Skip to content

Add performance data sdxl for RTX 4090 24G and 48G #1041

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

Open
lijunliangTG wants to merge 11 commits into
base: main
Choose a base branch
from
Open

Add performance data sdxl for RTX 4090 24G and 48G #1041

Changes from all commits
Commits
Show all changes
11 commits
Select commit Hold shift + click to select a range
eb122f2
sdxl data
lijunliangTG Jul 25, 2024
29dfb2a
remove extra chars
lijunliangTG Jul 25, 2024
73cffe4
Fix typo in CPU comment: change 'core' to 'cores'
lijunliangTG Jul 25, 2024
d9500e7
add 4090(48G)
lijunliangTG Jul 25, 2024
a833209
Merge branch 'add_Performance_data_sdxl' of github.com:siliconflow/on…
lijunliangTG Jul 25, 2024
d127c51
remove extra black
lijunliangTG Jul 25, 2024
facf6c3
Merge branch 'main' into add_Performance_data_sdxl
lijunliangTG Jul 26, 2024
c9b1751
add reserved CUDA memory
lijunliangTG Jul 26, 2024
efabda7
Merge branch 'add_Performance_data_sdxl' of github.com:siliconflow/on…
lijunliangTG Jul 26, 2024
85c67c1
add Max reserved CUDA memory Used of 4090(32G) 4090(48G)
lijunliangTG Jul 26, 2024
e7becc6
Merge branch 'main' into add_Performance_data_sdxl
lijunliangTG Jul 26, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
27 changes: 15 additions & 12 deletions onediff_diffusers_extensions/examples/sdxl/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -66,23 +66,26 @@ python3 benchmarks/text_to_image.py \
## Performance comparison

Testing on NVIDIA GeForce RTX 3090 / 4090, with image size of 1024*1024, iterating 20 steps:
| Metric | RTX 3090 1024*1024 | RTX 4090 1024*1024 |
| ------------------------------------ | --------------------- | --------------------- |
| Data update date (yyyy-mm-dd) | 2024-07-10 | 2024-07-10 |
| PyTorch iteration speed | 4.08 it/s | 6.93 it/s |
| OneDiff iteration speed | 7.21 it/s (+76.7%) | 13.92 it/s (+100.9%) |
| PyTorch E2E time | 5.60 s | 3.23 s |
| OneDiff E2E time | 3.41 s (-39.1%) | 1.67 s (-48.3%) |
| PyTorch Max Mem Used | 10.467 GiB | 10.467 GiB |
| OneDiff Max Mem Used | 12.004 GiB | 12.021 GiB |
| PyTorch Warmup with Run time | | |
| OneDiff Warmup with Compilation time | 474.36 s <sup>1</sup> | 236.54 s <sup>2</sup> |
| OneDiff Warmup with Cache time | 306.84 s | 104.57 s |
| Metric | RTX 3090 1024*1024 | RTX 4090 1024*1024 |RTX 4090(32G) 1024*1024|RTX 4090(48G) 1024*1024|RTX 4090(48G) 2048*2048|
| ------------------------------------ | --------------------- | --------------------- | --------------------- | --------------------- |---------------------- |
| Data update date (yyyy-mm-dd) | 2024-07-10 | 2024-07-10 |2024-07-25 |2024-07-25 |2024-07-25 |
| PyTorch iteration speed | 4.08 it/s | 6.93 it/s |6.158 it/s |7.585 it/s |1.649 it/s |
| OneDiff iteration speed | 7.21 it/s (+76.7%) | 13.92 it/s (+100.9%) |11.789 it/s (+91.4%) |14.895 it/s (+96.3%) |2.967 it/s (+79.9%) |
Comment thread
strint marked this conversation as resolved.
| PyTorch E2E time | 5.60 s | 3.23 s |3.674s |2.972 s |13.422s |
| OneDiff E2E time | 3.41 s (-39.1%) | 1.67 s (-48.3%) |2.029s (-44.8%) |1.571s (-47.2%) |7.688s(-42.8%) |
| PyTorch Max Mem Used | 10.467 GiB | 10.467 GiB |10.465 GiB |10.471 GiB |21.723 GiB |
| OneDiff Max Mem Used | 12.004 GiB | 12.021 GiB |12.002 GiB |12.013 GiB |24.015 GiB |
| PyTorch Max reserved CUDA memory Used| | |14.078 GiB |14.078 GiB |35.615 GiB |
| OneDiff Max reserved CUDA memory Used| | |14.873 GiB |14.859 GiB |35.666 GiB |
| PyTorch Warmup with Run time | | | | | |
| OneDiff Warmup with Compilation time | 474.36 s <sup>1</sup> | 236.54 s <sup>2</sup> |142.691 s <sup>3</sup> |287.011 s <sup>3</sup> |502.223 s <sup>3</sup> |
| OneDiff Warmup with Cache time | 306.84 s | 104.57 s |142.992s |132.207 s |363.051 s |

