Welcome, Please Sign In

Get in touch with your rep, view past orders, save configurations and more. Don't have an account? Create one in seconds below.

login

Atlas AI Cluster

Solutions for your AI Workload

Get to Result Faster with the Atlas AI Cluster

Looking for a fast, powerful system designed from the ground up to optimize large AI datasets? Consider the Silicon Mechanics Atlas AI Cluster™.

This custom-engineered, Linux-based cluster includes best-of-breed technology (including the NVIDIA® HGX™ A100 and AMD EPYC™), configured for fast deployment and powerful results. Best of all? The Atlas AI Cluster can give you GPU-accelerated computing that scales to any size and still provides stronger ROI compared to the equally powerful but more costly NVIDIA® DGX™.

Get More Information

Ideal Use Cases

  • Natural Language Processing
  • Machine Learning
  • Deep Learning
  • Predictive Analytics
  • Cybersecurity
  • Business Intelligence
  • Virtual Assistants
  • Robotics

Relevant Industries

Aerospace/Defense
Healthcare/Life Sciences
Design & Manufacturing (inc. Automotive)
Financial Services
Retail
Supply Chain
Government

Pre-Configured for Your AI Workloads

Quick setup

Atlas AI Clusters are pre-configured for AI, reducing time to result even on the largest workloads.

Low TCO

Save in the long run with the low total cost of ownership compared to traditional supercomputers.

Grows with you

Each cluster is designed to give you seamless, linear scaling as your computing needs grow.

Superior performance

Optional 3rd Gen AMD EPYC™, the world’s highest-performing x86 server CPUi.

No platform lock-in

Lower initial cost and no platform lock-in compared to the NVIDIA® DGX A100.

Blazing fast memory

Includes NVIDIA A100 GPUs, providing world’s fastest memory bandwidth (over 2 TB/s) to run the largest models and datasetsii and GPU partitioning.

Resources

Learn More About the Atlas Cluster

Silicon Mechanic engineers were able to achieve so much in a single AI platform by using a building block approach, where computing, storage, and networking components were optimized for specific AI needs.

However, because of the way they are integrated, end users can still request changes for today and scale as needed tomorrow.

How a Building Block Approach Enables AI Optimization

Download this white paper from InsideHPC and get a deep dive on the various modules inside the Silicon Mechanics Atlas AI Cluster and see how it can help you achieve your AI goals.

Download White Paper

GPU-Accelerated Compute Node Components

PROCESSOR

2x AMD EPYC 7742 64-core CPUs (128 cores total)

NETWORK INTERFACE

NVIDIA Mellanox® Spectrum® SN2000 GbE HDR switches

STORAGE

2x 1.92TB M.2 NVMe storage

POWER

4x 3000W power supplies (3+1)

OPTION 1
  • 4U server with 8x NVIDIA HGX™ A100 GPUs and 320GB GPU memory
  • 1TB DDR4 system memory
  • 15TB NVMe storage
OPTION 2
  • 4U server with 8x NVIDIA HGX™ A100 GPUs and 640GB GPU memory
  • 2TB DDR4 system memory
  • 30TB NVMe storage

Benchmark Your Performance

Not sure how well the combination of AMD EPYC and NVIDIA HGX A100 will meet your specific workload needs?

The Silicon Mechanics engineering team has set up remote compute modules for secure log in so that you can try both a 4x A100 GPUs (160GB memory) version and an 8x A100 GPUs (320GB memory) version.

Request a Test Drive

Storage Components

Includes 8 Weka.IO storage nodes + 1 additional Weka.IO node per additional GPU node. Additional optional storage nodes available as needed.

Weka All-Flash Tier
  • 4x High-density, all NVMe Flash storage arrays: 1U dual-node storage server with 36x E1.S (EDSFF) NVMe SSDs
  • Base configuration starts at 200TB NVMe SSD capacity
  • Base configuration scales to 360TB by adding drives, or >360TB by adding storage nodes
  • 2x Mellanox 25GbE adapter ConnectX-6 Dx (2x SFP28)
  • 2x Mellanox 200Gb/s HDR InfiniBand adapter ConnectX-6 VPI (1x QSFP56)
  • 1300W 1+1 Redundant power supplies - 80 PLUS Platinum
S3 Object Storage Capacity Tier
  • Node count and drive density based on capacity requirements

Supported Software & AI Frameworks

Includes the Silicon Mechanics AI Stack, Silicon Mechanics’ Scientific Computing Stack, and support for popular frameworks

Get a Quote
We build each of our systems to “zero defect” standards in our U.S.-based manufacturing facilities. Then we hand-inspect every order, testing them to ensure they are 100% operational and optimized to support rapid deployment. What kind of testing? We use a suite of automated tools that simulate real world operations (including convenience and update utilities) and stress tests to validate system functionality. And, because we’re ISO 9001 certified and we’ve been in your shoes, we document the heck out of it so you can easily replicate hardware settings, firmware updates, and software versioning. Standardized installs of new systems mean efficient system deployment. We like that. We think you will too.
We offer a comprehensive 3-year warranty standard, with every system purchased. But you can add extended or custom warranties if your situation calls for it. We know you are literally curing cancer, finding water on Mars, designing airplanes that will carry thousands of people, and ensuring we have live, on-screen data about opposing players during a critical game. We don’t want you to stop working any more than you do.
We offer customer support at different levels, each based on a detailed service level agreement (SLA) that fits your needs. And, with each one, you get the added benefit of knowing that the people who take your call are in-house Silicon Mechanics’ staff whose sole focus is customer support. When we say “Expert Included,” we mean it.

Expert Included

Our experts can survey the complex technology landscape and develop a solution that fits your unique needs. Read more about our Expert Included approach, or contact us now to get started.

i MLN-016: Results as of 01/28/2021 using SPECrate®2017_int_base. The AMD EPYC 7763 measured estimated score of 798 is higher than the current highest 2P server with an AMD EPYC 7H12 and a score of 717, https://spec.org/cpu2017/results/res2020q2/cpu2017-20200525-22554.pdf. OEM published score(s) for EPYC may vary. SPEC®, SPECrate® and SPEC CPU® are registered trademarks of the Standard Performance Evaluation Corporation. See www.spec.org for more information. ii https://www.nvidia.com/en-us/data-center/a100