AI Chips · Product page

Cerebras CS-2 (WSE-2): The Complete Guide

The generation that proved the wafer-scale bet could be repeated — more than doubling every headline spec, then going on to power a national-scale AI supercomputer. Simple to expert, every number links to its source.

Stuck at any point? Ask an AI about this page →
Ask an AI about this chip:CAGPX
Cerebras CS-2 (WSE-2) 10 sections · 2 levels 5 linked sources Checked September 2026
How to read this page. Each section starts Simple, then goes to a Deep dive: stop wherever you have what you need. The small numbers are sources: click one to open the original document. Where the maker has not published a figure -- a price, a die size, a factory address -- this page says so rather than estimate it.

1.At a glance

SimpleStart here

WSE-2, launched 20 April 2021, more than doubled the first Wafer-Scale Engine’s core count — 2.6 trillion transistors and 850,000 cores on the same dinner-plate-sized wafer 1. Two years later it became the chip behind Condor Galaxy 1, a 4-exaFLOP AI supercomputer built with G42 of Abu Dhabi 2.

Deep diveThe technical detail

Cerebras’s CEO Andrew Feldman said the generational jump moved “about 2.3X on every performance dimension” 3. The chip moved to a smaller manufacturing process too — TSMC 7 nm, down from 16 nm on the first Wafer-Scale Engine (see that page on this site) 1 4.

2.Launch and history

SimpleStart here

WSE-2 is the generation that proved Cerebras’s wafer-scale approach wasn’t a one-off: the same fundamental design — a whole silicon wafer as one chip — scaled to more than double the cores and memory two years after the first version shipped 1.

Deep diveThe technical detail

Its biggest deployment came from outside Cerebras’s own launch materials: Condor Galaxy 1, announced 20 July 2023 with G42, combined 64 CS-2 systems into what Cerebras called “a 4 exaFLOPS, 54 million core, 64-node AI supercomputer” 2.

3.What’s inside it

SimpleStart here

850,000 AI-optimized cores — more than double the first generation’s 400,000 — linked by an upgraded on-wafer fabric 1 3.

Deep diveThe technical detail

Cerebras built in spare capacity for manufacturing defects from the start: EE Times reported the chip includes “1% redundant cores purely for this eventuality,” with Feldman saying wafer yields were “much, much higher than much smaller parts like the GPU” because defects are spread randomly across the wafer rather than clustered 3. Feldman did concede one harder case: a defect pattern that knocks out several adjacent cores at once, which he called “extraordinarily rare” but did not claim was solved by the redundant cores alone 3.

4.Full spec table

SimpleStart here
SpecCerebras WSE-2
Transistors2.6 trillion 1
Cores850,000 AI-optimized cores 1
On-chip memory40 GB SRAM 3
Memory bandwidth20 petabytes/s 3
Fabric bandwidth220 petabits/s 3
Chip sizeMore than 46,225 mm² 1
Process nodeTSMC 7 nm 1
Deep diveThe technical detail

Against the first Wafer-Scale Engine (see that page on this site): memory more than doubled, 18 GB to 40 GB, and memory bandwidth grew from 9 to 20 petabytes/s — while fabric bandwidth grew from 100 petabits/s to 220 petabits/s 3 4.

5.Where it’s made

SimpleStart here

Designed by Cerebras; the wafer made by TSMC, now on a 7 nm process, up from 16 nm on the first generation 1 4.

Deep diveThe technical detail

Cerebras’s approach to manufacturing defects was explained publicly at this generation’s launch for the first time in the sources checked for this site: rather than discarding a wafer with a flaw, its on-chip network routes around bad cores, using the built-in 1% redundancy 3.

6.Which systems use it

SimpleStart here

Sold as the CS-2 system. Its largest confirmed deployment: 64 CS-2 systems in Condor Galaxy 1, built with G42 of Abu Dhabi and announced 20 July 2023 2.

Deep diveThe technical detail

Condor Galaxy 1 launched in two phases — 32 CS-2 systems, then a second phase bringing the total to 64 — reaching a combined 4 exaFLOPS of AI training compute 2. It was Cerebras’s largest publicly named system built on this generation of chip.

7.Official pricing

SimpleStart here

Not published by the company. Neither the April 2021 launch materials nor the Condor Galaxy 1 announcement give a CS-2 system price.

Deep diveThe technical detail

This matches every other Cerebras generation checked on this site: the company has never published a system list price, for CS-1, CS-2, CS-3 or CS-4 (see those pages on this site).

8.Real-world performance

SimpleStart here

Cerebras’s own generational claim: “more than double the amount of processor cores” versus the first generation, and CEO Andrew Feldman’s statement that the chip moved “about 2.3X on every performance dimension” 3.

Deep diveThe technical detail

This is Cerebras’s own comparison against its own prior generation, not an independently run benchmark against a competing GPU. No named third-party test of WSE-2 against a specific NVIDIA or AMD chip was found in the sources checked for this page.

9.What came before, what came next

SimpleStart here
Came before: WSE-1 (CS-1), covered on its own page on this site. Came after: WSE-3 (CS-3), covered on its own page on this site.
Deep diveThe technical detail

WSE-2 shipped about 20 months after the first Wafer-Scale Engine’s August 2019 unveiling 1 4. WSE-3 followed nearly three years later, in March 2024 — a longer gap than the one between the first two generations 5.

10.Hidden in plain sight

SimpleStart here

Condor Galaxy 1’s “54 million core” figure is simple multiplication, not a new design: 64 CS-2 systems at 850,000 cores each comes to 54.4 million cores, exactly matching Cerebras’s own published total 2 1.

Deep diveThe technical detail

EE Times reported Feldman saying Cerebras’s strategy “began with the assumption we would face flaws” in the silicon — in other words, the same fault-tolerant philosophy behind the first Wafer-Scale Engine, made explicit and quantified (1% redundant cores) for the first time at this generation’s launch 3.

11.Sources

5 sources, checked September 2026. Where NVIDIA or another maker has not published a figure, this page says so rather than estimate it.

  1. Cerebras: Second-generation Wafer Scale Engine, 2.6 trillion transistors (20 Apr 2021)Official
  2. Cerebras Blog: Introducing Condor Galaxy 1: a 4 exaFLOP supercomputer (20 Jul 2023)Official
  3. EE Times (press): Cerebras Crams More Compute Into Second-Gen ‘Dinner Plate Sized’ ChipPress
  4. Cerebras: Cerebras Systems Unveils the Industry’s First Trillion Transistor Chip (19 Aug 2019)Official
  5. Cerebras: Cerebras announces third-generation Wafer Scale Engine (13 Mar 2024)Official

Ask an AI about this page

Opens your assistant with this page as the source, and a question rather than a summary. It will ask what you are building before it answers.

ChatGPTClaudeGeminiPerplexityGrok

Nothing is sent from here. The link carries only this page’s title and address.