The AMD second generation of EPYC processors, code-named “Rome” and based on the Zen 2 microarchitecture, are emerging as an alternative processor for high-performance computing (HPC) server applications.
The processor, which was released last August, features up to 64 cores per socket with 128 PCIe 4.0 lanes and 8-channel memory. AMD touts it as offering a very high performance per dollar in the marketplace. For organizations familiar with AMD desktop technology, EPYC Rome is to the server what AMD Ryzen 3000 was to the desktop. Namely, the second-generation AMD EPYC Rome processor offers significantly improved instructions per cycle over the first-generation processor, as well as more cores and better thermal efficiency. Additionally, the processor features a higher base frequency and a higher boost frequency than the previous AMD top-of-the-line processor.
Workloads That Will Benefit from AMD EPYC
When launched, AMD noted that Rome offers far more CPU threads per socket than Intel’s Xeon Scalable CPUs do. The processor also supports a higher DDR4 clockrate (up to 3200 MHz). Most important for demanding data center applications, the 128 PCIe 4.0 lanes each have twice the bandwidth of a PCIe 3.0 lane. Such performance is critical in HPC and large data center applications because data ingestion by CPUs is often a bottleneck. The higher performance capacity means workloads will not be degraded, and jobs will run faster. Most Host Bus Adapters may not be able to immediately take advantage of this larger pipe, but we don’t expect that to be the case for very long. Over time, more and more network adapters, controllers, etc. will be able to fully take advantage of PCI Express® 4.0. That’s not to say that right now there are no benefits. In fact, the immediate impact is significantly more lanes for PCI devices, such as NVMe devices. This will dramatically improve IO performance and increase the number of devices connected to the server.
Another very basic advantage for HPC use cases is that since AMD EPYC processors have more cores than their comparable competitive processors, they deliver more computing power. Processors with more cores are ideal for advanced modeling and simulation algorithms, such as a finite element simulation. For these workloads, a job needs to use a very large number of cores at a given time. And the workloads often require large amounts of shared memory. This changes the game for HPC environments (ie. Design & Engineering, Weather Modeling, Life Science, and Artificial Intelligence), where total core count directly relates to program success. AMD® is realizing their investment in the 7 nm technology is finally paying off.
AMD EPYC processors deliver other benefits for HPC workloads and environments. By virtue of having more cores, the processors can run more virtual machines (VMs). As a result, a company gets more efficient compute resource use out of a single system. This is ideal for virtual desktop environments (VDI) as well as converged / hyper-converged applications.
Selecting the Right Technology Platform
AMD use for HPC server workloads has had a mixed track record over the years. However, the introduction of AMD EPYC processors seems to have caught the attention of the marketplace.
Noting this interest in the user community, PSSC Labs is rolling out solutions based on the processor this year. PSSC Labs has a more than 30 years history of delivering systems that meet the most demanding workloads across industries, government, and academia.
Its offerings include the PowerServe Uniti Servers line that leverages the latest components from Intel® and Nvidia®. These servers are ideal for a wide range of applications, including AI and deep learning, as well as for computational and data analysis.
PSSC Labs also offers CloudOOP Big Data Servers that deliver the highest level of performance one would expect in an enterprise server combined with the cost-effectiveness of direct attach storage for Big Data applications. The servers deliver 200+ MB/sec sustained IO speeds per hard drive (which is 30%+ faster than other OEMs.)
Due to the increase in the number of PCI Express lanes, PSSC Labs is upgrading their CloudSeek Database Servers as well. The products we plan to bring to market in the near future will be built to support up to 36 x NVMe devices
PSSC Labs plans to have an AMD EPYC version of all its platforms for HPC and Big Data by the second half of the year. User interest is sparking this move. The company has not sold an AMD-based server in more than ten years. The impression PSSC gets from the end-user community and its own evaluation of the new processors is that AMD is a viable choice for the most demanding HPC workloads.