Tuesday, August 18, 2026
Ryan Scherbarth (nvidia) joined the channel
David Ozog joined the channel
Sayan Ghosh joined the channel
Jiaqi Lou joined the channel
bsahu joined the channel
Fabian joined the channel
Francois Labonte joined the channel
Daud Arslan joined the channel
Shahbozbek Hakimov joined the channel
Tong Xu joined the channel
myoung-gyun.suh joined the channel
xi.chen joined the channel
Mahfuz joined the channel
Varun Chotalia joined the channel
pierre-louis.benard joined the channel
Matthew Fricke joined the channel
Advm joined the channel
ahmadrazarehman033 joined the channel
Howard Wang joined the channel
rashid-ahmed.kukkady joined the channel
Wednesday, August 19, 2026
A
Really interesting split between ME/NMC offload for bandwidth-heavy training and PE-issued WQEs or PE collective kernels for the sub-6 µs inference path. How do you verify that ordering, completion and failure semantics stay equivalent across those paths, especially when an HCCL or firmware update changes which path a collective takes? The tenant and job isolation work is also interesting, especially as it moves from customer-specific implementation toward standardization. What negative-test contract across VXLAN, SRv6, adaptive routing, MRC and NIC reordering proves packets cannot cross tenant boundaries during reroute or a mixed-version rollout?