AI HARDWARE

Cerebras Launches Condor Galaxy 1 AI Hardware System

By

–

20 July 2023 19h34

Condor Galaxy 1 is up and running at the Colovore data center in Santa Clara, CA. We will be unveiling an exciting new model trained on CG-1 at ICML 2023. Follow us on Twitter and sign up to our newsletter to stay up to date! https://
info.cerebras.net/subscribe

Cerebras: 20x Memory Advantage Enables 100B Models Natively

By

–

20 July 2023 19h34

Because our compute and memory are ~20x larger, even 100B models fit entirely in memory without tensor/pipeline parallel gymnastics. Our GPT implementation uses 1/20th the lines of code of Nvidia Megatron. Huge models work out of the box on Cerebras. This is why G42 picked us.

Cerebras Cloud Democratizes AI Supercomputing Access Globally

By

–

20 July 2023 19h34

CG-1 will be served through the Cerebras Cloud to both G42 and external customers. Big picture – Cerebras is now providing AI supercomputing instances – the same kind that Azure provided OpenAI. Our goal is to democratize AI access so all teams can build large scale AI models.

Cerebras CS-2 Software Stack: Single Logical Accelerator Architecture

By

–

20 July 2023 19h34

But specs are just specs. The real question is: how's the software stack? What's it like to program 64 CS-2 machines? This is where Cerebras truly shines – it's the only AI supercomputer that works as *one* logical accelerator.

Cerebras CG-1 vs Nvidia Israel-1: GPU Architecture Comparison

By

–

20 July 2023 19h34

Let's compare CG-1 with Nvidia's Israel-1. Israel-1 has 2048 H100 GPUs. But each GPU shows up as a unique device. It's your job to break apart your model and farm them out to each GPU. On Cerebras, 1 to 64 CS-2s show as one accelerator. Just implement mini-batches and train!

G42 Cloud Plans Nine AI Supercomputers with 36 ExaFLOPs

By

–

20 July 2023 19h34

G42 Cloud is the largest public cloud provider of the UAE. To expand its AI offering, we are planning not one but *nine* AI supercomputers. When complete in 2024, the full Condor Galaxy system will have 9 instances, 576 CS-2s, for a total of 36 exaFLOPs of AI compute.

36 ExaFLOPs AI Supercomputer Outperforms Google TPU Pod

By

–

20 July 2023 19h34

How does 36 exaFLOPs compare to other AI supercomputers? It's 4x the performance of Google's latest TPU Pod v4 and 9x the performance of Nvidia's yet complete Israel-1.

Cerebras Launches Condor Galaxy-1 4 Exaflop AI Supercomputer

By

–

20 July 2023 19h34

Today we are announcing Condor Galaxy-1: a 4 exaflop AI supercomputer built in partnership with @G42ai
. Powered by 64 Cerebras CS-2 systems, 54M cores, and 82TB of memory – it's the largest AI supercomputer we've ever built. But that's not all: CG-1 is just the start..

→ View original post on X — @groqinc

Samsung Foundry Partners with Tenstorrent and Groq on AI Semiconductor R&D

By

@groqinc

–

20 July 2023 17h57

"Samsung Foundry has recently launched Research & Development projects with American AI semiconductor startups Tenstorrent and Groq… both companies are considered promising firms in the global AI industry."