Skip to content

Benchmark

Disty0 edited this page Feb 9, 2024 · 17 revisions

Benchmark

To run standardized benchmark, you can use UI -> System -> Benchmark feature or via CLI using cli/run-benchmark.py script.

It runs identical tests, but often CLI is faster due to lower overhead.

Environment

  • Hardware: nVidia RTX 4090 with i9-13900KF
  • Packages: Torch 2.1.0 with CUDA 12.1 and cuDNN 8.9
  • Params: model=SD15 | batch-size=4 | batch-count=4 | steps=50 | resolution=512px | sampler=Euler A

Results

Basic tests using UI:

Diffusers Original
Precision Params SDP xFormers SDP xFormers None
FP32 Default 33.0 20.0
BF16 Default 73.0 45.5
FP16 Default 73.0 75.0 48.0 48.6 17.3
NHWC (channels last) 72.0
HyperTile (256) 79.0
ToMe (0.5) 77.0
Model no-move (medvram) 85.0
VAE no-slicing, no-tiling 73.8
Sequential offload (lowvram) 27.0

Notes: Options

  • All numbers are in it/s and higher is better
  • Test matrix is not full as some options can be combined together (e.g. cuDNN + HyperTile)
    while others cannot (e.g. HyperTile + ToMe)
  • Results may differ on different GPU/CPU combinations
    For example, pairing better CPU with older GPU may benefit from more processing done on CPU and leaving GPU to do only core ML tasks while paring high-end GPU with older CPU may result in lower results since CPU cannot feed enough tasks to GPU
  • Diffusers perform significantly better than original backend on modern hardware since tasks remain on GPU for longer time
    Equally, original backend may perform better on older hardware
  • Running quick tasks such as single image generate at low steps may not be sufficient to fully saturate high-end GPU so results will be lower
  • xFormers have a slight performance advantage over SDP
    However, SDP is a built-in in Torch and "just works" while xFormers needs manual install and its highly version dependent
  • Some extensions can add significant overhead to pre/post processing even if they are not used
  • Not worth consideration: cuDNN, NHWC, inference mode, eval
    • cuDNN full bench finds best math algorithm for specific GPU, but default is nearly identical
    • channels-last should better trigger utilization of tensor cores, but in practise result is nearly identical
    • inference-mode should have more optimizations than default no_grad, but in practise result is nearly identical
    • eval mode should allow for removal of some params in the model, but in pracise result is nearly identical
  • Benefit of BF16 vs FP16 is not performance as much, its ability to run higher numerical ranges so it can perform calculations where FP16 may result in NaN
  • Running in FP32 results in 60% performance drop - if you need FP32, you're leaving a lot on the table
  • Cost of using lowvram is very high as it needs to swap parts of model in-memory. Even using medvram comes at noticeable cost
  • Best: xFormers, FP16, HyperTile, no-model-move, no-slicing/tiling

Compile

Compile type Performance Overhead
cudnn/default 73.5 4
inductor/default 89.0 40
inductor/reduce-overhead 92.0 40
inductor/max-autotune 91.0 220
nvfuser/default 84.0 5
cudagraphs/reduce-overhead 85.0 14
stable-fast/sdp 96.0 76
stable-fast/xformers 96.0 101
stable-fast/full-graph 94.0 96

Notes: Compile

  • Performance numbers is in it/s and higher is better
  • Overhead is time in seconds needed to optimize a model with specific params and lower is better
    Model needs compile on initial generate, but it may also need a recompile if params such as resolution of batch size change
  • Model compile may not be compatible with any method that modifies underlying model,
    including loading Lora weights on top of a model
  • stable-fast compile backend requires that package is manually installed on the system

Intel ARC

Environment

  • Hardware: Intel ARC 770 LE 16GB with R7 5800X3D & MSI B350M Mortar (PCI-E 3.0) & 48 GB 3200 MHz CL18 RAM
  • OS: Arch Linux with this Docker environment: https://github.com/Disty0/docker-sdnext-ipex
  • Packages: Torch 2.1.0a0+cxx11.abi with IPEX 2.1.10+xpu and MKL / DPCPP 2024.0.0
  • Params: model=SD15 | batch-size=1 | batch-count=1 | steps=40 | resolution=512px | sampler=Euler a | CFG 6