@strint strint Jul 26, 2024

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

oom 时的报错信息是什么，可以发下

@strint strint Jul 26, 2024

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

可以用这个观察下 2048*2048 时的显存占用：https://pytorch.org/tutorials/recipes/recipes/profiler_recipe.html#using-profiler-to-analyze-memory-consumption

@lijunliangTG lijunliangTG Jul 26, 2024

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

报错信息如下 
[2024-07-26 16:43:01,384] [INFO] [graphs.py:34:dynamic_graphed_callable] Dynamically CUDA graphing ModuleToBeGraphed
/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/cuda/graphs.py:83: UserWarning: The CUDA Graph is empty. This usually means that the graph was attempted to be captured on wrong device or stream. (Triggered internally at ../aten/src/ATen/cuda/CUDAGraph.cpp:222.)
  super().capture_end()
[2024-07-26 16:48:29,566] [ERROR] [graphs.py:112:make_graphed_callable] Failed to capture CUDA Graph, please try without it
[2024-07-26 16:48:29,567] [ERROR] [graphs.py:38:dynamic_graphed_callable] Failed to dynamically CUDA graph ModuleToBeGraphed
Traceback (most recent call last):
  File "/root/project/nexfort/src/nexfort/cuda/graphs.py", line 110, in make_graphed_callable
    static_outputs = func(*static_inputs, **static_kwarg_inputs)
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1553, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1562, in _call_impl
    return forward_call(*args, **kwargs)
  File "/root/project/nexfort/src/nexfort/fx_compiler/fx_compiler.py", line 88, in forward
    return self.compiled_fn(*args)
  File "/root/project/nexfort/src/nexfort/fx_compiler/overrides.py", line 74, in wrapper
    return func(*args, **kwargs)
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/_dynamo/eval_frame.py", line 600, in _fn
    return fn(*args, **kwargs)
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/_functorch/aot_autograd.py", line 987, in forward
    return compiled_fn(full_args)
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/_functorch/_aot_autograd/runtime_wrappers.py", line 217, in runtime_wrapper
    all_outs = call_func_at_runtime_with_args(
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/_functorch/_aot_autograd/utils.py", line 120, in call_func_at_runtime_with_args
    out = normalize_as_list(f(args))
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/_functorch/_aot_autograd/runtime_wrappers.py", line 451, in wrapper
    return compiled_fn(runtime_args)
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/_inductor/codecache.py", line 1131, in __call__
    return self.current_callable(inputs)
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/_inductor/compile_fx.py", line 944, in run
    return model(new_inputs)
  File "/tmp/torchinductor_root/cb/ccbplrs7ajzxwnuf4q4zztbyhyulafddo6say73bhvlyhhozur3c.py", line 2917, in call
    buf351 = torch.ops.nexfort_cuda.cudnn_convolution_bias_add_act.default(buf350, arg102_1, arg103_1, None, None, [1, 1], [1, 1], [1, 1], False, [0, 0], 1, None)
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/_ops.py", line 667, in __call__
    return self_._op(*args, **kwargs)
RuntimeError: FIND was unable to find an engine to execute this computation

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/root/project/nexfort/src/nexfort/cuda/graphs.py", line 36, in dynamic_graphed_callable
    cached_callable = simple_make_graphed_callable(func, args, kwargs, warmups=warmups)
  File "/root/project/nexfort/src/nexfort/cuda/graphs.py", line 58, in simple_make_graphed_callable
    return make_graphed_callable(
  File "/root/project/nexfort/src/nexfort/cuda/graphs.py", line 109, in make_graphed_callable
    with torch.cuda.graph(fwd_graph, pool=execution_env.mempool, stream=execution_env.stream):
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/cuda/graphs.py", line 185, in __exit__
    self.cuda_graph.capture_end()
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/cuda/graphs.py", line 83, in capture_end
    super().capture_end()
RuntimeError: CUDA error: operation failed due to a previous error during capture
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1
Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.

