Microsoft HPC, network throughput issue

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

RRB

New Member
Feb 5, 2019
7
10
3
Not sure how many Windows HPC users there are on here but having an issue and thought no better place to reach out to. My experience with HPC and IT is beginner level.

Running HPC Pack 2016 (Update 3) on Windows Server 2016, 8 nodes, Infiniband FDR (56gb/sec) on application network. Running MPI PingPong throughput and only getting 5480 MB/sec max between nodes. Last I checked I should be getting around 6800 MB/sec for FDR.

Tried multiple FDR network cards, problem persists. Switch and Cables are fully FDR capable. Scheduler says all nodes have 56gb/sec connection and Network Direct is showing True. Drivers and firmware are all correct.

Oddly when i change the HPC network to run on Gigabit only, i am seeing a similar speed bottleneck- MPI PingPong only returns 70 MB/sec max, which is very short of the 125 MB/sec maximum.

Perhaps i am interpreting something incorrectly or maybe someone has seen a similar issue?