gpu indexing

[ranjith502]

i have installed milvus using docker in azure vm
i have data in the data bricks
from databricks inserted data into milvus server (vm). i tested with hnsw ,ivflat index. it is working fine
now iam exploring gpu indexing (GPU_CAGRA)

to create this indexing do i need to have gpu in the vm or databricks

for hnsw indexing iam getting load collection error

MilvusException: <MilvusException: (code=65535, message=show collection failed: load segment failed, OOM if load, maxSegmentSize = 2732.4219522476196 MB, memUsage = 12341.8046875 MB, predictMemUsage = 15074.22663974762 MB, totalMem = 15525.1015625 MB thresholdFactor = 0.900000)>

how to resolve the issue

[yhmo]

To use GPU index, you need to install milvus GPU edition.
CPU edition installation: https://milvus.io/docs/install_standalone-docker-compose.md, it is using this image: docker pull milvusdb/milvus:v2.5.5
GPU edition installation: https://milvus.io/docs/install_standalone-docker-compose-gpu.md, it is using this image: docker pull milvusdb/milvus:v2.5.5-gpu

Yes, you need a nvidia GPU device on the vm since the GPU edition deployment requires:

standalone:
  ...
  deploy:
    resources:
      reservations:
        devices:
          - driver: nvidia
            capabilities: ["gpu"]
            device_ids: ["0"]

"show collection failed: load segment failed, OOM if load, maxSegmentSize = 2732.4219522476196 MB, memUsage = 12341.8046875 MB, predictMemUsage = 15074.22663974762 MB, totalMem = 15525.1015625 MB thresholdFactor = 0.900000" ----------------------------- Your vm has 15525MB of RAM, now the collection already occupies 12341MB RAM, Now it want to load the next segment, the segment size is 2732MB, it could be OOM if it continue to load the next segment. So, it throws this error and stops loading the collection. The collection requires more RAM to load.

[ranjith502]

thanks for the response @yhmo

in the segment failed error:- we need more ram in the databricks or azure vm
form the databricks iam doing searching

[xiaofan-luan]

if you are using GPU index, this may means you need more GPU memory, FYI