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.

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

Decrease default cuda architectures to build on cause the limitation of the host compiler. #10415

Merged
merged 4 commits into from
Jan 22, 2024

Conversation

levi131
Copy link
Contributor

@levi131 levi131 commented Jan 22, 2024

No description provided.

@levi131 levi131 enabled auto-merge (squash) January 22, 2024 14:58
@levi131 levi131 requested review from oneflow-ci-bot and removed request for oneflow-ci-bot January 22, 2024 14:59
Copy link
Contributor

@jackalcooper
Copy link
Collaborator

这是测试还是要合并

Copy link
Contributor

Speed stats:
GPU Name: NVIDIA GeForce RTX 3080 Ti 

❌ OneFlow resnet50 time: 43.9ms (= 4386.3ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 57.4ms (= 5735.5ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.31 (= 57.4ms / 43.9ms)

OneFlow resnet50 time: 26.2ms (= 2615.5ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 38.5ms (= 3851.2ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.47 (= 38.5ms / 26.2ms)

OneFlow resnet50 time: 19.1ms (= 3817.4ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 36.4ms (= 7278.5ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.91 (= 36.4ms / 19.1ms)

OneFlow resnet50 time: 17.7ms (= 3537.4ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 31.6ms (= 6320.6ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.79 (= 31.6ms / 17.7ms)

OneFlow resnet50 time: 17.3ms (= 3463.8ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 28.7ms (= 5735.3ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.66 (= 28.7ms / 17.3ms)

OneFlow swin dataloader time: 0.201s (= 40.154s / 200, num_workers=1)
PyTorch swin dataloader time: 0.128s (= 25.667s / 200, num_workers=1)
Relative speed: 0.639 (= 0.128s / 0.201s)

OneFlow swin dataloader time: 0.054s (= 10.869s / 200, num_workers=4)
PyTorch swin dataloader time: 0.033s (= 6.591s / 200, num_workers=4)
Relative speed: 0.606 (= 0.033s / 0.054s)

OneFlow swin dataloader time: 0.032s (= 6.307s / 200, num_workers=8)
PyTorch swin dataloader time: 0.017s (= 3.318s / 200, num_workers=8)
Relative speed: 0.526 (= 0.017s / 0.032s)

❌ OneFlow resnet50 time: 49.2ms (= 4917.8ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 64.7ms (= 6465.3ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.31 (= 64.7ms / 49.2ms)

OneFlow resnet50 time: 37.1ms (= 3708.3ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 46.0ms (= 4597.5ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.24 (= 46.0ms / 37.1ms)

OneFlow resnet50 time: 28.3ms (= 5652.6ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 39.1ms (= 7824.1ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.38 (= 39.1ms / 28.3ms)

OneFlow resnet50 time: 25.1ms (= 5011.7ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 38.9ms (= 7775.6ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.55 (= 38.9ms / 25.1ms)

OneFlow resnet50 time: 24.0ms (= 4799.0ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 35.9ms (= 7177.8ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.50 (= 35.9ms / 24.0ms)

@levi131 levi131 disabled auto-merge January 22, 2024 16:21
@levi131
Copy link
Contributor Author

levi131 commented Jan 22, 2024

这是测试还是要合并

推进合并

levi131 added 2 commits January 22, 2024 16:49
….com:Oneflow-Inc/oneflow into lml/decrese_default_CMAKE_CUDA_ARCHITECTURES
Copy link
Contributor

CI failed when running job: Build cu116. PR label automerge has been removed

@levi131 levi131 enabled auto-merge (squash) January 22, 2024 17:07
Copy link
Contributor

Copy link
Contributor

Speed stats:
GPU Name: NVIDIA GeForce RTX 3080 Ti 

❌ OneFlow resnet50 time: 43.7ms (= 4374.3ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 57.6ms (= 5757.9ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.32 (= 57.6ms / 43.7ms)

OneFlow resnet50 time: 26.2ms (= 2617.3ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 38.0ms (= 3800.6ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.45 (= 38.0ms / 26.2ms)

OneFlow resnet50 time: 18.9ms (= 3777.2ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 36.8ms (= 7358.1ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.95 (= 36.8ms / 18.9ms)

OneFlow resnet50 time: 16.9ms (= 3388.4ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 34.2ms (= 6843.3ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 2.02 (= 34.2ms / 16.9ms)

OneFlow resnet50 time: 15.9ms (= 3178.1ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 28.9ms (= 5776.0ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.82 (= 28.9ms / 15.9ms)

OneFlow swin dataloader time: 0.200s (= 40.082s / 200, num_workers=1)
PyTorch swin dataloader time: 0.128s (= 25.599s / 200, num_workers=1)
Relative speed: 0.639 (= 0.128s / 0.200s)

OneFlow swin dataloader time: 0.059s (= 11.848s / 200, num_workers=4)
PyTorch swin dataloader time: 0.032s (= 6.481s / 200, num_workers=4)
Relative speed: 0.547 (= 0.032s / 0.059s)

OneFlow swin dataloader time: 0.032s (= 6.304s / 200, num_workers=8)
PyTorch swin dataloader time: 0.017s (= 3.320s / 200, num_workers=8)
Relative speed: 0.527 (= 0.017s / 0.032s)

❌ OneFlow resnet50 time: 49.2ms (= 4919.7ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 64.6ms (= 6460.9ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.31 (= 64.6ms / 49.2ms)

OneFlow resnet50 time: 36.5ms (= 3646.8ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 45.6ms (= 4560.2ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.25 (= 45.6ms / 36.5ms)

OneFlow resnet50 time: 28.0ms (= 5607.2ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 40.0ms (= 8007.9ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.43 (= 40.0ms / 28.0ms)

OneFlow resnet50 time: 25.3ms (= 5059.9ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 38.6ms (= 7712.7ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.52 (= 38.6ms / 25.3ms)

OneFlow resnet50 time: 23.5ms (= 4697.7ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 37.5ms (= 7493.8ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.60 (= 37.5ms / 23.5ms)

@levi131 levi131 merged commit 06c9ead into master Jan 22, 2024
22 checks passed
@levi131 levi131 deleted the lml/decrese_default_CMAKE_CUDA_ARCHITECTURES branch January 22, 2024 18:56
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

2 participants