ComfyUI using more memory and crashing after upgrading GPU.

I'm in a weird situation right now....
For a long time I've been using RTX 2060 6GB, 27,9GB RAM (yes 27gb) and 60GB Page file. With this I was able to use Wan 2.2 14b Q8 (15gb) or Safetensors (20gb) models, T2V or I2V without any problems, 480x720, 81 frames, the generation was about 7~10mins.

Now I upgraded my GPU RTX 5060 Ti 16GB, using the same workflow, RAM and Pagefile Comfyui just crash without any error message, it crash after the first generation or it crash while loading the second model, I tried adding some Cleanup nodes but it didn't fixed. The only way I can fix it isto to use a 100GB page file... why is this happening? shouldn't it be better with 16GB Vram?

To mention it happens even with 13gb Wan 2.2 14b models, there isn't any problems using other models like Zimage, Flux Klein and LTX-2.

I reinstalled ComfyUI, I suspected it was the drivers so I did a clean install using DDU, next since the new gpu is Blackwell I updated CUDA from 12.8 to 13.00, same problems.
I did some search, I see that some people had the same problem and it was because of their PSU, so I decided to test that.
First I disabled XMP and Re-Bar on the mobo BIOS and check if I had overclocked my CPU, next I run Furmark it run smoothly, no artifacts everything nice. Next I decided to train a Lora, the GPU runs under full load for 1 or 2h.. no system freeze, BSOD, it finished without any problems. Next I upscaled a video with SeedVR2, again the GPU was running under full load four about 4 or 5h!!!!

I don't know what to do, anyone can suggest something? I could just use 100gb page file but my question is...
HOW ON EARTH a 6GB GPU and 27gb Ram can handle a 20gb wan 2.2 model (watching youtube, several browser tabs open, running background apps like Discord or Steam) without any problems while with a 16GB gpu it can't?

I was hoping to decrease a little the 60gb page file to reduce the load on NVME a little but it looks like I'm going to have to use it even more than I already was using before ☹️