Results

Diffusers Original
Precision Params it/s it/s
BF16 Default 8.54 7.75
FP16 Default 6.92 7.23
FP32 Default 3.73 3.74
BF16 HyperTile (256) 10.03 9.32
BF16 ToMe (0.5) 9.24 8.61
BF16 No IPEX Optimize 8.23 7.82
BF16 Model no-move (medvram) 9.04
BF16 VAE no-slicing, no-tiling 8.67
BF16 Sequential offload (lowvram) 1.60 0.67

API Benchmarks

2024-02-07 22:52:56,406 INFO: {'run-benchmark'}
2024-02-07 22:52:56,407 INFO: {'options': {'prompt': 'photo of two dice on a table', 'negative_prompt': 'foggy, blurry', 'steps': 50, 'sampler_name': 'Euler a', 'width': 512, 'height': 512, 'full_quality': True, 'cfg_scale': 0, 'batch_size': 1, 'n_iter': 1, 'seed': -1}}
2024-02-07 22:52:56,432 INFO: {'version': {'app': 'sd.next', 'updated': '2024-02-07', 'hash': '659ad2e7', 'url': 'https://github.com/vladmandic/automatic/tree/dev'}}
2024-02-07 22:52:56,434 INFO: {'platform': {'arch': 'x86_64', 'cpu': '', 'system': 'Linux', 'release': '6.7.3-arch1-2', 'python': '3.11.6', 'torch': '2.1.0a0+cxx11.abi', 'diffusers': '0.26.2', 'gradio': '3.43.2'}}
2024-02-07 22:52:56,437 INFO: {'model': 'SD1.5/SoteMixV3 [dcc16969a0]'}
2024-02-07 22:52:56,441 INFO: {'system': {'cpu': {'free': 48901079040.00001, 'used': 1533939712, 'total': 50435018752.00001}, 'gpu': {'system': {'free': 17079205888, 'used': 0, 'total': 17079205888}, 'session': {'current': 0, 'peak': 0}}}}
2024-02-07 22:52:56,441 INFO: {'batch-sizes': [1, 1, 2, 4, 8, 12, 16, 24, 32]}
2024-02-07 22:53:10,362 INFO: {'warmup': 13.92}
2024-02-07 22:53:18,182 INFO: {'batch': 1, 'its': 12.81, 'img': 3.9, 'wall': 3.9, 'peak': 2.61, 'oom': False}
2024-02-07 22:53:31,723 INFO: {'batch': 2, 'its': 15.49, 'img': 3.23, 'wall': 6.45, 'peak': 3.07, 'oom': False}
2024-02-07 22:53:55,512 INFO: {'batch': 4, 'its': 17.18, 'img': 2.91, 'wall': 11.64, 'peak': 3.07, 'oom': False}
2024-02-07 22:54:39,504 INFO: {'batch': 8, 'its': 18.4, 'img': 2.72, 'wall': 21.74, 'peak': 3.07, 'oom': False}
2024-02-07 22:55:43,500 INFO: {'batch': 12, 'its': 18.93, 'img': 2.64, 'wall': 31.7, 'peak': 3.07, 'oom': False}
2024-02-07 22:56:58,086 INFO: {'batch': 16, 'its': 21.61, 'img': 2.31, 'wall': 37.01, 'peak': 3.07, 'oom': False}
2024-02-07 22:58:48,560 INFO: {'batch': 24, 'its': 21.92, 'img': 2.28, 'wall': 54.74, 'peak': 3.64, 'oom': False}
2024-02-07 23:01:09,184 INFO: {'batch': 32, 'its': 22.82, 'img': 2.19, 'wall': 70.12, 'peak': 4.06, 'oom': False}

OpenVINO

Environment

  • Hardware: Intel ARC 770 LE 16GB with R7 5800X3D & MSI B350M Mortar (PCI-E 3.0) & 48 GB 3200 MHz CL18 RAM
  • OS: Arch Linux
  • Packages: Torch 2.1.2+cpu and OpenVINO 2023.2.0
  • Params: model=SD15 | batch-size=1 | batch-count=1 | steps=20 | resolution=512px | sampler=Euler a | CFG 6

