deepseek-ai / deepseek-ai/DeepEP
Would deepep work in a hybrid GPU networking topology inside a single node(intronode)?
Open
- Dominant language
- Cuda
- Stars
- 10.1k
- Forks
- 1.4k
- Avg merge
- 4d 1h
- Merged PRs (30d)
- 2
Description
Hi,
I have a node as follows:
NUMA 0:
GPU0-GPU1: NVLink pair
GPU2-GPU3: NVLink pair
Cross between those pairs: NODE/PCIe
NUMA 1:
GPU4-GPU5: NVLink pair
Cross NUMA:
GPU0-3 <-> GPU4-5: SYS/PCIe + CPU interconnect
Would I be able to use DeepEP for EP = 4 in this setup? Please do let me know if it's possible.
Contributor guide
No contributing guide indexed for this repository
Assessment
This issue has not been assessed yet.