From 4d6a417f176e620e12406b55d2dc8eb0f8c1c1d1 Mon Sep 17 00:00:00 2001 From: Thomas Huber Date: Tue, 27 Oct 2026 11:16:07 -0400 Subject: [PATCH] fix: raise rccl_send_recv ctest timeout to 300s The 60s TIMEOUT was calibrated on 8 GPUs. On a 64-GPU MI350X CPX node the sample needs about as long as the other RCCL examples and is killed before ncclCommInitAll finishes. QA confirmed 300s completes. Co-authored-by: Cursor --- Libraries/RCCL/send_recv/CMakeLists.txt | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/Libraries/RCCL/send_recv/CMakeLists.txt b/Libraries/RCCL/send_recv/CMakeLists.txt index 928ca8a4e..0b445745f 100644 --- a/Libraries/RCCL/send_recv/CMakeLists.txt +++ b/Libraries/RCCL/send_recv/CMakeLists.txt @@ -47,7 +47,9 @@ find_package(rccl REQUIRED) add_executable(${example_name} main.cpp) # Make example runnable using ctest add_test(NAME ${example_name} COMMAND ${example_name}) -set_tests_properties(${example_name} PROPERTIES TIMEOUT 60) +# 60s was calibrated on <=8 GPUs (~15s). 64-GPU CPX init takes about as long +# as the other RCCL samples (~3 min); 300s is the timeout QA confirmed. +set_tests_properties(${example_name} PROPERTIES TIMEOUT 300) # Link to example library target_link_libraries(${example_name} PRIVATE roc::rccl)