NVIDIA Datacenter GPU Solutions
Clusters, servers and workstations with NVIDIA® GPUs for AI & HPC
NVIDIA Hopper Whitepaper
PLATFORMS
NVIDIA GPU Solutions from Microway
NVIDIA Datacenter GPUs are the leading acceleration platform for HPC and AI. They offer overwhelming speedups on thousands of applications and are the computational engine for nearly every accelerated computing workload. As an NVIDIA NPN Elite partner, Microway can custom-architect a bleeding-edge NVIDIA GPU solution tailored to your application or code.
WhisperStation – RTX Professional
Ultra-Quiet workstation with NVIDIA RTX Professional series GPUs for visualization, simulation, and AI
NVIDIA GPU Servers
GPU Servers with up to 10 NVIDIA Datacenter GPUs
GPU Clusters
Custom AI & HPC clusters from 5-500 GPU nodes
NVIDIA DGX
Leadership AI deployments built upon the NVIDIA DGX™ Platform
NVIDIA H100 & NVIDIA H100 NVL GPUs
The NVIDIA H100 and NVIDIA H100 NVL Tensor Core GPUs powered by the NVIDIA Hopper™ architecture provide the utmost in GPU acceleration for your deployment and groundbreaking features:
- A dramatic leap in performance for HPC
Up to 34 TFLOPS FP64 double-precision floating-point performance (67 TFLOPS via FP64 Tensor Cores) - Unprecedented performance for AI
Speedups of up to 9X for AI training vs the previous generation and up to 30X for inference with Transformers, approaching 2 PFLOPS/4PFLOPS performance - 2nd Gen Multi Instance GPU (MIG)
Provides up to 7X the secure tenants for more, fully isolated applications - Fastest GPU memory
94GB of HBM3 operating at up to 3.9TB/s for each NVIDIA H100 NVL; 80GB of HBM3 operating at up to 3.35TB/s for NVIDIA H100 - Class-leading connectivity
PCI-E Gen 5 for up to 128GB/sec of transfer BW to DPUs/InfiniBand and 4th-generation NVIDIA NVLink® for up to 900GB/sec of GPU-GPU communication
NVIDIA L40S GPUs
NVIDIA L40S GPUs are outstanding general purpose GPU accelerators for a variety of AI & Graphics workloads
- Leading Graphics and FP32 performance
Up to 91.6 TFLOPS FP32 single-precision floating-point performance - Excellent Deep Learning throughput
362.05 TFLOPS of FP16 Tensor Core (733 TFLOPS with sparsity) AI performance; 733 TOPS Int8 performance (1,466 TOPS with Sparsity) - Large, Fast GPU memory
1.2-1.7X the AI training performance of NVIDIA A100
Why NVIDIA Datacenter GPUs?
Full NVIDIA NVLink Capability, Up to 900GB/sec
Only NVIDIA Datacenter GPUs deploy the most robust implementations of NVIDIA NVLink for the highest bandwidth data transfers. At up to 900GB/sec per GPU, your data moves freely throughout the system and nearly 7X the rate of PCI-E x16 5.0 GPUs.
Unique Instructions for AI Training, AI Inference, & HPC
Datacenter GPUs support the latest TF32, BFLOAT16, FP64 Tensor Core, Int8, and FP8 instructions that dramatically improve application performance.
Unmatched Memory Capacity, up to 94GB per GPU
Support your largest datasets with up to 94GB of GPU memory, far greater capacity than available on consumer offerings.
Full GPU Direct Capability
Only datacenter GPUs support the complete array of GPU Direct P2P, RDMA, and Storage features. These critical functions remove unnecessary copies and dramatically improve data flow.
Explosive Memory Bandwidth up to 3.9TB/s and ECC
NVIDIA Datacenter GPUs uniquely feature HBM3 GPU memory with up to 3.9TB/sec of bandwidth and full ECC protection.
Superior Monitoring & Management
Full GPU integration with the host system’s monitoring and management capabilities such as IPMI. Administrators can manage datacenter GPUs with their widely-used cluster/grid management tools.
Get a Free NVIDIA AI Enterprise License
Every NVIDIA H100 PCI-Express GPU comes with a free license for NVIDIA AI Enterprise software tools.
NVIDIA AI Enterprise simplifies the building of an AI-ready solution, accelerating AI development and deployment of production-ready generative AI, computer vision, speech AI, and more.
NVIDIA Ebook: HPC for the Age of AI & Cloud
Read how HPC is evolving in the age of AI and Cloud Computing and the NVIDIA framework for HPC