GPU Results

Diffusers
Precision Params it/s
Default Default 9.21

CPU Results

Diffusers
Precision Params s/it
Default Default 3.00
Default LCM & CFG 0 1.60
INT8 Default 3.30
INT4_SYM Default 4.00
INT4_ASYM Default 4.30
NF4 Default 5.25
FP32 Diffusers & No OpenVINO 4.20

DirectML

  • Hardware: Intel Core i9-14900K, SAPPHIRE AMD Radeon RX 7900 XTX NITRO+ Vapor-X 24GB, SAMSUNG DDR5 32GBx4
  • Operating System: Windows 11 Build 22631
  • Packages: PyTorch 2.0.0 (built with CPU), torch-directml 0.2.0.dev230426
  • Performed using cli/run-benchmark.py script

Peak: 9.36 with batch size 8.

Possible max batch size: 12 (Slow with 12, OOM with 16)

2024-02-08 20:09:31,923 INFO: {'run-benchmark'}
2024-02-08 20:09:31,924 INFO: {'options': {'prompt': 'photo of two dice on a table', 'negative_prompt': 'foggy, blurry', 'steps': 50, 'sampler_name': 'Euler a', 'width': 512, 'height': 512, 'full_quality': True, 'cfg_scale': 0, 'batch_size': 1, 'n_iter': 1, 'seed': -1}}
2024-02-08 20:09:32,005 INFO: {'version': {'app': 'sd.next', 'updated': '2024-02-07', 'hash': '659ad2e7', 'url': 'https://github.com/vladmandic/automatic/tree/dev'}}
2024-02-08 20:09:32,007 INFO: {'platform': {'arch': 'AMD64', 'cpu': 'Intel64 Family 6 Model 183 Stepping 1, GenuineIntel', 'system': 'Windows', 'release': 'Windows-10-10.0.22631-SP0', 'python': '3.10.11', 'torch': '2.0.0+cpu', 'diffusers': '0.26.2', 'gradio': '3.43.2'}}
2024-02-08 20:09:32,013 INFO: {'system': {'cpu': {'free': 136382431232.00002, 'used': 708612096, 'total': 137091043328.00002}, 'gpu': {'error': 'unavailable'}}}
2024-02-08 20:09:32,013 INFO: {'batch-sizes': [1, 1, 2, 4, 8, 12, 16]}
2024-02-08 20:09:51,463 INFO: {'warmup': 19.45}
2024-02-08 20:10:03,837 INFO: {'batch': 1, 'its': 8.06, 'img': 6.2, 'wall': 6.2, 'peak': 0.0, 'oom': False}
2024-02-08 20:10:27,845 INFO: {'batch': 2, 'its': 9.02, 'img': 5.54, 'wall': 11.09, 'peak': 0.0, 'oom': False}
2024-02-08 20:11:12,886 INFO: {'batch': 4, 'its': 9.04, 'img': 5.53, 'wall': 22.12, 'peak': 0.0, 'oom': False}
2024-02-08 20:12:38,582 INFO: {'batch': 8, 'its': 9.36, 'img': 5.34, 'wall': 42.76, 'peak': 0.0, 'oom': False}
2024-02-08 20:15:22,610 INFO: {'batch': 12, 'its': 7.31, 'img': 6.84, 'wall': 82.04, 'peak': 0.0, 'oom': False}
2024-02-08 20:15:23,465 ERROR: {'requested': 16, 'received': 0}
2024-02-08 20:15:24,161 ERROR: {'requested': 16, 'received': 0}
2024-02-08 20:15:24,164 INFO: {'batch': 16, 'its': 1150.12, 'img': 0.04, 'wall': 0.7, 'peak': 0.0, 'oom': False}

ONNX Runtime

  • Hardware: Intel Core i9-14900K, SAPPHIRE AMD Radeon RX 7900 XTX NITRO+ Vapor-X 24GB, SAMSUNG DDR5 32GBx4
  • Operating System: Windows 11 Build 22631
  • Packages: PyTorch 2.2.0 (built with CPU), onnxruntime 1.17.0, onnxruntime-directml 1.17.0
  • Performed using cli/run-benchmark.py script

Peak: 17.58

Possible max batch size: 8 (Not OOM, but very slow with 12 or higher)

