Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Speed Test #2

Open
3 of 4 tasks
78Alpha opened this issue Mar 6, 2023 · 31 comments
Open
3 of 4 tasks

Speed Test #2

78Alpha opened this issue Mar 6, 2023 · 31 comments
Assignees
@78Alpha
Labels
documentation Improvements or additions to documentation

Comments

@78Alpha
Copy link
Owner

78Alpha commented Mar 6, 2023 •
edited
Loading

Testing to see the maximum performance of the available cards and what settings are needed to hit those top speeds. Direct code alterations are not permitted. This is meant to test end user speed.

Commit: 0cc0ee1

Standards:

  • V1-5-pruned-emaonly
  • 512 x 512
  • cfg 7
  • Euler a
  • Steps 20

Links will direct to peak performance settings
@78Alpha 78Alpha added the documentation Improvements or additions to documentation label Mar 6, 2023
@78Alpha 78Alpha self-assigned this Mar 6, 2023
@78Alpha
Copy link
Owner Author

78Alpha commented Mar 6, 2023 •
edited
Loading

RTX 3070

Test 1

Command Line Arguments: N/A
XFORMERS: False
Torch: 1.13.1+cu117
CuDNN: Default
Triton: False
Opt-channelslast: False
Precision: Autocast
Batch Size: 1

Result:

9.16 it/s raw | 7.90 it/s load
9.16 it/s avg @ BS1

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 6, 2023

RTX 3070

Test 2

Command Line Arguments: N/A
XFORMERS: False
Torch: 1.13.1+cu117
CuDNN: Default
Triton: False
Opt-channelslast: False
Precision: Autocast
Batch Size: 8

Result:

1.38 it/s raw | 1.20 it/s load
11.04 it/s avg @ BS8

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 7, 2023

RTX 3070

Test 3

Command Line Arguments: N/A
XFORMERS: False
Torch: 1.13.1+cu117
CuDNN: 8.8.0
Triton: False
Opt-channelslast: False
Precision: Autocast
Batch Size: 8

Result:

1.4 it/s raw | 1.21 it/s load
11.2 it/s avg @ BS8

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 7, 2023 •
edited
Loading

RTX 3070

Test 4

Command Line Arguments: --opt-channelslast
XFORMERS: False
Torch: 1.13.1+cu117
CuDNN: 8.8.0
Triton: False
Opt-channelslast: True
Precision: Autocast
Batch Size: 8

Result:

1.36 it/s raw | 1.16 it/s load
10.88 it/s avg @ BS8

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 7, 2023

RTX 3070

Test 5

Command Line Arguments: --opt-channelslast --xformers
XFORMERS: 0.016rc425
Torch: 1.13.1+cu117
CuDNN: 8.8.0
Triton: False
Opt-channelslast: True
Precision: Autocast
Batch Size: 8

Result:

1.77 it/s raw | 1.44 it/s load
14.16 it/s avg @ BS8

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 7, 2023 •
edited
Loading

RTX 3070

Test 6

Command Line Arguments: --xformers
XFORMERS: 0.016rc425
Torch: 1.13.1+cu117
CuDNN: 8.8.0
Triton: False
Opt-channelslast: False
Precision: Autocast
Batch Size: 8

Result:

1.8 it/s raw | 1.51 it/s load
14.4 it/s avg @ BS8

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 7, 2023

RTX 3070

Test 7

Command Line Arguments: --xformers
XFORMERS: 0.016rc425
Torch: 1.13.1+cu117
CuDNN: 8.8.0
Triton: False
Opt-channelslast: False
Precision: Autocast
Batch Size: 1

Result:

10.64 it/s raw | 9.17 it/s load
10.64 it/s avg @ BS1

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 7, 2023

RTX 3070

Test 7

Command Line Arguments: --xformers --opt-channelslast
XFORMERS: 0.016rc425
Torch: 1.13.1+cu117
CuDNN: 8.8.0
Triton: False
Opt-channelslast: True
Precision: Autocast
Batch Size: 1

Result:

11.4 it/s raw | 9.58 it/s load
11.4 it/s avg @ BS1

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 7, 2023

RTX 3070

Test 8

Command Line Arguments: --xformers --opt-channelslast
XFORMERS: 0.0.17+b89a493.d20230306
Torch: 1.13.1+cu117
CuDNN: 8.8.0
Triton: False
Opt-channelslast: True
Precision: Autocast
Batch Size: 1

Result:

