Positron
Contact Sales
background top linesbackground top lines
background right lines

Atlas

Atlas

Transformer Inference Server

  • >4x Performance per Watt versus GPUs

  • >3x Performance per Dollar vs NVIDIA Hopper

Contact Sales
Now Available
background big renderbackground big render
01render 1
02render 2
03render 3

System Specifications

Accelerator8x Positron Archer Transformer Accelerators
Memory per Accelerator32 GB HBM Each
Total Memory
For All Accelerators256 GB Each
System memory24-Channel RDIMM DDR5, 24 x 16GB DIMMs = 384 GB Can support up to 2TB system memory.
System Storage1x 1.92TB NVMe M.2, PCIe Gen3 x4 interface (OS/boot drive)
Data Storage1x 1.92TB U.2 2.5” SSD (model storage drive) Four total 2.5" SATA/SAS hot-swappable bays
CPUDual AMD® EPYC® Genoa 9374F Processors
64 Cores total, 3.85 GHz (Base)
4.3 GHz (Max Boost)
Networking1x 10Gb/s LAN port (X550 Compatible)
Management Network1x 1Gb/s LAN port (Intel® I210-AT)
SoftwarePositron Inference Engine, Ubuntu 22.04.4 LTS (Jammy)
System Power Usage/PSU2+2 2000W Redundant Titanium Level (96%) power supplies
Expansion2x FHFL PCIe Gen5 x16 slots for GPUs, NICs, etc.
1x FHHL PCIe Gen4 x16 slot
System dimensionsHeight: 7.0in (177mm)
Width: 19.0in (482.2mm)
Length: 29.25” (743mm)
System weight100 pounds (45.4 kg)
Support24h SLA Response Time from Washington-/US-based team

Head to Head Systems Comparison


(Llama 3.1 8B with BF16 compute, no speculation or paged attention)

Positron delivers leading Performance per Dollar and Performance per Watt compared to NVIDIA

01

NVIDIA DGX H100

System Power ⚡ 5900W
060120180240300
182.00
Tokens/sec/User
Perf/Dollar: 1.00x
Perf/Watt: 1.00x
02

Positron Atlas

System Power ⚡ 2000W
060120180240300
280.00
Tokens/sec/User
Perf/Dollar: 3.08x
Perf/Watt: 4.54x

Every Transformer Runs on Positron

Supports all Transformer models
seamlessly with zero time and zero effort

Positron maps any trained HuggingFace Transformers Library model directly onto hardware for maximum performance and ease of use

Step 1
Model files

.pt

.safetensors

Hugging Face

Develop or procure a model using the HuggingFace Transformers Library

Step 2

Drag & Drop to Upload

or

Upload or link trained model file (.pt or .safetensors) to Positron Model Manager

Step 3
from openai import OpenAI
client = OpenAI(uri="api.positron.ai")

client.chat.completions
  .create(
    model="my_model"
  )

Update client applications to use Positron's OpenAI API-compliant endpoint