Search results

Why is a mechanism like PXN in NCCL not used in low-latency kernel? It should be able to achieve even lower latency.

Status: Open.

#31 In deepseek-ai/DeepEP;

· weishenmezhememan opened on Feb 28, 2025

When testing test_internode.py, what is the effect of setting NVSHMEM_DISABLE_P2P=1? Will NVLink be disabled?

Status: Open.

#30 In deepseek-ai/DeepEP;

· MoringKing opened on Feb 28, 2025

test_low_latency.py bandwidth extremely low

Status: Open.

#28 In deepseek-ai/DeepEP;

· Qizhi697 opened on Feb 28, 2025

why your guess? "st.global.L1::no_allocate" that doesn't mean L2 won't be allocated cache.

Status: Open.

#27 In deepseek-ai/DeepEP;

· zzJingHong opened on Feb 26, 2025

Why "without costing any GPU SMs from the computation part"

Status: Open.

#26 In deepseek-ai/DeepEP;

· zhang662817 opened on Feb 26, 2025

"gdrdrv-dkms:amd64 is not configured" in container when installing NVSHMEM

Status: Open.

#24 In deepseek-ai/DeepEP;

· JoeyYoung opened on Feb 26, 2025

why enabling adaptive routing for normal internode kernels may lead to deadlocks or data corruption issues？

Status: Open.

#20 In deepseek-ai/DeepEP;

· VincentHITlemon opened on Feb 26, 2025

Test test_low_latency.py failed on H20 ROCE

Status: Open.

#17 In deepseek-ai/DeepEP;

· HanHan009527 opened on Feb 25, 2025

Test test_low_latency.py failed

Status: Open.

#15 In deepseek-ai/DeepEP;

· Nekofish-L opened on Feb 25, 2025

Test fail use 2node on tests/test_internode.py

Status: Open.

#13 In deepseek-ai/DeepEP;

· xesdiny opened on Feb 25, 2025

Benifits of FP8 cast and AR

Status: Open.

#12 In deepseek-ai/DeepEP;

· xiaoyao9933 opened on Feb 25, 2025

Installing issue with gdrdrv

Status: Open.

#11 In deepseek-ai/DeepEP;

· kevinhuangxf opened on Feb 25, 2025