Over at the Lenovo Blog, the company has unveiled a new ThinkSystem based on a modular infrastructure targeted at Ai workloads.
With the release of the new ThinkSystem SR670, Lenovo has delivered on a new approach to server design by providing an architecture that can be customized for your workload needs. We call this concept “integrated modularity.” It means that the system was designed from the ground up, optimized for connection to value-added devices, such as GPUs and accelerators. The ThinkSystem SR670 completely re-thinks traditional board-level design: no longer are devices in fixed positions on a large, static motherboard. Instead, the system is compartmentalized, with different functions placed throughout the system, optimized for access, cooling and performance. Integrated modularity is achieved by leveraging flexible PCI lanes instead of PCIe slots. This allows Dynamic PCIe port splitting or aggregation, depending on customer requirements and determined by how the devices are connected to the processor. By separating the peripherals from the CPU, customers have the flexibility to take advantage of advancements in those subsystems without being tied to the CPU/motherboard complex.
ThinkSystem SR670 highlights:
- Accelerating artificial intelligence performance
- Up to 4 best-of-breed GPUs for accelerated performance
- Scalable cluster solution for AI and HPC workloads, supporting the LiCO platform
- Perfect balance of performance, density, and TCO for AI & GPU-intensive workloads
- Supports high-speed Mellanox EDR InfiniBand, Intel 2x 10GbE & Intel 2x 1GbE networking
- Supports standard SATA HDDs & SSDs, and M.2 boot SSDs for even greater storage flexibility
From recognizing speech to training virtual assistants to converse naturally, from detecting lanes on the road to teaching autonomous cars to drive, data scientists are taking on increasingly complex challenges with Artificial Intelligence. Tackling some of humanities greatest challenges by leveraging AI will change what we expect from technology and specifically the hardware that powers it.
By developing the ThinkSystem SR670 platform with integrated modularity, Lenovo has recognized that technologies around the CPU are evolving rapidly and driving massive performance gains for many types of workloads. System designs need to evolve to provide the capability and flexibility for customers to add more revolutionary technologies to the system faster. Customers shouldn’t be gated by a system that can’t change and grow with them as they build their intelligence platforms of the future.
The ThinkSystem SR670 is a 2U two-processor system with capacity for up to 4 high-powered NVIDIA©Tesla© V100, that will empower organizations with the hardware performance needed in a world where more accelerated computing is required. The system is perfect for emerging Artificial Intelligence workloads, which require the additional processing power for model training and algorithm development. For larger AI or HPC environments, the ThinkSystem SR670 uses open standards, not proprietary technology, to scale out to an economical cluster that optimizes both system and GPU performance. The ThinkSystem SR670, coupled with NVIDIA Tesla V100, supports parallel processing for the most commonly used AI frameworks such as Caffe2, TensorFlow, Torch, and PaddlePaddle, as well as others. This unique design also has the capacity to support high speed networking including two high speed adapters or the future 200Gb/s HDR InfiniBand adapter, ensuring you have the perfect balance of connectivity and performance.