# Cerebras explains how to run 3T models on WSE-3T chips

Digest AI · Hardware & Compute · published 2026-10-22T00:00:00Z

Canonical: https://digestai.news/story/cerebras-explains-how-to-run-3t-models-on-wse-3t-chips

## Summary

Cerebras published a technical guide detailing how to deploy large language models on its WSE-3T wafer-scale chips. The company used the open-source Kimi K3 model, which has approximately 3 trillion parameters, as a case study. The post explains that a single WSE-3T wafer contains 44 GB of on-chip memory, which is insufficient for the terabytes of weights required by frontier models. To solve this, Cerebras splits the model across multiple wafers, creating a pipeline where each wafer handles specific layers.

The guide breaks down the memory calculations for 4-bit weight quantization and 8-bit KV caching. It notes that linear-attention layers do not require large KV caches, allowing two such layers to fit on one wafer, while dense layers require one wafer each to accommodate the cache. For Kimi K3, this configuration requires 59 wafers in total. The system uses RDMA over Ethernet to pass activations between wafers with roughly 2 microseconds of latency per hop.

Cerebras emphasizes that this pipeline design allows for high concurrency. Unlike traditional batch processing, requests stream through the pipeline continuously, with different stages handling different requests simultaneously. The company states that this approach enables multi-trillion parameter models to run at speeds of 1000-2000+ tokens per second. The post concludes by noting that the same methodology scales to even larger models by adding more wafers to the pipeline.

## Key points

- Cerebras uses 59 WSE-3T wafers to run the 3T parameter Kimi K3 model.
- The system achieves 1000-2000+ tokens per second using a pipelined architecture.
- Linear-attention layers fit two per wafer, while dense layers require one wafer each.

## Why it matters

This demonstrates a viable hardware strategy for running ultra-large open-source models at high speed, potentially lowering the barrier for deploying frontier-scale AI without massive GPU clusters.

## Sources

1. [Cerebras Deploys Trillion-Parameter AI Models](https://cerebras.ai/blog/how-we-deploy-trillion-parameter-frontier-models-on-cerebras) (cerebras.ai, 2026-10-22)

## Cite

Digest AI, "Cerebras explains how to run 3T models on WSE-3T chips", 22 October 2026, https://digestai.news/story/cerebras-explains-how-to-run-3t-models-on-wse-3t-chips

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/cerebras-explains-how-to-run-3t-models-on-wse-3t-chips.json