Cerebras Wafer Scale (WSE)
Vendor: Cerebras (IPO completed)
Category: ASIC Dedicated Accelerator
Architecture: CS-3
Introduction
The Cerebras Wafer-Scale Engine (WSE-3) is the largest AI processor in existence, featuring 4 trillion transistors and 900,000 AI cores on a single chip. Focused on large model training, it delivers performance comparable to traditional clusters without requiring distributed programming. On May 14, 2026, Cerebras went public on NASDAQ with a first-day valuation of approximately $560 billion.
Specifications
| Model | Compute | Memory | Interface | TDP | Process |
|---|---|---|---|---|---|
| WSE-3 | 4 trillion transistors, 900K cores | 44GB on-chip SRAM | CS-2 Fabric | 20000+ | 5nm |
| WSE-2 | 2.6 trillion transistors, 850K cores | 40GB on-chip SRAM | CS-2 Fabric | 20000+ | 7nm |
Official Website
Driver Downloads
Linux
Related Documentation
OS Support
| Windows | Linux | macOS | Android |
|---|---|---|---|
| ❌ | ✅ (Cerebras Cloud) | ❌ | ❌ |
Version History
| Version | Release Date | Description |
|---|---|---|
| Cerebras Cloud 2.0 | 2024 | WSE-3 launched + API access |
Performance Benchmarks
| Model | Task | Performance Metric |
|---|---|---|
| WSE-3 | Llama 3 70B Inference | ~1800 tok/s (pipeline) |
| WSE-3 | Ultra-long context inference | Supports 128K context |
| WSE-2 | GPT-J 6B Training | Completed in minutes |
Pricing
| Model | Reference Price | Notes |
|---|---|---|
| WSE-3 | Cloud API pay-per-use | Cerebras Cloud / CSP |
| WSE-3 | Contact vendor | Bare-metal deployment |
Quick Installation
Cerebras Cloud (API)
# Install Cerebras SDK
pip install cerebras-cloud-sdk
# Use Cerebras Inference API
# OpenAI API-compatible interface
WSE series is primarily accessed through Cerebras Cloud API or partner cloud services. Bare chips are not sold to end users.
Code Examples
Python (Cerebras API)
from cerebras.cloud.sdk import Cerebras
# OpenAI-compatible interface
client = Cerebras(api_key="your-key")
response = client.chat.completions.create(
model="llama3.1-8b",
messages=[{"role": "user", "content": "Hello"}],
max_tokens=100
)
print(response.choices[0].message.content)
Architecture Highlights
- Wafer-Scale Engine (WSE): An entire wafer as a single chip. WSE-3 features 4 trillion transistors and 900K AI cores
- CS-3 System: 16 WSE-3 chips interconnected, supporting trillion-parameter model training
- Zero Fragmentation: Wafer-scale design eliminates traditional inter-chip interconnect bottlenecks with ultra-low latency
Model Compatibility
| Model/Framework | Support | Notes |
|---|---|---|
| Llama Series | ✅ Native | Official Cerebras deployment |
| Large Language Models | ✅ | API inference |
| Training Frameworks | ⚠️ | Via Cerebras PyTorch extensions |
| Custom Models | ⚠️ | Requires Cerebras customization |
Large-Scale Cluster Deployments
Based on global AI supercomputing cluster statistics, Cerebras WSE has accumulated over 65 chips deployed across 2 publicly disclosed clusters.
Chip Model Statistics
| Chip Model | Total Deployed | Cluster Count |
|---|---|---|
| Cerebras CS-2 | 64 | 1 |
| Cerebras CS-1 | 1 | 1 |
Notable Deployment Clusters Top 10
| # | Cluster Name | Total Chips | Chip Model | Operator |
|---|---|---|---|---|
| 1 | Condor Galaxy 1 (CG-1) | 64 | Cerebras CS-2 ×64 | G42,Cerebras Systems, United States of America |
| 2 | Lawrence Livermore NL Lassen Phase 2 | 1 | Cerebras CS-1 ×1 | US Department of Energy, United States of America |
Related Products
If you're evaluating alternatives, the following products may also fit your scenario:
- Groq LPU v1 — Groq (LPU Language Processor)
- Google Cloud TPU — Google (TPU Tensor Processor)
- SambaNova SN40L — SambaNova (ASIC Dedicated Accelerator)
- NVIDIA GPU / CUDA — NVIDIA (GPU Graphics Processor)
- AWS Trainium / Inferentia — Amazon AWS (ASIC Dedicated Accelerator)
- Etched Sohu ASIC — Etched (ASIC Dedicated Accelerator)
- Cambricon Siyuan MLU — Cambricon (ASIC Dedicated Accelerator)