Traceback (most recent call last):
  File "/root/project/onediff/benchmarks/text_to_image.py", line 428, in <module>
    main()
  File "/root/project/onediff/benchmarks/text_to_image.py", line 360, in main
    pipe(**get_kwarg_inputs())
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
    return func(*args, **kwargs)
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py", line 1289, in __call__
    image = self.vae.decode(latents, return_dict=False)[0]
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/diffusers/utils/accelerate_utils.py", line 46, in wrapper
    return method(self, *args, **kwargs)
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/diffusers/models/autoencoders/autoencoder_kl.py", line 314, in decode
    decoded = self._decode(z).sample
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/diffusers/models/autoencoders/autoencoder_kl.py", line 285, in _decode
    dec = self.decoder(z)
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1553, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1562, in _call_impl
    return forward_call(*args, **kwargs)
  File "/root/project/onediff/src/onediff/infer_compiler/backends/nexfort/deployable_module.py", line 27, in forward
    return self._deployable_module_model(*args, **kwargs)
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1553, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1562, in _call_impl
    return forward_call(*args, **kwargs)
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/_dynamo/eval_frame.py", line 433, in _fn
    return fn(*args, **kwargs)
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1553, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1562, in _call_impl
    return forward_call(*args, **kwargs)
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/diffusers/models/autoencoders/vae.py", line 284, in forward
    def forward(
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/_dynamo/eval_frame.py", line 600, in _fn
    return fn(*args, **kwargs)
  File "/root/project/nexfort/src/nexfort/cuda/graphs.py", line 43, in dynamic_graphed_callable
    return cached_callable(*args, **kwargs)
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1553, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1562, in _call_impl
    return forward_call(*args, **kwargs)
  File "/root/project/nexfort/src/nexfort/fx_compiler/fx_compiler.py", line 88, in forward
    return self.compiled_fn(*args)
  File "/root/project/nexfort/src/nexfort/fx_compiler/overrides.py", line 74, in wrapper
    return func(*args, **kwargs)
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/_dynamo/eval_frame.py", line 600, in _fn
    return fn(*args, **kwargs)
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/_functorch/aot_autograd.py", line 987, in forward
    return compiled_fn(full_args)
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/_functorch/_aot_autograd/runtime_wrappers.py", line 217, in runtime_wrapper
    all_outs = call_func_at_runtime_with_args(
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/_functorch/_aot_autograd/utils.py", line 120, in call_func_at_runtime_with_args
    out = normalize_as_list(f(args))
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/_functorch/_aot_autograd/runtime_wrappers.py", line 451, in wrapper
    return compiled_fn(runtime_args)
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/_inductor/codecache.py", line 1131, in __call__
    return self.current_callable(inputs)
  File "/root/anaconda3/envs/sd2/lib/python3.10/site-packages/torch/_inductor/compile_fx.py", line 944, in run
    return model(new_inputs)
  File "/tmp/torchinductor_root/cb/ccbplrs7ajzxwnuf4q4zztbyhyulafddo6say73bhvlyhhozur3c.py", line 2726, in call
    buf267 = empty_strided_cuda((1, 512, 1024, 1024), (536870912, 1, 524288, 512), torch.float32)
torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 2.00 GiB. GPU 0 has a total capacity of 31.51 GiB of which 196.06 MiB is free. Including non-PyTorch memory, this process has 31.31 GiB memory in use. Of the allocated memory 28.34 GiB is allocated by PyTorch, with 19.83 GiB allocated in private pools (e.g., CUDA Graphs), and 2.63 GiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)

@strint strint Jul 26, 2024

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more. 

torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 2.00 GiB. GPU 0 has a total capacity of 31.51 GiB of which 196.06 MiB is free. Including non-PyTorch memory, this process has 31.31 GiB memory in use. Of the allocated memory 28.34 GiB is allocated by PyTorch, with 19.83 GiB allocated in private pools (e.g., CUDA Graphs), and 2.63 GiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)

this process has 31.31 GiB memory in use

看起来是否会 oom，Max reserved CUDA memory Used 更有参考价值。

@lijunliangTG lijunliangTG Jul 26, 2024

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

image


<sup>1</sup> OneDiff Warmup with Compilation time is tested on Intel(R) Xeon(R) Silver 4314 CPU @ 2.40GHz. Note this is just for reference, and it varies a lot on different CPU.

<sup>2</sup> AMD EPYC 7543 32-Core Processor.

<sup>3</sup> Intel(R) Xeon(R) Gold 6150 CPU @ 2.70GHz (8 cores).

## Dynamic shape for SDXL

Expand Down