Benchmarking

Benchmarking

Some people may be interested in the generation speed, vram usage of different settings (sampler, resolution, batch size)

This page details the start of a standardized benchmarking method and is subject to change

Sampler test

Example run

Hardware

• CPU: Xeon E5-2630v2
• RAM: 96GB DDR3
• GPU: NVIDIA RTX 3060 12GB

Software

• OS: Windows 10 LTSC 21H2

prompt:

anime girl holding a giant NVIDIA Tesla A100 GPU graphics card, Anime Blu-Ray boxart, super high detail

seed:

hlky

settings:

height, width =  512
cfg scale = 7.5
batch count, batch size = 1
sampling steps = 50

k_lms
Took 13.58s total (13.58s per image) Peak memory usage: 7615 MiB / 12288 MiB / 61.964%

Output:

---

k_heun
Took 24.09s total (24.09s per image) Peak memory usage: 7667 MiB / 12288 MiB / 62.388%

Output:

---

k_euler
Took 12.35s total (12.35s per image) Peak memory usage: 7663 MiB / 12288 MiB / 62.357%

Output:

---

k_euler_a
Took 12.36s total (12.36s per image) Peak memory usage: 7665 MiB / 12288 MiB / 62.373%

Output:

---

k_dpm_2
Took 24.18s total (24.18s per image) Peak memory usage: 7655 MiB / 12288 MiB / 62.291%

Output:

---

k_dpm_2_a
Took 24.16s total (24.16s per image) Peak memory usage: 7670 MiB / 12288 MiB / 62.417%

Output:

---

PLMS
Took 13.01s total (13.01s per image) Peak memory usage: 7665 MiB / 12288 MiB / 62.371%

Output:

---

DDIM
Took 12.4s total (12.4s per image) Peak memory usage: 7668 MiB / 12288 MiB / 62.395%

Output:

---

A template will be added and a Discussion opened to share results