ScaleMP and AMD Benchmark Best-in-Class Large-System Performance on SPEC CPU 2017

Print Friendly, PDF & Email

Today ScaleMP announced that it has accomplished best-ever recorded results for the SPEC CPU 2017 benchmark for any system size of four processors and higher. ScaleMP’s results for 32-, 24- and 16-processor systems are now holding the top three positions for best results ever.

In 2019, AMD and ScaleMP announced a strategic collaboration to enable AMD server OEMs to create systems with four, eight and up to 128 processor sockets, up to 8,192 CPUs and 256 terabytes of shared memory using ScaleMP’s vSMP Foundation line of products.

We are extremely pleased with these new benchmark results, which show superiority across the board,” said Shai Fultheim, founder and CEO of ScaleMP. “Customers can now scale their systems on demand for more computing or more memory, pay only for what they need, and be sure that they are buying the best and most cost-effective solution available that delivers the best performance.”

The following table summarizes the current best SPEC CPU2017 Integer Rates and SPEC CPU2017 Floating Point Rates for every scale of server system and contrasts it with the previous best system/result for that scale.

ScaleMP’s unique software-defined server technology enables AMD to capture the leading positions for the system-scale SPEC CPU2017 benchmark without any custom engineering,” said Raghu Nambiar, corporate vice president, Datacenter Ecosystems & Application Engineering at AMD. “Through our collaboration with ScaleMP, we expect to continue to dominate the scale-up benchmarks as this innovation continues.”

ScaleMP solidifies the leading position of its patented vSMP Foundation technology and line of products as the best scale-up computing architecture for compute-intensive workloads, with up to 96% better performance compared to alternative systems. The results, made public by the, show that ScaleMP’s on-demand composed computing architecture provides better scaling than hardware-only approaches.

Sign up for our insideHPC Newsletter