2024-02-08 19:20:45,235 INFO: {'run-benchmark'}
2024-02-08 19:20:45,236 INFO: {'options': {'prompt': 'photo of two dice on a table', 'negative_prompt': 'foggy, blurry', 'steps': 50, 'sampler_name': 'Euler a', 'width': 512, 'height': 512, 'full_quality': True, 'cfg_scale': 0, 'batch_size': 1, 'n_iter': 1, 'seed': -1}}
2024-02-08 19:20:45,317 INFO: {'version': {'app': 'sd.next', 'updated': '2024-02-07', 'hash': '659ad2e7', 'url': 'https://github.com/vladmandic/automatic/tree/dev'}}
2024-02-08 19:20:45,318 INFO: {'platform': {'arch': 'AMD64', 'cpu': 'Intel64 Family 6 Model 183 Stepping 1, GenuineIntel', 'system': 'Windows', 'release': 'Windows-10-10.0.22631-SP0', 'python': '3.10.12', 'torch': '2.2.0+cpu', 'diffusers': '0.26.2', 'gradio': '3.43.2'}}
2024-02-08 19:20:45,324 INFO: {'system': {'cpu': {'free': 136392728576.00002, 'used': 698314752, 'total': 137091043328.00002}, 'gpu': {'error': 'unavailable'}}}
2024-02-08 19:20:45,324 INFO: {'batch-sizes': [1, 1, 2, 4, 8, 12, 16]}
2024-02-08 19:21:03,553 INFO: {'warmup': 18.23}
2024-02-08 19:21:12,036 INFO: {'batch': 1, 'its': 11.81, 'img': 4.23, 'wall': 4.23, 'peak': 0.0, 'oom': False}
2024-02-08 19:21:26,618 INFO: {'batch': 2, 'its': 13.79, 'img': 3.62, 'wall': 7.25, 'peak': 0.0, 'oom': False}
2024-02-08 19:21:54,400 INFO: {'batch': 4, 'its': 14.46, 'img': 3.46, 'wall': 13.83, 'peak': 0.0, 'oom': False}
2024-02-08 19:22:40,407 INFO: {'batch': 8, 'its': 17.58, 'img': 2.84, 'wall': 22.75, 'peak': 0.0, 'oom': False}
2024-02-08 19:30:30,903 INFO: {'batch': 12, 'its': 2.56, 'img': 19.57, 'wall': 234.8, 'peak': 0.0, 'oom': False}
2024-02-08 19:40:08,391 INFO: {'batch': 16, 'its': 2.77, 'img': 18.05, 'wall': 288.86, 'peak': 0.0, 'oom': False}

With optimized model using Olive

  • Package: olive-ai 0.4.0

Peak: 54.08

Possible max batch size: Unknown (at least 48)

2024-02-08 18:51:28,096 INFO: {'run-benchmark'}
2024-02-08 18:51:28,097 INFO: {'options': {'prompt': 'photo of two dice on a table', 'negative_prompt': 'foggy, blurry', 'steps': 50, 'sampler_name': 'Euler a', 'width': 512, 'height': 512, 'full_quality': True, 'cfg_scale': 0, 'batch_size': 1, 'n_iter': 1, 'seed': -1}}
2024-02-08 18:51:28,167 INFO: {'version': {'app': 'sd.next', 'updated': '2024-02-07', 'hash': '659ad2e7', 'url': 'https://github.com/vladmandic/automatic/tree/dev'}}
2024-02-08 18:51:28,168 INFO: {'platform': {'arch': 'AMD64', 'cpu': 'Intel64 Family 6 Model 183 Stepping 1, GenuineIntel', 'system': 'Windows', 'release': 'Windows-10-10.0.22631-SP0', 'python': '3.10.12', 'torch': '2.2.0+cpu', 'diffusers': '0.26.2', 'gradio': '3.43.2'}}
2024-02-08 18:51:28,174 INFO: {'system': {'cpu': {'free': 136385822719.99998, 'used': 705220608, 'total': 137091043327.99998}, 'gpu': {'error': 'unavailable'}}}
2024-02-08 18:51:28,174 INFO: {'batch-sizes': [1, 1, 2, 4, 8, 12, 16]}
2024-02-08 18:51:42,445 INFO: {'warmup': 14.27}
2024-02-08 18:51:46,603 INFO: {'batch': 1, 'its': 23.63, 'img': 2.12, 'wall': 2.12, 'peak': 0.0, 'oom': False}
2024-02-08 18:52:00,527 INFO: {'batch': 2, 'its': 35.06, 'img': 1.43, 'wall': 2.85, 'peak': 0.0, 'oom': False}
2024-02-08 18:52:18,711 INFO: {'batch': 4, 'its': 40.34, 'img': 1.24, 'wall': 4.96, 'peak': 0.0, 'oom': False}
2024-02-08 18:52:42,958 INFO: {'batch': 8, 'its': 50.51, 'img': 0.99, 'wall': 7.92, 'peak': 0.0, 'oom': False}
2024-02-08 18:53:13,677 INFO: {'batch': 12, 'its': 53.81, 'img': 0.93, 'wall': 11.15, 'peak': 0.0, 'oom': False}
2024-02-08 18:53:51,700 INFO: {'batch': 16, 'its': 54.08, 'img': 0.92, 'wall': 14.79, 'peak': 0.0, 'oom': False}

