HackerPapers

arXiv:2503.11698cs.AR

A Comparison of the Cerebras Wafer-Scale Integration Technology with Nvidia GPU-based Systems for Artificial Intelligence

Yudhishthira Kundu, Manroop Kaur, Tripty Wig, Kriti Kumar, Pushpanjali Kumari, Vivek Puri, Manish Arora

Abstract

Cerebras' wafer-scale engine (WSE) technology merges multiple dies on a single wafer. It addresses the challenges of memory bandwidth, latency, and scalability, making it suitable for artificial intelligence. This work evaluates the WSE-3 architecture and compares it with leading GPU-based AI accelerators, notably Nvidia's H100 and B200. The work highlights the advantages of WSE-3 in performance per watt and memory scalability and provides insights into the challenges in manufacturing, thermal management, and reliability. The results suggest that wafer-scale integration can surpass conventional architectures in several metrics, though work is required to address cost-effectiveness and long-term viability.

arXiv comments
11 pages
Journal reference
DOI
Subjects
cs.ARHardware Architecture

On Hacker News

Linked by 3 identifiable commenters in 4 comments across 4 threads, from May 2025 to Oct 2025. Found in 1 HN submission from Mar 2025 to Mar 2025. Open submission discussions.

Comments linking this paper, per year.

Discussed in comments4 threads

  1. Memory access is O(N^[1/3]) 1 comment

    • > L3 cache is not built for mass throughput in the same way that DRAM is, and so it has roughly identical mass throughput despite its much closer distance to the computation. "The von Neumann bottleneck is impeding AI computing?" (2025)…

      westurnerView on HNParent context

  2. Cerebras systems raises $1.1B Series G 1 comment

    • > What about for inference? Same techniques apply. > but did not dispute that they store the entire model on SRAM. No idea what they did or did not do for that specific test (which was about delivering 1800 tokens/sec though, not simply running qwen-3) since they didn't provide…

      sailingparrotView on HNParent context

  3. In a first, Google has released data on how much energy an AI prompt uses 1 comment

    • TPUs are more efficient at LLMs because they do more TOPS/KWh. "OpenTPU: Open-Source Reimplementation of Google Tensor Processing Unit (TPU)" (2025) https://news.ycombinator.com/item?id=44111452 "A Comparison of the Cerebras Wafer-Scale Integration Technology with Nvidia…

      westurnerView on HNParent context

  4. Cerebras achieves 2,500T/s on Llama 4 Maverick (400B) 1 comment

    • Performance per watt is better than h100 and b200, performance per watt per $ is worse than B200, and it does fp8 just fine https://arxiv.org/pdf/2503.11698

      skrylView on HNParent context

Submitted to Hacker News

Connected through shared HN threads, not inferred similarity or a recommendation.