11.43 it/s raw | 9.55 it/s load
11.43 it/s avg @ BS1

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 7, 2023

RTX 3070

Test 9

Command Line Arguments: --xformers --opt-channelslast
XFORMERS: 0.0.17+b89a493.d20230306
Torch: 1.13.1+cu117
CuDNN: 8.8.0
Triton: False
Opt-channelslast: True
Precision: Autocast
Batch Size: 8

Result:

1.76 it/s raw | 1.42 it/s load
14.08 it/s avg @ BS8

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 7, 2023

TESLA P40

Test 1

Command Line Arguments: N/A
XFORMERS: False
Torch: 1.13.1+cu117
CuDNN: Default
Triton: False
Opt-channelslast: False
Precision: Autocast
Batch Size: 1

Result:

2.54 it/s raw | 2.38 it/s load
2.54 it/s avg @ BS1

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 7, 2023 •
edited
Loading

TESLA P40

Test 2

Command Line Arguments: N/A
XFORMERS: False
Torch: 1.13.1+cu117
CuDNN: Default
Triton: False
Opt-channelslast: False
Precision: Autocast
Batch Size: 8

Result:

1.83 s/it raw | 2.02 s/it load
4.37 it/s avg @ BS8

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 7, 2023

TESLA P40

Test 3

Command Line Arguments: N/A
XFORMERS: False
Torch: 1.13.1+cu117
CuDNN: Default
Triton: False
Opt-channelslast: False
Precision: Autocast
Batch Size: 1

Other: Upcast Cross Attention to FP32

Result:

2.25 it/s raw | 2.12 it/s load
2.25 it/s avg @ BS1

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 7, 2023

TESLA P40

Test 4

Command Line Arguments: N/A
XFORMERS: False
Torch: 1.13.1+cu117
CuDNN: Default
Triton: False
Opt-channelslast: False
Precision: Autocast
Batch Size: 6

Other: Upcast Cross Attention to FP32

Result:

1.73 s/it raw | 1.88 s/it load
3.46 it/s avg @ BS6

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 7, 2023

TESLA P40

Test 5

Command Line Arguments: N/A
XFORMERS: False
Torch: 1.13.1+cu117
CuDNN: Default
Triton: False
Opt-channelslast: False
Precision: Autocast
Batch Size: 4

Other: Upcast Cross Attention to FP32

Result:

1.27 s/it raw | 1.36 s/it load
3.14 it/s avg @ BS4

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 7, 2023

TESLA P40

Test 6

Command Line Arguments: --no-half --precision full --no-half-vae
XFORMERS: False
Torch: 1.13.1+cu117
CuDNN: Default
Triton: False
Opt-channelslast: False
Precision: FULL
Batch Size: 1

Result:

2.71 it/s raw | 2.54 it/s load
2.71 it/s avg @ BS1

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 7, 2023

TESLA P40

Test 7

Command Line Arguments: --no-half --precision full --no-half-vae
XFORMERS: False
Torch: 1.13.1+cu117
CuDNN: Default
Triton: False
Opt-channelslast: False
Precision: FULL
Batch Size: 8

Result:

2.18 s/it raw | 2.38 s/it load
3.66 it/s avg @ BS8

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 8, 2023

TESLA P40

Test 8

Command Line Arguments: --no-half --precision full --no-half-vae --opt-channelslast
XFORMERS: False
Torch: 1.13.1+cu117
CuDNN: Default
Triton: False
Opt-channelslast: True
Precision: FULL
Batch Size: 1

Result:

2.44 it/s raw | 2.24 it/s load
2.44 it/s avg @ BS1

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 8, 2023

TESLA P40

Test 9

Command Line Arguments: --no-half --precision full --no-half-vae --opt-channelslast
XFORMERS: False
Torch: 1.13.1+cu117
CuDNN: Default
Triton: False
Opt-channelslast: True
Precision: FULL
Batch Size: 8

Result:

2.69 s/it raw | 2.97 s/it load
2.97 it/s avg @ BS8

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 8, 2023 •
edited
Loading

TESLA P40

Test 10

Command Line Arguments: --no-half --precision full --no-half-vae --xformers
XFORMERS: 0.016rc425
Torch: 1.13.1+cu117
CuDNN: Default
Triton: False
Opt-channelslast: False
Precision: FULL
Batch Size: 1

Result:

3.31 it/s raw | 3.04 it/s load
3.31 it/s avg @ BS1

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 8, 2023 •
edited
Loading

TESLA P40

Test 11

Command Line Arguments: --no-half --precision full --no-half-vae --xformers
XFORMERS: 0.016rc425
Torch: 1.13.1+cu117
CuDNN: Default
Triton: False
Opt-channelslast: False
Precision: FULL
Batch Size: 8

