world scale, hand tracking...) - Release thread

Don’t quite get the relevance of TV smoothing here. TV content is fixed camera. You are always moving the camera in VR. This is what makes MR difficult. This is what makes DLSSG not usable as-is.

DLSSG is backward interpolation. This technique is highly incompatible with VR due to latency.

(from my other post about reprojection: Motion Reprojection explained - General Discussion & Interests / Virtual Reality (VR) - Microsoft Flight Simulator Forums)

(and here a reference to the latency of DLSSG: Will DLSS 3.0 be supported in VR msfs? - General Discussion & Interests / Virtual Reality (VR) - Microsoft Flight Simulator Forums)


Though if latency is acceptable and solved in 2d games with DLSS3, why could not be developed in the VR with similar latencies or how? more dedicated hardware for DLSS3 in the GPU?

DLSS3 is gonna be implemented in some point in VR sure sure, but like you have experienced, there isn´t interest in NVIDIA staff side apparently nowadays.

Any information about DLSS version updates in MSFS?
Has DLSS version 2.5.1 been implemented already?

AFAIK curent version is

I didn’t try yet, but I guess we can just drop the new dll version to replace the nvngx_dlss.dll file.

1 Like

5800X3D is probably the best CPU selling now (because of its huge L3 cache) but it’s a purely gaming CPU with only 8 cores. A new 7800X3D and more importantly 7900X3D (12-core) and 7950X3D (16-core) CPUs are releasing in February. 7950X3D will absolutely crush Intel in both single-core and multi-core (by 10% to 20% most likely) and will definitely be the best CPU for MSFS due to X3D L3 cache on one chiplet (8 cores). The core management may demand Windows 11 though. But I guess MSFS could be manually set for CPU affinity to the correct X3D cores on Win10. However, because of the new AM5 socket that means a new expensive motherboard and new expensive DDR5 RAM. On the other hand, if one decides to upgrade to an expensive CPU, the rest of the system should be a match.

1 Like

The DLSS version is now discussed here:


I am running 2.5.1 in VR. There is a -noticeable- improvement with ghosting of rapidly changing numbers (on the PFD for example), and the “vasoline smear” is a little bit better too. I can notice that smearing effect on DLSS Performance at night, looking at the city lights and it’s a lot better.

As for frames, for a properly tuned system, I don’t see substantial gains, and without hard data, I can’t be sure where performance increases are. Anecdotally, there appears to be some improvement in frame time (in milliseconds) but the sim is so dynamic it’s hard to really benchmark. I wanted to report back and avoid placebo syndrome.

Disclaimer: In my testing, I ran in to a severe frame rate issue, where all of a sudden frames went from 45 FPS (MR locked) to 8. Reported frame time in OXR Toolkit was CPU=12ms and GPU=23ms, VMEM=63%, CPU utilization was low, and Windows Task Manager reported 33% GPU utilization, and system memory was 53%. I have NO idea why the frame rate was so bad. Exiting WMR and re-entering didn’t solve it. Changing location in the sim via slewing didn’t solve it. I had to completely reload the flight. (Not the sim) Important to note here, I have NEVER seen this behavior before… until I started testing out this DLSS 2.5.1 DLL.

Regardless, the money seems to be in DLSS Balanced in terms of visual clarity, fidelity and performance. It looks great, and overall image/number ghosting appears to have improved slightly and the vasoline smear is reduced.

1 Like

I have also tested the new DLSS dll and can see slight visual improvements (I must say I am not very sensitive to such slight visual changes).
I have not compared the performance, but I have not noticed any degradation.
MSFS seems to run stable with the new DLSS version (test flight of about 40 minutes in TBM).


Hey Matt. I’m sorry for the off topic topic (or is it?)

Would you care to comment on this please? Is this your team? :open_mouth: what does this mean for the common wmr user?

Best regards

7950X3D… will definitely be the best CPU for MSFS due to X3D L3 cache on one chiplet (8 cores).

From what I’ve read, the 7800x3D should be just as good for MSFS since the v-cache CCDs on both chips are expected to be the same, same clocks etc.

Hi @mbucchia just wondering did you ever consider adding the dynamic frame rate limiter discussed a few weeks ago, that could try to limit ones frame rate to multiples of their display refresh automatically.

I’m usually between 35 and 55fps depending on where I’m flying, but currently have to lock to 30fps to make things look smooth in vr although I know in the back of my mind, half of the time I could lock to 45fps but it’d be a pain to keep changing the toolkit settings from 30 to 45 manually per area.

