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

Facing scaling issue for GSPMM kernel operation on X86 machine #7845

Open
choudhary-devang opened this issue Dec 10, 2024 · 1 comment
Open

Comments

@choudhary-devang
Copy link

choudhary-devang commented Dec 10, 2024

ICELAKE

Input Shape | 1 Core | 2 Cores | 4 Cores | 8 Cores | 16 Cores | 32 Cores
[11658, 128] | 150.9 ms |113.41ms | 89.54ms | 81.23ms | 82.71ms | 166.92ms
[11410, 128] | 150.56ms | 109.93ms | 87.33ms | 80.08ms | 82.58ms | 157.84ms
[11070, 128] | 145.78ms | 109.69ms | 88.50ms | 81.51ms | 82.56ms | 165.82ms
[11172, 128] | 144.81ms | 109.81ms | 87.17ms | 79.12ms | 81.70ms | 162.82ms
[10938, 128] | 143.74ms | 109.54ms | 86.50ms | 79.06ms | 81.25ms | 160.59ms
[11096, 128] | 141.87ms | 109.44ms | 86.16ms | 78.73ms | 82.35ms | 164.38ms
[10740, 128] | 136.91ms | 105.59ms | 85.67ms | 78.73ms | 80.23ms | 165.82ms
[10464, 128] | 139.91ms | 105.08ms | 84.59ms | 77.68ms | 79.87ms | 159.14ms
[10168, 128] | 139.27ms | 102.07ms | 83.99ms | 76.57ms | 80.32ms | 166.89ms
[10066, 128] | 131.42ms | 101.85ms | 83.10ms | 76.48ms | 78.29ms | 162.82ms
[10092, 128] | 134.06ms | 102.55ms | 83.88ms | 75.25ms | 78.29ms | 164.52ms
[10482, 128] | 133.30ms | 104.64ms | 84.94ms | 77.78ms | 81.43ms | 165.23ms
[10206, 128] | 133.43ms | 103.74ms | 83.10ms | 76.26ms | 80.88ms | 159.14ms
[10398, 128] | 132.03ms | 102.55ms | 83.88ms | 76.85ms | 79.89ms | 157.77ms
[10344, 128] | 135.24ms | 103.36ms | 83.99ms | 76.64ms | 80.07ms | 160.32ms
[10266, 128] | 134.60ms | 102.71ms | 84.27ms | 76.64ms | 79.05ms | 157.83ms
[10590, 128] | 132.03ms | 104.56ms | 84.73ms | 77.95ms | 78.88ms | 165.23ms
[10144, 128] | 136.40ms | 100.83ms | 83.52ms | 76.44ms | 78.88ms | 160.32ms
[10194, 128] | 133.39ms | 102.65ms | 82.80ms | 76.33ms | 92.43ms | 164.38ms

this is the bench marking results which I took from MESHGNN model
as it can observed from the results that the spmm operation is not scaling properly after 8 cores these are the inference results

script which i used:-

image

setup details :-

dgl = 2.5 (build form source)
torch = 2.5 (pip installed)

@choudhary-devang
Copy link
Author

@itaraban, @jermainewang, @BarclayII can you please look into it!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant