WDDM mode is extremely slower compared to TCC mode and Linux doesn't have this issue

[FurkanGozukara]

Is this a duplicate?

• I confirmed there appear to be no duplicate issues for this bug and that I agree to the Code of Conduct

Type of Bug

Performance

Component

Not sure

Describe the bug

We are working on generative AI models training

We have noticed that we are getting massive speed loss when we do big data transfer between RAM and GPU on Windows compared to Linux

The hit is such a big scale that Linux runs 2x faster than Windows even more

Same GPU RTX 5090

You can read more info here : kohya-ss/musubi-tuner#700

It turns out if we enable TCC mode on Windows, it gets equal speed as Linux

However again NVIDIA blocked this at driver level

I found a Chinese article with just changing few letters, via patching Patching nvlddmkm.sys, the TCC mode fully becomes working on consumer GPUs

Now my question is, why we can't get Linux speed on Windows?

Moreover it seems like Microsoft added this feature : MCDM

https://learn.microsoft.com/en-us/windows-hardware/drivers/display/mcdm-architecture

Why it is still not available for consumer GPUs?

How can we solve this slowness on Windows compared to Linux?

Thank you so much

How to Reproduce

Do big data transfer between GPU and RAM and compare speed on Windows and Linux

Expected behavior

Same speed as Linux on Windows

Operating System

Windows 11

nvidia-smi output

Microsoft Windows [Version 10.0.26200.7019]
(c) Microsoft Corporation. All rights reserved.

C:\Users\Furkan>nvidia-smi
Sun Nov 2 13:18:29 2025
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 581.57 Driver Version: 581.57 CUDA Version: 13.0 |
+-----------------------------------------+------------------------+----------------------+
| GPU Name Driver-Model | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 NVIDIA GeForce RTX 5090 WDDM | 00000000:01:00.0 Off | N/A |
| 0% 40C P8 10W / 575W | 641MiB / 32607MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
| 1 NVIDIA GeForce RTX 3090 Ti WDDM | 00000000:0E:00.0 On | Off |
| 30% 52C P0 102W / 450W | 8847MiB / 24564MiB | 4% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+

[leofang]

(This issue has nothing to do with any packages provided by the CUDA Python team. Moving to discussion for later triage.)

[FurkanGozukara]

(This issue has nothing to do with any packages provided by the CUDA Python team. Moving to discussion for later triage.)

thanks

so do you have any idea or recommendation of how to solve this issue?

Linux being way faster than Windows