Supercharge your next cluster with NVIDIA A100, Tesla V100 or T4 GPUs
Microway NVIDIA GPU Clusters
Microway’s fully integrated NVIDIA GPU clusters deliver supercomputing & AI performance at a lower power, lower cost, and using many fewer systems than CPU-only equivalents. These clusters are powered by NVIDIA A100, Tesla V100 “Volta” or T4 GPUs. NVIDIA datacenter GPUs scale to solve the world’s most important computing challenges more quickly and accurately.
Successfully deployed in demanding applications at research institutes, universities, and enterprises, NVIDIA GPUs power the most powerful supercomputers worldwide.
Installed Software
A Microway NVIDIA GPU cluster comes installed, integrated, and tested with:
- Linux distribution of your choice, including Red Hat, CentOS, Ubuntu, Debian, openSUSE or Gentoo
- NVIDIA HPC SDK and CUDA® SDK
- Bright Cluster Manager, OpenHPC, or Microway Cluster Management Software (MCMS™) integrated with optional MPI Link-Checker™ Fabric Validation Suite
- Optional User-level application and library installations
What Makes a Microway Cluster Different?
Expert Guidance, Expert Design
Share the details of your application or code. Microway experts will help you evaluate hardware platforms for your application. Then, they’ll help design a custom configuration tuned to your specific needs and budget.
Intensive Burn-in Testing
Every Microway cluster receives up to 1 week of burn-in testing. This includes GPU stress tests designed to identify hardware faults and “infant mortality.” Your cluster is qualified at our facility, not yours. So you can get to work faster.
Complete Software Integration
Our team integrates all the drivers, packages, and SDKs that enable you to start working from day 1. Your cluster is also delivered with GPU-aware cluster management software and schedulers tested on real accelerated HPC/AI jobs.
Our Experience in GPU Systems
We’ve been delivering NVIDIA GPU Clusters for longer than NVIDIA datacenter GPUs have existed. Microway has hundreds of satisfied customers & thousands of GPUs in the field, and we’ll apply that expertise to your successful deployment.
Sample Microway NVIDIA GPU Cluster Specifications
Density with NVIDIA A100
Two CPUs and Four NVIDIA NVLink™ GPUs with 2U Navion compute nodes
| GPUs per Node | (4) NVIDIA A100 with NVLink 3.0 |
|---|---|
| Sample Cluster Size | One fully-integrated 42U rackmount cabinet with 20 Nodes (80 GPUs) |
| Base Platform | Navion 2U NVIDIA A100 GPU Server with NVLink Dense configuration with the latest NVIDIA A100 GPUs and NVLink 3.0 interconnect |
| System Memory per Node | 4 TB DDR4 |
| Total GPU Memory per Node |
160GB HBM2 |
| Head Node | Dual AMD EPYC Server (1U – 4U) with up to 4 TB memory Optional NVIDIA Quadro™ Graphics |
| Storage | Head Node: up to 648 TB Compute Nodes: up to 8 TB Optional Storage Servers or Parallel HPC Storage System |
| Ethernet Network | Dual 10 Gigabit Ethernet built-in Optional 100Gb Ethernet |
| Interconnect (optional) | ConnectX-6 200Gb HDR or ConnectX-5 100Gb EDR InfiniBand Fabric |
| Cabinet | 42U APC NetShelter Cabinet (extra-depth model required due to chassis depth) |
| Green HPC Features | High-efficiency (80PLUS Platinum-Level) power supplies Software/firmware to reduce power consumption on idle cores Optional liquid-cooled rack doors (for thermally-neutral HPC) |
Lowest Cost
One CPU and Two GPUs with 1U NumberSmasher compute nodes
| GPUs per Node | (2) Tesla V100 |
|---|---|
| Sample Cluster Size | One fully-integrated 42U rackmount cabinet with 32 Nodes (64 GPUs) |
| Base Platform | 1U Rackmount Server Lowest cost platform for NVIDIA Datacenter GPUs |
| System Memory per Node | 1.50 TB DDR4 |
| Total GPU Memory per Node |
32GB or 64GB (Tesla V100) |
| Head Node | Dual Xeon Server (1U – 4U) with up to 3 TB memory Optional NVIDIA Quadro Graphics/GRID Remote Visualization |
| Storage | Head Node: up to 648 TB Compute Nodes: up to 54 TB Optional Storage Servers or Parallel HPC Storage System |
| Network | Dual Gigabit Ethernet built-in Optional 10Gb or 100Gb Ethernet |
| Interconnect (optional) | ConnectX-5 100Gb EDR InfiniBand Fabric |
| Cabinet | 42U APC NetShelter Cabinet |
| Green HPC Features | High-efficiency (80PLUS Platinum-Level) power supplies Software/firmware to reduce power consumption on idle cores Optional liquid-cooled rack doors (for thermally-neutral HPC) |
Balanced & Flexible
Two CPUs and up to four mixed GPUs with 4U NumberSmasher compute nodes
| GPUs per Node | (4) Tesla V100 GPUs |
|---|---|
| Sample Cluster Size | One fully-integrated 42U rackmount cabinet with 10 Nodes (40 GPUs) |
| Base Platform | NumberSmasher 4U Tesla GPU Server Support NVIDIA Datacenter GPUs for compute, Quadro GPUs for visualization, or a mix for as appropriate for your workload. Add other PCI-E devices. |
| System Memory per Node | 2 TB DDR4 |
| Total GPU Memory per Node |
48GB or 96GB (Tesla V100) |
