Testbeds and GPUs
The Networking and Testbeds project provides research, performance testing, capability testing, and analysis for the file system, network, and interconnect subsystems in support of current and future systems and environments. This work relies heavily on an adequately provisioned testbed, skilled staff, and collaborations with vendors. This project tests various hardware and software components to quantify the features, performance, reliability, security, and interoperability of the products and broader technology base. The information acquired as a result of this project will be used to help determine an integrated architecture and resultant procurements for these subsystems.
Those with access to the LLNL yellow network can read more about the 2018 systems in our April 2018 Newsletter article, "Testbeds are Cropping Up."
For information about the IPA Testbed, contact Trent d'Hooge.
Production GPU Clusters
|Zone||Nodes||Cores/ Node||Total Cores||Memory/ Node||Total Memory (GB)||GPU Architecture||Total GPUs||GPUs per compute node||GPU peak performance (TFLOP/s double precision)||GPU global memory (GB)||Switch||Bandwith to Lustre (GB/sec)||Peak TFLOPs (GPUs)||Peak TFLOPS (CPUs+GPUs)|
|Max||SCF||324||16||5,184||256||82,944||NVIDIA Kepler K20X||40||2||78,336||IB QDR||52.4||160.2|
|Pascal||CZ||164||36||5,904||256||41,984||NVIDIA Tesla P100 (Pascal)||326||2||5.00||16||Omni-Path||1 727.8||1 926.1|
|RZHasGPU||RZ||20||16||320||128||2,560||NVIDIA Tesla K80||64||4||60.00||768||IB QDR||59.8||68.0|
|Surface||CZ||162||16||2,592||256||41,500||NVIDIA Tesla (mixed)||316||2||452.00||3,792||IB QDR||451.9||505.8|