30X Slowdown from 0.3.6 to 0.3.9

[astrologos]

Which version of LM Studio?
Example: LM Studio 0.3.9 Build 6

Which operating system?
Windows

What is the bug?
30X slowdown to 1 tok/sec after update from 0.3.6B8 to 0.3.9B6 on Intel B580 GPU

Logs
see attached

To Reproduce
Observe initial speed using 3.6 Build 8 with Vulkan engine with model DeepSeek-R1-Distill-Qwen-14B-GGUF/DeepSeek-R1-Distill-Qwen-14B-Q4_K_M.gguf
30.1 tok/s with full GPU offload

Update to from 3.6 Build 8 to 3.9 using Arc B580
Observe initial speed using 3.9 Build 6 with Vulkan engine with model DeepSeek-R1-Distill-Qwen-14B-GGUF/DeepSeek-R1-Distill-Qwen-14B-Q4_K_M.gguf
1.2 tok/s with full GPU offload

Problem has been recreated by a colleague using NVIDIA hardware.

What's going on?? And why can't I revert to the previous version??

main.log

[yagil]

Thanks for the bug report.

Can you please:

1. Share a screenshot of Ctrl + Shift + R
2. In addition, share a screenshot of Ctrl + Shift + H
3. Share a screenshot of your GPU tab in Task Manager (windows) when the model behaves this way

Thanks