Computer servers in data centers and enterprise offices are critical to the business functions of organizations. As more applications move to the cloud and virtual computing takes an increasingly larger role in day-to-day computing functions, computer server solutions for cloud and virtual computing must meet high functionality, performance, reliability, and availability requirements. Vendors supplying computer servers for these markets must ensure that their products meet these requirements or lose their competitiveness.
High performance computing requires computer server hardware to be fast in terms of processing speed and network interconnect throughput. Most high performance computing tasks involve parallel processing using multiple processors in one server and between servers as members of a cluster. Tasks can be accomplished faster with more than one or two processors and/or servers working on them. Some high performance computing server solutions use GPUs to perform massively parallel processing. These GPUs can contain thousands of computing cores (such as the Nvidia® Tesla™ Kepler-based GPUs). Racks loaded with these GPU-optimized servers can accomplish workloads on par with supercomputers.
When servers participate in a cluster, the network interconnects linking them must be fast enough to support the communication throughput between all the servers to avoid bottlenecks. As a result, each high performance computer server is configured with the fastest network interconnect available such as 56-Gigabit FDR InfiniBand or Fibre Channel.
A third performance factor in high performance computing environments is server storage. In addition to having enough storage capacity for targeted applications, storage in high performance computing servers must incorporate high throughput data transfer interfaces such as SAS 3.0 12Gb/s. Storage media also affect read and write speeds. Solid-state drives being faster but more expensive can be configured for a computer server if cost is not a consideration. Alternatively, hard disks can be used in conjunction with a solid-state drive cache mechanism with the cache accelerating most frequently accessed data.
Thinkmate is a premier supplier of high performance computing servers. It offers several server product lines for high performance computing including Supermicro servers. These product lines comprise high density servers, blade servers, and GPU-optimized servers.
Thinkmate’s HDX range of high density servers come in 1U, 2U, and 3U chassis. All footprints contain more than one computer server node. The HDX 1U servers contain two nodes with two multi-core Intel® Xeon® or AMD Opteron™ CPUs in each node. The HDX 2U servers include Supermicro 2U Twin and 2U Twin2® configurations. Supermicro 2U Twin includes two hot-swappable server nodes containing two CPUs in each node. Supermicro 2U Twin2® includes four hot-swappable nodes containing two CPUs in each node. The HDX 3U configuration is a Supermicro SuperServer® 3U MicroCloud. It has eight individual server nodes with one CPU in each node.
In addition to the HDX range, Thinkmate also sells Supermicro FatTwin™ servers. The FatTwin™ server design leverages the 4U chassis size to contain two, four, or eight modular and hot-pluggable computer server nodes (with two Intel Xeon CPUs per node). Configurations can include Nvidia® Tesla™ and Intel Xeon Phi™ co-processors. Each chassis has redundant power supplies shared between all the nodes.
Blade servers from Thinkmate include the Supermicro SuperBlade® range that comes in 7U enclosures that fit into standard 42U racks. The highest density SuperBlade configuration is the TwinBlade® that can pack 20 computer server nodes (40 CPUs) per 7U enclosure or 120 server nodes per 42U rack.
Most servers from Thinkmate can be configured with SAS 3.0 12Gb/s drives and drive controllers as well as 56-Gigabit FDR InfiniBand or Fibre Channel network connectivity making Thinkmate servers the ideal platform for high performance computing.