Strangely, flying through dense clouds is ok for me but when I fly through a high level thin film of cloud, I often get stutters. Not sure why, but this thin film of cloud is more taxing on my gpu than dense clouds. Not sure if others gave have experienced this

It is completely off-topic indeed :stuck_out_tongue_winking_eye:

I have explained here what MRTK is. I have personally not been let go, and I continue working on the same things I used to work on before. (no longer true after second wave in February)


It’s in one of my “ideas lists” but not done. This type of feature is unfortunately extremely controversial, think of the whole Motion Reprojection auto-lock and how many are not happy with it and prefer the manual lock. This feature is bound to have the same issues. The problem with throttling is that it makes it virtually impossible to know the actual “best framerate” you could achieve (because you intentionally not achieve it). So you have to make guesses. Most people won’t agree with your guesses. So they will want another 4-5 parameters to control the way things are guessed (sensitivity, delays, thresholds). This is how you make unmanageable features :frowning:


I’m sorry for the off topic. I just stumbled on that bit of news and in regards to everything you’ve done for us here, i wish you all the best in all aspects of your life, so i stopped to think if this has in any way affected you. Glad to see it hasn’t. Also, thank you for the explanation. The term “toolkit” in the news got me confused. I know “our” toolkit has no official ties with MS, but i thought it may have something to do with the OXR Tools, but that they have perhaps misspelled it in the article. Anyyyyway…carry on, thanks for the explanation and have a great day :smiley:


I don’t want to post too much off-topic text here, so let’s try to stick to OpenXR Toolkit. But yes, you are correct. The problem with it is that 7800X3D is a purely gaming CPU like 5800X3D. If you are using the same PC for productivity (video editing, graphic editing, photo editing, 3D rendering etc.) 8 cores are just not enough. 7950X3D covers both worlds. It may or may not be a bit better because of the slightly larger 3D cache size, but it’s impossible to really tell until benchmarks are available.

Yes, I have noticed the same. To me it has been like this all the way along since release.

I do not find the answer in the documentation.
When I use CAS for the sharpening you can’t change the resolution like with NIS and FSR why. ?

Also in the System tab you can also override resolution: why can you change resolution in two different tab (onglet) ?

When I lock the fps (to 22 for example) it does not seem to work (I have always my 30 fps)
I use a vive pro with steamVR.

Hello mbucchia,

just one question… , ok two :slight_smile:
when can we expect some update for oxr tools ( NOT Toolkit ) and what can be positive for MSFS?

ok and second… what about Toolkit? any new version to come?

Tough questions, but not for reasons you can imagine.

I will be going on a leave of absence to take care of family starting next week. I insist, do not read anything into it about my position at Microsoft. This leave had been planned for several months (EDIT: to be clear, I’m talking about a few weeks, I’m just not calling it “vacation” because it won’t be for pleasure). This will probably delay a bit the next release of OpenXR for WMR. The main changes I have plans for are improvements to the motion reprojection, which I honestly think look great and had very positive feedback so far, but I cannot speak whether they will be in 1 or 2 releases from now (for the same reason that I own these changes and I will not be at work for a little while to shepherd them to release). There will also be several HoloLens 2 and OpenXR standard features, but I’m pretty sure this audience cares less about that :smiley:

For my personal projects, there is currently no future release slated for OpenXR Toolkit. I don’t think there is any pending feature worth it, and between 1.2.2 and 1.2.3, most users should be all set depending on their needs. I acknowledge there are stability issues that surfaced in 1.2.3 for certain users only, but they do not have any easy solution. Perhaps there will be a 1.2.4 to revert a change or two from 1.2.3 for stability. The rest, I have no plans today.

For PimaxXR (OpenXR for Pimax), I am trying to put a finishing touch to it in the next 1-2 versions. It will be 99% conformant to the OpenXR standard, which should make it pretty self-sustaining for future users and applications.

I think from now on, I might focus more on developer education around OpenXR & API layers than development myself. You know, the “teach someone to fish and you feed them for a lifetime”. There’s a high demand for features like overlays, controller tracking, etc, all of them I do not have time for. My current OpenXR API layers have served as a basis for a lot of other projects: XRNS, OXRMC, OBSMirror… I will try to create a decent basis for developers to bring new API layers to offer new features. This is a better investment on the long term to ensure that many more developers can contribute their ideas and innovation.