API Benchmarks

Using latest version of SD.Next with Torch 2.2.0, CUDA 12.1
Note: Usage of SD.Next via API is faster than via UI due to lower overhead.

Environment: Intel i9-13900KF platform with nVidia RTX 4090 GPU

As you can see, we're reaching peak performance of ~110 it/s using simple settings:

vlado@wsl:~/dev/sdnext-dev $ python cli/run-benchmark.py --maxbatch 32
2024-02-07 11:19:53,026 INFO: {'run-benchmark'}
2024-02-07 11:19:53,027 INFO: {'options': {'prompt': 'photo of two dice on a table', 'negative_prompt': 'foggy, blurry', 'steps': 50, 'sampler_name': 'Euler a', 'width': 512, 'height': 512, 'full_quality': True, 'cfg_scale': 0, 'batch_size': 1, 'n_iter': 1, 'seed': -1}}
2024-02-07 11:19:53,046 INFO: {'version': {'app': 'sd.next', 'updated': '2024-02-07', 'hash': 'd967bd03', 'url': 'https://github.com/vladmandic/automatic/tree/dev'}}
2024-02-07 11:19:53,048 INFO: {'platform': {'arch': 'x86_64', 'cpu': 'x86_64', 'system': 'Linux', 'release': '5.15.146.1-microsoft-standard-WSL2', 'python': '3.11.1', 'torch': '2.2.0+cu121', 'diffusers': '0.26.2', 'gradio': '3.43.2'}}
2024-02-07 11:19:53,051 INFO: {'model': 'sd15/lyriel-v16 [ec6f68ea63]'}
2024-02-07 11:19:53,054 INFO: {'system': {'cpu': {'free': 49020043264.0, 'used': 1495736320, 'total': 50515779584.0}, 'gpu': {'system': {'free': 24110956544, 'used': 1645740032, 'total': 25756696576}, 'session': {'current': 0, 'peak': 0}}}}
2024-02-07 11:19:53,054 INFO: {'batch-sizes': [1, 1, 2, 4, 8, 12, 16, 24, 32]}
2024-02-07 11:19:59,394 INFO: {'warmup': 6.34}
2024-02-07 11:20:02,354 INFO: {'batch': 1, 'its': 33.63, 'img': 1.49, 'wall': 1.49, 'peak': 7.05, 'oom': False}
2024-02-07 11:20:06,213 INFO: {'batch': 2, 'its': 64.3, 'img': 0.78, 'wall': 1.56, 'peak': 7.1, 'oom': False}
2024-02-07 11:20:11,293 INFO: {'batch': 4, 'its': 90.87, 'img': 0.55, 'wall': 2.2, 'peak': 7.18, 'oom': False}
2024-02-07 11:20:19,416 INFO: {'batch': 8, 'its': 104.6, 'img': 0.48, 'wall': 3.82, 'peak': 7.18, 'oom': False}
2024-02-07 11:20:30,850 INFO: {'batch': 12, 'its': 111.96, 'img': 0.45, 'wall': 5.36, 'peak': 7.18, 'oom': False}
2024-02-07 11:20:46,236 INFO: {'batch': 16, 'its': 110.37, 'img': 0.45, 'wall': 7.25, 'peak': 7.18, 'oom': False}
2024-02-07 11:21:09,338 INFO: {'batch': 24, 'its': 109.75, 'img': 0.46, 'wall': 10.93, 'peak': 7.18, 'oom': False}
2024-02-07 11:21:39,623 INFO: {'batch': 32, 'its': 111.38, 'img': 0.45, 'wall': 14.37, 'peak': 7.18, 'oom': False}

