Andromeda, a 13.5 Million CoreAI Supercomputer
Andromeda is one of the largest AI supercomputers ever built. It delivers more than 1 exaFLOPS of AI compute and 120 Petaflops of dense compute.
Andromeda is the only AI supercomputer to ever demonstrate near-perfect linear scaling on large language model workloads, and is extremely simple to use.
By the numbers
Andromeda is a >1 exaFLOPS AI Supercomputer, comprised of 16 Cerebras CS-2systems.
>120
PetaFLOPs of dense compute
18,176
AMD Epyc™ Gen 3 cores
96.8
terabits-per-second node-node fabric bandwidth
Near-Perfect Performance Scaling
Unlike any known GPU-based cluster, Andromeda delivers near-perfect scaling across GPT-class large language models, including GPT-3, GPT-J and GPT-NeoX. Near-perfect scaling means that that as additional CS-2s are used, training time is reduced in near perfect proportion.
| Model | 1 CS-2 | 2 CS-2 | 4 CS-2 | 8 CS-2 | 16 CS-2 |
|---|---|---|---|---|---|
| GPT3-XL 1.3B | 1 | 1.99 | 3.94 | 7.87 | 15.50 |
| GPT3-XL 1.3B MSL 10K | 1 | 1.99 | 3.97 | 7.95 | 15.87 |
| GPT3 2.7B | 1 | 1.98 | 3.91 | 7.86 | 15.62 |
| GPT3 6.7B | 1 | 1.99 | 3.89 | 7.91 | 15.45 |
| GPT3 20B | 1 | 1.92 | 3.75 | 7.93 | 15.32 |
| GPT-J 6B | 1 | 1.97 | 3.65 | 7.69 | 14.52 |
| GPT NeoX 20B | 1 | 1.98 | 3.92 | 7.89 | 15.45 |
| GPT-J 25B MSL 10K | 1 | 1.98 | 3.95 | 7.92 | 15.85 |
What our customers are saying
Get a Test Drive
Access to Andromeda is available now, and customers and academic researchers are already running real workloads and deriving value from the leading AI supercomputer’s extraordinary capabilities. If you have an “impossible” problem, we can help.