Borealis GPU Servers

NVIDIA H100 GPU Server
The Worlds Fastest Server for AI Research

H100

Introducing Borealis H100 Servers

Experience ultimate GPU performance with newly released NVIDIA H100 Tensor Core GPUs. Available in 4U and 8U form factors.
Start AcceleratingBorealis H100 8U Server featuring 8 NVIDIA H100 GPUs
Up to
8
Nvidia H100 GPUs
Up to
128
CPU cores & 256 threads
Up to
4096
GB of memory
Up to
320
TB of NVME SSD

Transformational AI Training

H100 features fourth-generation Tensor Cores and the Transformer Engine with FP8 precision that provides up to 9X faster training over the prior generation for mixture-of-experts (MoE) models. The combination of fourth-generation NVlink, which offers 900 gigabytes per second (GB/s) of GPU-to-GPU interconnect; NVLINK Switch System, which accelerates communication by every GPU across nodes; PCIe Gen5; and NVIDIA Magnum IO™ software delivers efficient scalability from small enterprises to massive, unified GPU clusters.

Deploying H100 GPUs at data center scale delivers outstanding performance and brings the next generation of exascale high-performance computing (HPC) and trillion-parameter AI within the reach of all researchers.
Learn More
Up to 9X Higher AI Training on Largest Models
H100 vs A100 Performance
a100 Servers

Build Your Borealis A100 Server

Hyperborea

Experience 100% GPU Utilization

GVM Server, our proprietary hypervisor, is the only hypervisor that can be used to virtualize CPU and GPU across multiple GPU manufacturers. While NVIDIA, AMD, and Intel all have their own GPU virtualization tools ours is compatible with all of them.
Learn More
Arc Compute GPU Cloud Infrastructure
Arc Stack

Preinstalled Arc Stack

We've optimized our GPU servers for training and inference. Preinstalled in all Borealis servers, the Arc Stack is always ready to go with managed updates. Spin up a virtual machine with TensorFlow, TensorBoard, PyTorch, Keras, ONNX, Weights & Biases, NumPy, SciPy, JAX, CUDA, cuDNN, and all NVIDIA Drivers already waiting for you. Never waste time on set up again.

Get Started
TensorFlow
TensorFlow + TensorBoard

Neural Networks made easy by Google + the standard in experiment recording

Learn More
PyTorch
PyTorch

Neural Networks made easy by Meta

Learn More
Keras
Keras

Neural Networks made easier by Google

Learn More
ONNX
ONNX

Inter-framework model sharing + inference made easy

Learn More
Weights & Biases
Weights & Biases

Experiment recording made easy

Learn More
NumPy
NumPy

Math made easy

Learn More
SciPy
SciPy

Traditional data science made easy

Learn More
JAX
JAX

Neural Networks made faster by Google

Learn More
Tech Specs

Technical Specifications

Borealis 2U Server

GPUs

+
-
  • Up to 4 NVIDIA® A100 SXM4 GPUs (40 GB or 80 GB)
  • NVLink and PCIe 4.0 GPU-to-GPU interconnect

Processors

+
-
AMD and Intel processors available (contact us for options)

System memory

+
-
  • Up to 4096 GB of 3200 MHz DDR4 ECC RAM in 32 DIMM slots
  • 8 channels per processor

Operating system drive

+
-
  • Up to 15.36 TB of storage via one 2.5" NVME SSD (PCIe 3.x or 4.0)

Additional storage drives

+
-
  • Up to 46.08 TB of additional storage via 3 hot-swappable 2.5" NVMe SSDs (PCIe 3.x or 4.0)
  • Designed to store large data sets and model checkpoints

Processors

+
-
Built-in networking:
  • 2 Rj45 10 Gbps (aggregate) BASE-T LAN ports
  • 1 RJ45 1 Gbps BASE-T LAN out-of-band management (IPMI) port
Up to 4 high-speed NICs for GPUDirect RDMA. Options include:
  • 200 Gbps InfiniBand NVIDIA HDR ConnectX-6 VPI Adapter Card, 2x QSFP56, PCIe 4.0 x16
  • 200 Gbps InfiniBand NVIDIA HDR ConnectX-6 VPI Adapter Card, 1x QSFP56, PCIe 4.0 x16
  • 100 Gbps InfiniBand NVIDIA HDR100 ConnectX-6 VPI Adapter Card, 1x QSFP56, PCIe 4.0 x16
  • 100 Gbps InfiniBand NVIDIA HDR100 ConnectX-6 VPI Adapter Card, 2x QSFP56, PCIe 4.0 x16
  • 200 Gbps Ethernet NVIDIA ConnectX-6 EN Adapter Card, 2x QSFP56, PCIe 4.0 x16
  • 200 Gbps Ethernet NVIDIA ConnectX-6 Dx EN Adapter Card, 1x QSFP56, PCIe 4.0 x16
  • 100 Gbps Ethernet NVIDIA ConnectX-6 EN Adapter Card, 2x QSFP56, PCIe 4.0 x16
  • 100 Gbps Ethernet NVIDIA ConnectX-6 Dx EN Adapter Card, 1x QSFP56, PCIe 4.0 x16

