Cirrascale Cloud Services, a provider of public and private dedicated, multi-GPU cloud solutions enabling deep learning, has announced the availability of IBM Power Systems AC922 on its deep learning and HPC cloud platform.
The company’s deep learning and HPC cloud platform now supports multiple bare-metal configurations of IBM Power Systems AC922 Server with up to four Volta-based NVIDIA GPUs.
The IBM Power Systems AC922 is a complete redesign supporting the latest technologies including CAPI 2.0, PCIe Gen 4, as well as NVLink 2.0 which is embedded at the silicon level and tightly coupled with the CPU. This would enable up to 150GB/sec of bi-directional bandwidth for data-intensive workloads, nearly 5X the CPU:GPU bandwidth of PCIe. Each GPU would be serviced with 300GB/sec of NVLink bandwidth, “nearly 10X the data throughput of PCIe x16 3.0 platforms.”
Additionally, customers would notice reduced latencies with memory and NVMe devices. According to Cirrascale Cloud Services, overall, these optimizations accelerate deep learning and HPC workloads such as weather modeling, autonomous transportation, natural language processing, and speech recognition.
Bare Metal Servers
Cirrascale Cloud Services purposely offers dedicated GPUs as a bare metal offering. This would give users the full power of modern, dedicated, directly accessible GPUs and associated compute hardware, as well as ultimate flexibility in system configuration. The available configurations would ensure that a wide variety of application types can be addressed including deep learning, artificial intelligence, high-performance data analytics, and accelerated databases.
“Deep learning and HPC workloads require an advanced level of computing architecture between CPUs and GPUs to resolve bandwidth bottlenecks,” said Mike LaPan, Vice President, Cirrascale Cloud Services. “With the Cirrascale Cloud Services platform offering the new IBM POWER9 servers, customers can experience the next level of GPU-accelerated computing across industries.”