Skip to content

Commit 53e92ef

Browse files
rongbingzhoubwbarrett
authored andcommitted
aws: Default to sendrecv protocol on g7e.8xlarge
Switch to the sendrecv protocol on g7e.8xlarge. g7e.8xlarge does not support GPUDirect RDMA. The RDMA protocol requires FI_HMEM support without allowing the provider to use CUDA, which is impossible without GPUDirect RDMA support. Signed-off-by: Rongbing Zhou <rongbiz@amazon.com>
1 parent 0b8bc0f commit 53e92ef

File tree

1 file changed

+1
-1
lines changed

1 file changed

+1
-1
lines changed

src/platform-aws.cpp

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -207,7 +207,7 @@ const PlatformAWS::ec2_platform_data PlatformAWS::platform_data_map[] = {
207207
.default_dup_conns = 0,
208208
.latency = 35.0,
209209
.gdr_required = false,
210-
.default_protocol = PROTOCOL::RDMA,
210+
.default_protocol = PROTOCOL::SENDRECV,
211211
.env = {
212212
{ "NCCL_BUFFSIZE", "8388608" },
213213
{ "NCCL_P2P_NET_CHUNKSIZE", "524288" },

0 commit comments

Comments
 (0)