| Head Node | Dual Xeon Server (1U – 4U) with up to 3 TB memory Optional NVIDIA Quadro Graphics/NVIDIA GRID Remote Visualization |
| Storage | Head Node: up to 648 TB Compute Nodes: up to 20 TB Optional Storage Servers or Parallel HPC Storage System |
| Network | Dual Gigabit Ethernet built-in Optional 10Gb or 100Gb Ethernet |
| Interconnect (optional) | ConnectX-5 100Gb EDR InfiniBand Fabric |
| Cabinet | 42U APC NetShelter Cabinet |
| Green HPC Features | High-efficiency (80PLUS Platinum-Level) power supplies Software/firmware to reduce power consumption on idle cores Optional liquid-cooled rack doors (for thermally-neutral HPC) |
Max GPUs
Two CPUs and up to Ten GPUs with Navion 4U or Octoputer compute nodes
| GPUs per Node | 8 or 10 NVIDIA A100 or Tesla V100 Optional NVIDIA Quadro Graphics/NVIDIA GRID Remote Visualization in additional slot |
|---|---|
| Sample Cluster Size | One fully-integrated 42U rackmount cabinet with 9 Nodes (72 GPUs for RDMA; 90 GPUs for Density) |
| Base Platform |
Navion 4U GPU Server with NVIDIA A100 GPUs Octoputer 4U 8 GPU Server or Octoputer 4U 10-GPU Server with Single Root Complex for GPU-Direct High Density Compute/Visualization configuration with maximum number of GPUs per node |
| System Memory per Node | Up to 4 TB DDR4 |
| Total GPU Memory per Node |
400GB (NVIDIA A100) 160GB or 320GB (Tesla V100) |
| Head Node | Dual Xeon or AMD EPYC Server (1U – 4U) with up to 4 TB memory Optional NVIDIA Quadro Graphics/NVIDIA GRID Remote Visualization |
| Storage | Head Node: up to 648 TB Compute Nodes: up to 48 TB Optional Storage Servers or Parallel HPC Storage System |
| Network | Dual Gigabit Ethernet built-in Optional 10Gb or 100Gb Ethernet |
| Interconnect (optional) | ConnectX-6 200Gb HDR or ConnectX-5 100Gb EDR InfiniBand Fabric |
| Cabinet | 42U APC NetShelter Cabinet |
| Green HPC Features | High-efficiency (80PLUS Platinum/Titanium-Level) power supplies Software/firmware to reduce power consumption on idle cores Optional liquid-cooled rack doors (for thermally-neutral HPC) |
Max GPUs+NVLink
Two CPUs and 8 GPUs with Octoputer with NVLink compute nodes
| GPUs per Node | 8 Tesla V100 with NVLink 2.0 |
|---|---|
| Sample Cluster Size | One fully-integrated 42U rackmount cabinet with 9 Nodes (72 GPUs) |
| Base Platform |
Octoputer 4U Tesla 8-GPU Server with NVLink High Density Compute with NVIDIA NVLink 2.0 Interconnect |
| System Memory per Node | Up to 3 TB DDR4 |
| Total GPU Memory per Node |
160GB or 320GB (Tesla V100) |
| Head Node | Dual Xeon Server (1U – 4U) with up to 3 TB memory Optional NVIDIA Quadro Graphics/NVIDIA GRID Remote Visualization |
| Storage | Head Node: up to 648 TB Compute Nodes: up to 48 TB Optional Storage Servers or Parallel HPC Storage System |
| Network | Dual Gigabit Ethernet built-in Optional 10Gb or 100Gb Ethernet |
| Interconnect (optional) | ConnectX-5 100Gb EDR InfiniBand Fabric |
| Cabinet | 42U APC NetShelter Cabinet |
| Green HPC Features | High-efficiency (80PLUS Platinum/Titanium-Level) power supplies Software/firmware to reduce power consumption on idle cores Optional liquid-cooled rack doors (for thermally-neutral HPC) |
CPU:GPU Coherence
Two POWER9 with NVLink CPUs and 4/6 Tesla V100 GPUs with AC922 compute nodes
| GPUs per Node | (4) Tesla V100 SXM2.0 (air cooled) (6) Tesla V100 SXM2.0 (liquid cooled) |
|---|---|
| Sample Cluster Size | One fully-integrated 42U rackmount cabinet with 18 Power Systems AC922 Nodes (64 GPUs and 32 CPUs) |
| Base Platform | Power Systems AC922 with Tesla V100 with NVLink nodes World’s first CPU: Tesla GPU Coherence— POWER9 CPU and Tesla V100 GPU share same memory space Only Platform with CPU:GPU NVLink—No PCI-E data bottleneck between POWER9 CPU and Tesla GPU |
| System Memory per Node | Up to 2TB DDR4 |
| Total GPU Memory per Node |
64GB (4 GPU node, air cooled) 96GB (6 GPU node, liquid cooled) |
| Head Node | Dual POWER9 Server (1U – 2U) with up to 1TB memory |
| Storage | Head Node: up to 216 TB Compute Nodes: up to 4 TB Optional Storage Servers or Parallel HPC Storage System |
| Network | Dual Gigabit Ethernet built-in Optional 10Gb/100Gb Ethernet |
| Interconnect (optional) | ConnectX-6 200Gb HDR or ConnectX-5 100Gb EDR InfiniBand Fabric |
| Cabinet | 42U APC NetShelter Cabinet |
| Green HPC Features | High-efficiency (80PLUS Platinum-Level) power supplies Software/firmware to reduce power consumption on idle cores Optional liquid cooling of nodes Optional liquid-cooled rack doors (for thermally-neutral HPC) |
Supported for Life
Our technicians and sales staff consistently ensure that your entire experience with Microway is handled promptly, creatively, and professionally.
Telephone support is available for the lifetime of your cluster by Microway’s experienced technicians. After the initial warranty period, hardware warranties are offered on an annual basis. Out-of-warranty repairs are available on a time & materials basis.
