Product attributes
Other attributes
Condor Galaxy is an AI supercomputer developed by Cerbras in partnership with G42. Condor Galaxy 1 (CG-1), the first of nine planned supercomputers, was made available on July 20, 2023. CG-1 is a 4 exaFLOPS AI supercomputer for training large generative models. Cerebras states that the nine inter-connected supercomputers will be completed in 2024, offering 36 exaFLOPS of AI compute.
CG-1 is managed and operated by Cerebras for G42. The supercomputer is available through the Cerebras Cloud White Glove Service and through G42 Cloud for commercial customers training generative AI models. CG-1 is located in Santa Clara California at the Colovore data center. CG-1 is comprised of 64 Cerebras CS-2 systems, linked together via Cerebras SwarmX technology. It supports up to 600 billion parameters and is extensible up to 100 trillion.
CG-1 specifications:
- 4 exaFLOPS of AI compute at FP16 with sparsity
- 54 million AI-optimized compute cores
- 82 terabytes of memory
- 64 Cerebras CS-2 systems
- Base configuration supports 600 billion parameters, extendable up to 100 trillion.
- 386 terabits of internal cluster fabric bandwidth
- 72,704 AMD EPYC Gen 3 processor cores
- Native hardware support for training with 50,000-token sequence length, no third-party libraries needed.
- Data parallel programming model with linear performance scaling
The introduction of CG-1 in July 2023 is phase 1 of Cerebras's Condor Galaxy roadmap. Phase 2 will see the expansion of CG-1 to 64 CS-2 systems at 4 exaFLOPS. Phase 3 is building two more supercomputers across the United States, bringing the total deployed compute to three centers at 12 exaFLOPS. Phase 4 is building six more supercomputers, bringing the full install base to nine instances at thirty-six exaFLOPS of AI compute. Cerebras states that Condor Galaxy will be fully deployed in 2024, becoming one of the largest cloud AI supercomputers in the world. Thirty-six exaFLOPS makes it nine times more powerful than Nvidia’s Israel-1 supercomputer and four times more powerful than Google’s largest announced TPU v4 pod.
