Editor's Pick: New 1U Server for HPC Features Cascade Lake Xeon and Tesla GPU

Dear DE Reader:

Once upon a time, engineering IT people wondered if in-house high-performance computing (HPC) clusters were on their way out, due to the rise of cloud services. Then engineering discovered artificial intelligence and deep learning, and found a new value driver for in-house HPC clusters. 

Intel has released the newest generation of Xeon Scalable processors. As usual the first out of the gate with products based on this new Xeon CPU is boutique power computing vendor Microway. Their latest NumberSmasher clusters are our Editor’s Pick of the Week. 

Computer vendors work closely with the CPU vendors, timing releases to coincide with new CPUs. This is the “Cascade Lake” update for Intel Xeon. For Microway’s customers, it means increased CPU core counts, higher memory capacity and bandwidth, and new clock speeds. It also means new features from Microway of interest to HPC deployments in engineering. 

Why does HPC matter to engineering? Because there is an increased reliance on deep learning to tackle the big problems. For example, in the past an issue like jet engine airflow would start with an assumption, then iterate toward a solution. Now the problem is assigned to a deep neural net, which can examine thousands of assumptions to reach an approximation of the solution—which may or may not match the initial assumption. Engineers can then improve on the deep learning “suggestion” and move forward much faster with an ideal solution.

Like most workstations and server products, there is not one Microway NumberSmasher, but a complete line serving various use cases. In this report, we’re focused on the NumberSmasher 1U 4 GPU Single Socket Server as the reference model. The 1U offers Microway’s highest GPU/CPU density in a single socket compact rack footprint. It is designed for accelerated workloads that require a lower level of computation on the host CPU and most processing assigned to the GPUs. 

The 1U can support up to 4 NVIDIA Tesla V100 or P100 GPUs. The V100 is currently NVIDIA’s graphics processing unit (GPU) for data centers, ideal for engineering HPCs. The 1U NumberSmasher is available with up to 28 Intel processor cores. There are six RAM slots, holding up to 768 GB memory. For storage, there are up to 4 PCI-Express 3.0 double-width slots and 2 hot-swap 2.5 inch drive bays running at 6 Gbps. The Cascade Lake SP CPU driving this is available at various clock speeds up to 3.8 GHz.  

Learn more details on the 1U NumberSmasher and other products in the Microway HPC line in this week’s Product Brief. Thanks for reading; we’ll see you next week with another Editor’s Pick of the Week. 

—The DE Editors

[RECOMMENDATIONS FOR ADDITIONAL IN-TEXT URLS TO NEW PRODUCT WRITE-UP]

Second paragraph, second sentence, the phrase “boutique power computing vendor Microway”

Third paragraph, fourth sentence, the phrase “of particular interest to HPC deployments in engineering”

Fifth paragraph, second sentence, the phrase “NumberSmasher 1U 4 GPU Single Socket Server”

Tenth paragraph, second sentence, the phrase, “this week’s Product Brief”