Result:

1.65 s/it raw | 1.85 s/it load
4.84 it/s avg @ BS8

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 8, 2023 •
edited
Loading

TESLA P40

Test 12

Command Line Arguments: --no-half --precision full --no-half-vae --xformers
XFORMERS: 0.016rc425
Torch: 1.13.1+cu117
CuDNN: 8.8.0
Triton: False
Opt-channelslast: False
Precision: FULL
Batch Size: 1

Result:

3.78 it/s raw | 3.51 it/s load
3.78 it/s avg @ BS1

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 8, 2023 •
edited
Loading

TESLA P40

Test 13

Command Line Arguments: --no-half --precision full --no-half-vae --xformers
XFORMERS: 0.016rc425
Torch: 1.13.1+cu117
CuDNN: 8.8.0
Triton: False
Opt-channelslast: False
Precision: FULL
Batch Size: 8

Result:

1.65 s/it raw | 1.82 s/it load
4.84 it/s avg @ BS8

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 8, 2023 •
edited
Loading

TESLA P40

Test 14

Command Line Arguments: --no-half --precision full --no-half-vae --xformers
XFORMERS: 0.0.17+b89a493.d20230307
Torch: 1.13.1+cu117
CuDNN: 8.8.0
Triton: False
Opt-channelslast: False
Precision: FULL
Batch Size: 1

Result:

3.74 it/s raw | 3.46 it/s load
3.74 it/s avg @ BS1

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 8, 2023

TESLA P40

Test 15

Command Line Arguments: --no-half --precision full --no-half-vae --xformers
XFORMERS: 0.0.17+b89a493.d20230307
Torch: 1.13.1+cu117
CuDNN: 8.8.0
Triton: False
Opt-channelslast: False
Precision: FULL
Batch Size: 8

Result:

1.67 s/it raw | 1.84 s/it load
4.79 it/s avg @ BS8

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 19, 2023

RTX 3090

Test 1

Command Line Arguments: N/A
XFORMERS: False
Torch: 1.13.1+cu117
CuDNN: Default
Triton: False
Opt-channelslast: False
Precision: Autocast
Batch Size: 1

Result:

12.91 it/s raw | 11.34 it/s load
12.91 it/s avg @ BS1

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 19, 2023

RTX 3090

Test 2

Command Line Arguments: N/A
XFORMERS: False
Torch: 1.13.1+cu117
CuDNN: Default
Triton: False
Opt-channelslast: False
Precision: Autocast
Batch Size: 8

Result:

2.24 it/s raw | 1.88 it/s load
17.92 it/s avg @ BS8

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 19, 2023 •
edited
Loading

RTX 3090

Test 3

Command Line Arguments: --autolaunch --xformers
XFORMERS: 0.0.16rc425
Torch: 1.13.1+cu117
CuDNN: Default
Triton: False
Opt-channelslast: False
Precision: Autocast
Batch Size: 1

Result:

16.95 it/s raw | 14.20 it/s load
16.95 it/s avg @ BS1

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 19, 2023

RTX 3090

Test 4

Command Line Arguments: --autolaunch --xformers
XFORMERS: 0.0.16rc425
Torch: 1.13.1+cu117
CuDNN: Default
Triton: False
Opt-channelslast: False
Precision: Autocast
Batch Size: 8

Result:

3.10 it/s raw | 2.46 it/s load
24.8 it/s avg @ BS8

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 19, 2023 •
edited
Loading

RTX 3090

Test 5

Command Line Arguments: --autolaunch --xformers --opt-channelslast
XFORMERS: 0.0.16rc425
Torch: 1.13.1+cu117
CuDNN: Default
Triton: False
Opt-channelslast: True
Precision: Autocast
Batch Size: 1

Result:

16.37 it/s raw | 13.39 it/s load
16.37 it/s avg @ BS1

@78Alpha
Copy link
Owner Author

78Alpha commented Mar 19, 2023

RTX 3090

Test 5

Command Line Arguments: --autolaunch --xformers --opt-channelslast
XFORMERS: 0.0.16rc425
Torch: 1.13.1+cu117
CuDNN: Default
Triton: False
Opt-channelslast: True
Precision: Autocast
Batch Size: 8

Result:

3.07 it/s raw | 2.37 it/s load
24.56 it/s avg @ BS8

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees

@78Alpha 78Alpha

Labels
documentation Improvements or additions to documentation
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
1 participant
@78Alpha