profile
viewpoint
If you are wondering where the data of this site comes from, please visit https://api.github.com/users/cadedaniel/events. GitMemory does not store any data, but only uses NGINX to cache data for a period of time. The idea behind GitMemory is simply to give users a better reading experience.
Cade Daniel cadedaniel Amazon Web Services Palo Alto SDE at AWS Deep Learning

thomasdunn/judo 6

Java IDE for Children and Beginning Programmers

cadedaniel/zstd 1

Zstandard - Fast real-time compression algorithm

cadedaniel/amazon-sagemaker-operator-for-k8s 0

Amazon SageMaker operator for Kubernetes

cadedaniel/pyresttest 0

Python Rest Testing

issue commentNVIDIA/nccl

Implications of increasing NCCL_BUFFSIZE

If the NCCL_BUFFSIZE is not large enough for tensor communication, how does NCCL communicate the data? Is it as simple as chunking the source tensor into 4MB sections and sending them one-by-one? (e.g. memcpy from source into NCCL buffer+transmit each 4MB chunk).

anirudh2290

comment created time in 2 months