Supply, Installation and Commissioning of Artificial Intelligence Lab Solution for the use of Computer Science Department of the University

Amount

Supply, Installation & Commissioning of Artificial Intelligence Lab Solution for the use of Computer Science Department of the University

GPU Accelerated AI Computing and Developer Workstation Processor:Single AMD EPYC 7742 64 core, 2.25GHz CPU or better System Memory:512 GB DDR4 GPU: 4 x NVIDIA A100 NVlink, 40GB Memory per GPU Performance:2.5 PetaFLOPS AI, 5 petaOPS INT8 CUDA Cores:Minimum5000 per GPU Tensor Cores: Minimum400 per GPU Power Requirements:1.5KW or less Storage: 1X 1.92 TB NVMe Drive and Internal Storage 1x7.68TB NVMe drive System Network: Dual 10 GbE, Single Port 1 GbE GPU communications protocol:NVLink, 600 gigabytes per second (GB/s) Bidirectional bandwidth OS Support: Ubuntu Linux USB Port:3 Noise level: < 40 db Display: 4X Mini Display port, 4GB GPU Memory Cooling:Workstation will be liquid cooled. No additional cooling support will be provided for the system. Software with Support (Directly from OEM with updates& upgrades): NVIDIA NGC with 3 years support, CUDA tuned Neural Network (cu DNN) Primitives Tensor RT Inference Engine Deep Stream SDK Video Analytics CUDA tuned BLAS CUDA tuned Sparse Matrix Operations (cu SPARSE) Multi-GPU Communications (NCCL), Kubernetes Tensor Flow, Caffe ,Py Torch, Theano,Keras, caffe2, CNTK software containers Warranty & Support: 3 Years, Standard Form factor:Workstation

ML & DL Benchmarking Time to Train a BERT model on Wikipedia DataSet: Not more than 130 minutes. Time to Train a SSD model on COCO DataSet: Not more than 30 minutes. HPL Benchmarking: FP64 45TF HPL Sustained Performance.