DRS 20 beta 4 - AMD Linux performance not matching Windows

Issue: Running same system in both Windows / Linux completely different performance is observed with Linux lagging significantly especially in AI tasks and Fusion effects.
Setup:
Basic timeline 4k @ 30fps using 30fps h264 30fps footage recorded via screen recording.
Create magic mask to blend two clips overlaid.
Windows 11 24H2, AMD Adren Driver 25.4.1
Linux Ubuntu 24.04, AMD ROCM 6.4.0, kernel 6.11.0-25, official AMD amdgpu-dkms driver
Davinci Resolve Studio 20b3
Result:
Windows 11: Magic Mask detection speed = 27fps+
Linux: Magic Mask detection speed = 5fps
Similar results can be seen with just adding text to a timeline, on Windows text will play at full speed without render cache, on linux timeline will stall for ~20-30seconds before playing then play at 7fps there after until the text effect has ended in the timeline then it will return to normal frame rate.
Also the same system using an RTX 5080, windows and Linux speed are basically the same, perhaps linux is slightly faster.
System
Software: Davinci Resolve Studio 20.0 Beta 3
Windows: 11 24H2, AMD Adren 25.4.1
Linux: Ubuntu 24.04 (official rocm 6.4.0 using official amdgpu-dkms), CachyOS (rocm 6.4.0)
CPU: AMD 9950X
RAM: 96G DDR5 6000
GPU: AMD Radeon RX 9070 XT @ PCIE5 16X
Disk: Kingston KC3000 4TB SSD
Motherboard: ASUS X870E ProArt
Setup:
Basic timeline 4k @ 30fps using 30fps h264 30fps footage recorded via screen recording.
Create magic mask to blend two clips overlaid.
Windows 11 24H2, AMD Adren Driver 25.4.1
Linux Ubuntu 24.04, AMD ROCM 6.4.0, kernel 6.11.0-25, official AMD amdgpu-dkms driver
Davinci Resolve Studio 20b3
Result:
Windows 11: Magic Mask detection speed = 27fps+
Linux: Magic Mask detection speed = 5fps
Similar results can be seen with just adding text to a timeline, on Windows text will play at full speed without render cache, on linux timeline will stall for ~20-30seconds before playing then play at 7fps there after until the text effect has ended in the timeline then it will return to normal frame rate.
Also the same system using an RTX 5080, windows and Linux speed are basically the same, perhaps linux is slightly faster.
System
Software: Davinci Resolve Studio 20.0 Beta 3
Windows: 11 24H2, AMD Adren 25.4.1
Linux: Ubuntu 24.04 (official rocm 6.4.0 using official amdgpu-dkms), CachyOS (rocm 6.4.0)
CPU: AMD 9950X
RAM: 96G DDR5 6000
GPU: AMD Radeon RX 9070 XT @ PCIE5 16X
Disk: Kingston KC3000 4TB SSD
Motherboard: ASUS X870E ProArt