Cirrascale Cloud Services has launched its Cerebras Cloud @ Cirrascale platform, which provides deep learning infrastructure solutions for autonomous vehicles, Natural Language Processing (NLP), and computer vision workflows. Using Cirrascale’s scalable and accessible cloud service, Cerebras’ Wafer-Scale Engine (WSE-2) deep learning processor would deliver the highest performance to more users via Cirrascale’s scalable and accessible cloud service.
Cerebras Cloud delivers the wall-clock training performance of many 10s to 100s of GPUs for training and inference workloads on a single CS-2 system, enabling cluster-scale performance with single-node programming simplicity with standard ML frameworks.
In a recent comparison, a Cerebras system showed 9.5x faster training time to accuracy on a BERT-type model for NLP compared to an 8-GPU server, according to Cirrascale. Since training time does not scale linearly with GPUs, users would need more than 120 GPUs to train the model to the same accuracy as a single CS-2. The WSE-2 powering the Cerebras Cloud uniquely handles fine-grained and dynamic sparsity that are common in neural network workloads, but are challenging for traditional processors.
“Cirrascale is thrilled to bring Cerebras’ world-class artificial intelligence (AI) compute power to the cloud – we are truly democratizing AI by broadening deep learning access and enabling large-scale commercial deployments across leading Fortune 500 customers, research organizations and innovative startups,” said PJ Go, CEO at Cirrascale Cloud Services. “The Cerebras Cloud @ Cirrascale will advance the future of deep learning and greatly accelerate the performance of AI workloads, all through a single device that is easy to use and simple to deploy.”
With every component optimized for AI work, the Cerebras Cloud @ Cirrascale would deliver more compute performance at less space and less power. Depending on workload, from AI to HPC, it would deliver hundreds or thousands of times more performance than legacy alternatives, while only using a fraction of the space and power. Cerebras Cloud is designed to enable fast, flexible training and low-latency data center inference, thanks to greater compute density, faster memory, and higher bandwidth interconnect.
“We are excited to partner with Cirrascale and introduce the Cerebras Cloud @ Cirrascale, bringing the power and performance of our CS-2 system to more customers looking to glean world-changing insights from massive datasets sooner,” said Andrew Feldman, CEO and Co-Founder of Cerebras. “We designed the WSE-2 with 850,000 AI-optimized cores so that the CS-2 system at the foundation of the Cerebras Cloud @ Cirrascale offering can deliver cluster-scale acceleration of AI workloads easily with a single system, thereby enabling customers to deploy solutions faster, using large, state-of-the-art models for training or inference.”
The Cerebras Cloud @ Cirrascale is now accessible in weekly or monthly flat-rate allotments. Cirrascale’s Cerebras Cloud may be readily integrated with a client’s existing online cloud-based process at multiple hyperscalers, resulting in a secure multi-cloud solution.