Tensorrt actually slows down my render speed?

[lxrswdd]

Hi, I am running the sdxl checkpoint animagineXLV3 using a Nividia 2060s and 32GB RAM.
I have exported a 1024x1024 Tensorrt static engine.

The prompts and hyperparameters are fixed :

(art by shexyo:0.6), solo, 1girl, smile, looking at viewer, standing, arms behind head, long hair, white hair, blue eyes, cat lingerie, choker, neck bell, breasts, masterpiece, best quality, newest
Negative prompt: (worst quality, low quality:1.4), (interlocked fingers, badly drawn hands and fingers, anatomically incorrect hands), blurry, blurry background, bokeh, text, signature, speech bubble
Steps: 24, Sampler: Euler a, CFG scale: 6, Seed: 2982216394, Size: 1024x1024, Model hash: 1449e5b0b9, Model: sdxl_animagineXLV3_v30, VAE hash: 63aeecb90f, VAE: sdxl_vae.safetensors, Clip skip: 2, ENSD: 31337, RNG: NV, Version: v1.7.0

The time taken is actually doubled with Tensorrt:

Activating unet: [TRT] sdxl_animagineXLV3_v30
Loading TensorRT engine: D:\Stable Diffusion\sd.webui\webui\models\Unet-trt\sdxl_animagineXLV3_v30_75f2f05b_cc75_sample=2x4x128x128-timesteps=2-encoder_hidden_states=2x77x2048-y=2x2816.trt

Loaded Profile: 0
        sample = [(2, 4, 128, 128), (2, 4, 128, 128), (2, 4, 128, 128)]
        timesteps = [(2,), (2,), (2,)]
        encoder_hidden_states = [(2, 77, 2048), (2, 77, 2048), (2, 77, 2048)]
        y = [(2, 2816), (2, 2816), (2, 2816)]
        latent = [(2, 4, 128, 128), (2, 4, 128, 128), (2, 4, 128, 128)]
Couldn't find VAE named sdxl_vae_updated.safetensors; using None instead██████████████████████████████████████████████████████████████████████████| 24/24 [02:47<00:00,  7.10s/it]
Restoring base VAE
Applying attention optimization: xformers... done.
VAE weights loaded.
Total progress: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 24/24 [02:59<00:00,  7.47s/it]
Loading VAE weights specified in settings: D:\Stable Diffusion\sd.webui\webui\models\VAE\sdxl_vae.safetensors█████████████████████████████████████| 24/24 [02:59<00:00,  7.10s/it]

Without Tensorrt:

Dectivating unet: [TRT] sdxl_animagineXLV3_v30
*** Error running process_batch: D:\Stable Diffusion\sd.webui\webui\extensions\Stable-Diffusion-WebUI-TensorRT\scripts\trt.py
    Traceback (most recent call last):
      File "D:\Stable Diffusion\sd.webui\webui\modules\scripts.py", line 742, in process_batch
        script.process_batch(p, *script_args, **kwargs)
      File "D:\Stable Diffusion\sd.webui\webui\extensions\Stable-Diffusion-WebUI-TensorRT\scripts\trt.py", line 302, in process_batch
        if self.idx != sd_unet.current_unet.profile_idx:
    AttributeError: 'NoneType' object has no attribute 'profile_idx'

---
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 24/24 [01:24<00:00,  3.54s/it]
Couldn't find VAE named sdxl_vae_updated.safetensors; using None instead██████████████████████████████████████████████████████████████████████████| 24/24 [01:20<00:00,  3.43s/it]
Restoring base VAE
Applying attention optimization: xformers... done.
VAE weights loaded.
Total progress: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 24/24 [01:37<00:00,  4.07s/it]
Total progress: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 24/24 [01:37<00:00,  3.43s/it] 

What is the possible explanation?

[showerstomp]

Batch sizing or incorrect tensor setup.