Since August 2026 Cerebras also ships a Turbo bin of this same wafer, the WSE-3 Turbo, in the CS-4 rack. It is not a new processor: same four trillion transistors, same 46,225 mm2 of TSMC 5 nm silicon, same 900,000 cores and same 44 GB of on-chip SRAM, but with roughly twice as much power delivered to it, which Cerebras rates at 250 PFLOPS and 43.2 PB/s of memory bandwidth per wafer against 125 PFLOPS and 21 PB/s here. Both figures are the vendor's sparse FP16 numbers. There is no WSE-4.
Cerebras logo

Cerebras WSE-3

Type: Wafer-scaleArchitecture: Wafer Scale Engine 3Released: 2024Process: TSMC 5nmSpec confidence: Official

Overview

The Cerebras Wafer Scale Engine 3 is a single chip the size of a wafer: 46,225 mm2 of TSMC 5nm silicon carrying 4 trillion transistors and 900,000 cores optimised for sparse linear algebra, with 44 GB of on-chip SRAM at 21 PB/s and 214 Pb/s of on-wafer fabric bandwidth. It has no external memory at all — the 44 GB of SRAM is the entire memory system, which is why the usual DRAM capacity and bandwidth figures are blank rather than missing. Cerebras publishes a single compute figure of 125 petaFLOPS, and a footnote on its own datasheet states plainly that FLOPS are shown as sparse. That sparsity is unstructured rather than the 2:4 structured kind, so no dense figure can be derived from it by halving, and none is published; this entry therefore carries no precision figures at all rather than a number that would be compared against dense figures elsewhere on this site.

Performance

Peak theoretical throughput by precision type

PrecisionPeak
FP64
No verified data available
FP32
No verified data available
TF32
No verified data available
BF16
No verified data available
FP16
No verified data available
FP8
No verified data available
FP6
No verified data available
FP4
No verified data available
INT8
No verified data available

Flopper holds no verified throughput figures for this part yet. The rows stay so the gaps are visible.

Specifications

Architecture

Wafer Scale Engine 3

Form Factor

Wafer-Scale

Launch Year

2024

Process Node

TSMC 5nm

Memory

No verified data available On-die SRAM

Bandwidth

No verified data available

TDP

No verified data available

Max power

No verified data available

Transistors

4000.0 billion

Sparse linear algebra core

900,000

Spec Confidence

Official

Full Specifications

Compute Engine
Sparse linear algebra core 900,000
Chip Design
Transistors 4000.0 billion
Die Size 46225 mm²
Process Node TSMC 5nm
Memory
Memory No verified data available
Memory Type On-die SRAM
Bandwidth No verified data available
Interface Width No verified data available
On-Die SRAM 44 GB
Power & Thermal
TDP No verified data available
Max power No verified data available
General
Form Factor Wafer-Scale
Architecture Wafer Scale Engine 3
Process Node TSMC 5nm
Launch Year 2024

Datasheet & Resources

Data Provenance

Every figure traced to a source

Primary Source

Publisher
Cerebras
Published
No verified data available

Data Quality

Spec confidence
Official
Core precisions with figures
0 of 9
Normalization
All values in TFLOPS

Compare cloud providers offering on-demand GPU instances for AI training, inference, and HPC workloads.

Browse GPU Cloud Providers

Systems Using This GPU

Pre-configured systems featuring the Cerebras WSE-3

SystemGPU CountPeak PerformanceTotal Power
Cerebras CS-4
CS-4 · rack · 2026
3x WSE-3 No verified data available No verified data available View System
Cerebras CS-3
CS-3 · rack · 2024
1x WSE-3 No verified data available 27.0 kW View System
Cerebras CS-3 Inference Cluster (16-node)
CS-3 · cluster · 2024
16x WSE-3 No verified data available 475.0 kW View System
Cerebras CS-3 Inference Cluster (32-node)
CS-3 · cluster · 2024
32x WSE-3 No verified data available 945.0 kW View System
Cerebras CS-3 Inference Cluster (64-node)
CS-3 · cluster · 2024
64x WSE-3 No verified data available 1860.0 kW View System
Cerebras CS-3 Inference Cluster (8-node)
CS-3 · cluster · 2024
8x WSE-3 No verified data available 240.0 kW View System
Cerebras CS-3 Inference Cluster (88-node)
CS-3 · cluster · 2024
88x WSE-3 No verified data available 2540.0 kW View System
Cerebras CS-3 Rack
CS-3 · rack · 2024
2x WSE-3 No verified data available 54.0 kW View System

Frequently Asked Questions

How many TFLOPS does the Cerebras WSE-3 have?

Flopper does not currently have any verified throughput figures for the Cerebras WSE-3.

What is the power consumption of the Cerebras WSE-3?

Flopper does not currently have a verified TDP (Thermal Design Power) figure for the Cerebras WSE-3.

How much memory does the Cerebras WSE-3 have?

Flopper does not currently have verified memory capacity or bandwidth figures for the Cerebras WSE-3.

What architecture is the Cerebras WSE-3 based on?

The Cerebras WSE-3 is based on the Wafer Scale Engine 3 architecture, launched in 2024.

Stay Updated on GPU Releases

Get notified when new GPUs are added or specifications are updated.

Loading verification...

No spam, unsubscribe anytime.

Back to GPUs

© 2026 Flopper.io - Compare the hardware powering AI