Cerberas & G42 have unveiled the 4 Exaflops Condor Galaxy 1 AI supercomputer with thousands of AMD EPYC PUs & millions of AI cores.
Cerebras in Partnership With G42, Is Set to Open New Gateways in Cloud Computing & Generative AI
Cerebras Systems, one of the industry-leading companies in AI, and G42, a technology holding group, have unveiled a new standard in interconnected supercomputers. The companies will develop nine different supercomputers, with the highlight being the CG-1 (Condor Galaxy 1), which is expected to feature the world's first 4 exaFLOPs computing power.
Condor Galaxy 1 AI supercomputer 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
You may have heard about the above companies for the first time, but both have been involved in technological development for a long time. Cerebras Systems is known for its efforts in the generative AI industry, primarily through its CS-2 Wafer Scale Engine, which has achieved an astonishing milestone of 2.6 trillion transistors. The chip is said to be the largest ever built and is quoted as a "single CS-2 system, akin to a supercomputer all on its own".
However, Cerebras, in partnership with G42 has gone one step ahead through their new supercomputers. The companies have provided details on CG-1, which is expected to set a new benchmark in the industry. According to the blog post, it is said that CG-1 is linked with 64 Cerebras CS-2 systems, incorporating them into a single supercomputer with a combined power of 4 exaFLOPs.
Well, if you want to know about the immense potential of the CG-1, the current active supercomputer, the AMD-powered Frontier, is reportedly equipped with 1.1 exaFLOPs. The CG-1 holds almost a four times increase, an enormous number in the supercomputer industry. Here is what Cerebras's CEO, Andrew Feldman has to say about the development:
Delivering 4 exaFLOPs of AI compute at FP 16, CG-1 dramatically reduces AI training timelines while eliminating the pain of distributed compute.
Many cloud companies have announced massive GPU clusters that cost billions of dollars to build, but that are extremely difficult to use. Distributing a single model over thousands of tiny GPUs takes months of time from dozens of people with rare expertise. CG-1 eliminates this challenge
Diving into the specifications of the CG-1 supercomputer, it features 4 exaFLOPs of 16-bit compute power along with 600 billion parameter models. The system is reportedly equipped with 54 million AI-optimized compute cores, 388 terabits per second of fabric bandwidth, and 72,704 AMD EPYC CPU cores. The onboard equipment is nothing like what we have seen, and judging by the metrics revealed, the CG-1 is ready to dominate the industry.
AMD is committed to accelerating AI with cutting edge high-performance computing processors and adaptive computing products as well as through collaborations with innovative companies like Cerebras that share our vision of pervasive AI.
Driven by more than 70,000 AMD EPYC processor cores, Cerebras’ Condor Galaxy 1 will make accessible vast computational resources for researchers and enterprises as they push AI forward
-Forrest Norrod, VP of Data Center Solutions Business Group AMD
Cerebras and G42's ambitions don't stop here, as the company intends to interconnect three supercomputers (CG-1, CG-2, and CG-3), aiming to reach 12 exaFLOPs, an enormous milestone. Moreover, the company plans to bring six further supercomputers into action by 2024, setting the bar up to 36 exaFLOPs of total computing power. The disclosed ambitions are indeed gigantic in the modern world, and this could potentially open up a new era in cloud computing and the generative AI industry.
| Condor Galaxy | Phase 1 Delivered | Phase 2 Q4 2023 | Phase 3 H1 2024 | Phase 4 H2 2024 |
| ExaFLOPS | 2 | 4 | 12 | 36 |
| CS-2 Systems | 32 | 64 | 192 | 576 |
| Supercomputer Centers | 1 | 1 | 3 | 9 |
| Milestone | Largest CS-2 deployment to date | First 64-node Cerebras AI supercomputer | First distributed supercomputer network | Largest distributed supercomputer network |
from Wccftech https://wccftech.com/cerebras-4-exaflops-condor-galaxy-1-ai-supercomputer-over-70000-amd-epyc-cores/
0 Comments