With a full optimizations and custom compiled Stable-Fast:
We're reaching peak performance of ~150 it/s (and ~165 it/s using TAESD instead of full VAE):

vlado@wsl:~/dev/sdnext-dev $ python cli/run-benchmark.py --maxbatch 32
2024-02-07 11:29:23,431 INFO: {'run-benchmark'}
2024-02-07 11:29:23,432 INFO: {'options': {'prompt': 'photo of two dice on a table', 'negative_prompt': 'foggy, blurry', 'steps': 50, 'sampler_name': 'Euler a', 'width': 512, 'height': 512, 'full_quality': True, 'cfg_scale': 0, 'batch_size': 1, 'n_iter': 1, 'seed': -1}}
2024-02-07 11:29:23,451 INFO: {'version': {'app': 'sd.next', 'updated': '2024-02-07', 'hash': 'd967bd03', 'url': 'https://github.com/vladmandic/automatic/tree/dev'}}
2024-02-07 11:29:23,453 INFO: {'platform': {'arch': 'x86_64', 'cpu': 'x86_64', 'system': 'Linux', 'release': '5.15.146.1-microsoft-standard-WSL2', 'python': '3.11.1', 'torch': '2.2.0+cu121', 'diffusers': '0.26.2', 'gradio': '3.43.2'}}
2024-02-07 11:29:23,456 INFO: {'model': 'sd15/lyriel-v16 [ec6f68ea63]'}
2024-02-07 11:29:23,459 INFO: {'system': {'cpu': {'free': 49373564927.99999, 'used': 1142214656, 'total': 50515779583.99999}, 'gpu': {'system': {'free': 24110956544, 'used': 1645740032, 'total': 25756696576}, 'session': {'current': 0, 'peak': 0}}}}
2024-02-07 11:29:23,459 INFO: {'batch-sizes': [1, 1, 2, 4, 8, 12, 16, 24, 32]}
2024-02-07 11:29:38,504 INFO: {'warmup': 15.04}
2024-02-07 11:29:38,965 INFO: {'batch': 1, 'its': 78.16, 'img': 0.67, 'wall': 0.23, 'peak': 7.11, 'oom': False}
2024-02-07 11:29:42,630 INFO: {'batch': 2, 'its': 98.91, 'img': 0.51, 'wall': 1.01, 'peak': 7.11, 'oom': False}
2024-02-07 11:29:47,192 INFO: {'batch': 4, 'its': 117.92, 'img': 0.42, 'wall': 1.7, 'peak': 7.11, 'oom': False}
2024-02-07 11:29:54,028 INFO: {'batch': 8, 'its': 142.42, 'img': 0.35, 'wall': 2.81, 'peak': 7.11, 'oom': False}
2024-02-07 11:30:03,161 INFO: {'batch': 12, 'its': 153.29, 'img': 0.33, 'wall': 3.91, 'peak': 7.11, 'oom': False}
2024-02-07 11:30:14,921 INFO: {'batch': 16, 'its': 153.41, 'img': 0.33, 'wall': 5.21, 'peak': 7.11, 'oom': False}
2024-02-07 11:30:33,534 INFO: {'batch': 24, 'its': 144.65, 'img': 0.35, 'wall': 8.3, 'peak': 7.11, 'oom': False}
2024-02-07 11:30:56,914 INFO: {'batch': 32, 'its': 150.59, 'img': 0.33, 'wall': 10.63, 'peak': 7.11, 'oom': False}

Additional performance may be reached by experimenting with different settings, but combination of such may lead to unstable results
For example: channels-last, hyper-tile, tomesd, fused-projections

Clone this wiki locally