Power

+
-
  • Two 2200 watt 80 PLUS™ Platinum PSUs
  • 2 + 0 redundancy
AC input:
  • 220-240Vac / 11-10 amperes / 50-60 Hz
Mac output per PSU:
  • 220-240 Vac: 2200 watts

Front I/O

+
-
  • Power button
  • Reset button
  • Power LED
  • Information LED
  • Power supply failure LED
  • 2 LAN activity LEDs, one for each RJ45 10 Gbps (aggregate) BASE-T LAN port
  • Storage drive activity LED

Rear I/O

+
-
  • UID button with LED
  • VGA port
  • 2 USB 3.0 ports
  • 2 RJ45 10 Gbps (aggregate) BASE-T LAN ports
  • 1 RJ45 1 Gbps BASE-T LAN out-of-band management (IPMI) port

Accessories

+
-
  • Rackmounting kit
  • 4 configurable power cables

Dimensions & weight

+
-
  • Form factor: 2U rackmount
  • Width: 17.2 inches (437 mm)
  • Height: 3.5 inches (89 mm)
  • Depth: 32.7 inches (830.3 mm)
  • Server weight: 78.5 lbs (35.6 kg)
  • Rackmounting kit weight: 5 lbs (2.3 kg)
  • Total weight with packaging: 88.5 lbs (40.1 kg)
Borealis 4U Server

GPUs

+
-
  • 8 NVIDIA® A100 SXM4 GPUs (40 GB or 80 GB)
  • NVLINK™ and NVIDIA® NVSwitch™ fabric

Processors

+
-
AMD and Intel processors available (contact us for options)

System memory

+
-
  • Up to 4096 GB of 3200 MHz DDR4 ECC RAM in 32 DIMM slots
  • 8 channels per processor

Operating system drive

+
-
  • Up to 15.36 TB of storage via one 2.5" NVME SSD (PCIe 3.x or 4.0)

Additional storage drives

+
-
  • Up to 138.24 TB of additional storage via 9 NVMe SSDs (PCIe 3.x or 4.0)
  • Designed to store large data sets and model checkpoints

Networking

+
-
Built-in networking:
  • 2 RJ45 10 Gbps BASE-T LAN ports
  • 1 RJ45 1 Gbps BASE-T LAN out-of-band management (IPMI) port
Up to 8 high-speed NICs for GPUDirect RDMA. Options include:
  • 200 Gbps InfiniBand NVIDIA HDR ConnectX-6 VPI Adapter Card, 2x QSFP56, PCIe 4.0 x16
  • 200 Gbps InfiniBand NVIDIA HDR ConnectX-6 VPI Adapter Card, 1x QSFP56, PCIe 4.0 x16
  • 100 Gbps InfiniBand NVIDIA HDR100 ConnectX-6 VPI Adapter Card, 1x QSFP56, PCIe 4.0 x16
  • 100 Gbps InfiniBand NVIDIA HDR100 ConnectX-6 VPI Adapter Card, 2x QSFP56, PCIe 4.0 x16
  • 200 Gbps Ethernet NVIDIA ConnectX-6 EN Adapter Card, 2x QSFP56, PCIe 4.0 x16
  • 200 Gbps Ethernet NVIDIA ConnectX-6 Dx EN Adapter Card, 1x QSFP56, PCIe 4.0 x16
  • 100 Gbps Ethernet NVIDIA ConnectX-6 EN Adapter Card, 2x QSFP56, PCIe 4.0 x16
  • 100 Gbps Ethernet NVIDIA ConnectX-6 Dx EN Adapter Card, 1x QSFP56, PCIe 4.0 x16

Power

+
-
  • 4 hot-swappable 3000 watt 80 PLUS™ Titanium PSUs
  • 2 + 2 redundancy
AC input:
  • 200-207 Vac / 16-15.7 amperes / 50-60 Hz
  • 207.1-240 Vac / 16-14.5 amperes / 50-60 Hz
Mac output per PSU:
  • 200-207 Vac: 2880 watts
  • 207.1–240 Vac: 3000 watts

Front I/O

+
-
  • Power button
  • Reset button
  • Power LED
  • UID button with LED
  • Information LED
  • 1 RJ45 1 Gbps BASE-T LAN out-of-band management (IPMI) port
  • LAN activity LED for out-of-band management (IPMI) port

Rear I/O

+
-
  • 2 RJ45 10 Gbps BASE-T LAN ports

Accessories

+
-
  • Rackmounting kit
  • 4 C19 TO C20 power cables (4 foot)

Dimensions & weight

+
-
  • Form factor: 4U rackmount
  • Width: 17.6 inches (446 mm)
  • Height: 6.9 inches (174 mm)
  • Depth: 35.4 inches (900 mm)
  • Server weight: 166 lbs (75.3 kg)
  • Rackmounting kit weight: 5 lbs (2.3 kg)
  • Total weight with packaging: 225 lbs (102.1 kg)