search

DeepGraphLearning / graphvite

GraphVite: A General and High-performance Graph Embedding System
https://graphvite.io
Apache License 2.0
1.21k stars 151 forks source link

Check failed: gpu_memory_cost < gpu_memory_limit Can't satisfy the specified GPU memory limit #64

Open etorf opened 4 years ago

etorf commented 4 years ago

So I am trying to run graphvite to embed a very large graph - 800M edges. I am running on 4GPUs each with 16GB. My setup is as follows:

app = gap.GraphApplication(dim=128)
app.build(episode_size=2500)

But this fails with "Check failed: gpu_memory_cost < gpu_memory_limit Can't satisfy the specified GPU memory limit".

Any ideas?

KiddoZhu commented 4 years ago

How many nodes do you have in the graph?

It seems like the nodes can't be hold in 16GB memory. You may increase num_partition to some multiplier of your GPU count to reduce the memory cost.