Skip to content

fix: nvls all reduce correction factor #239

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.

Sign up for GitHub

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

Open
functionstackx wants to merge 1 commit into
base: master
Choose a base branch
from
Open

fix: nvls all reduce correction factor #239

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions doc/PERFORMANCE.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -136,6 +136,7 @@ And :
To obtain a bus bandwidth which should be independent of the number of ranks _n_, we apply a correction factor to the algorithm bandwidth :

* AllReduce : 2*(_n_-1)/_n_
* AllReduce (In-Network Reduction): (n-1)/(n+1)
* ReduceScatter : (_n_-1)/_n_
* AllGather : (_n_-1)/_n_
* Broadcast : 1
Expand Down
11 changes: 10 additions & 1 deletion src/all_reduce.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -36,7 +36,16 @@ void AllReduceGetBw(size_t count, int typesize, double sec, double* algBw, doubl
double baseBw = (double)(count * typesize) / 1.0E9 / sec;

*algBw = baseBw;
double factor = ((double)(2*(nranks - 1)))/((double)nranks);

const char* nccl_algo = getenv("NCCL_ALGO");

double factor;
if (nccl_algo != nullptr && (strcmp(nccl_algo, "NVLS") == 0 || strcmp(nccl_algo, "NVLSTREE") == 0)) {
factor = ((double)(nranks - 1)) / ((double)(nranks + 1));
} else {
factor = ((double)(2 * (nranks - 1))) / ((double)nranks);
}

*busBw = baseBw * factor;
}

Expand Down