Every year it’s inspiring to see an eclectic group of students, scientists, researchers and technological innovators from the high performance computing (HPC) community come together at SuperComputing 2017 (SC17). It’s the one place where some of the brightest minds in computing can share ideas and map out of the future of cutting edge technologies.
At the heart of all the work and collaboration accomplished this week is a foundation set in mathematics and the ability of computers to handle a mind-boggling number of computations and an incredible amount of data. Modeling, simulation, and data analysis now underpin everything from protein folding and gene research to high particle physics and cosmology.
When you think about math in the context of the datacenter and with high performance CPUs, more often than not you’re really talking about how you leverage the utility of floating point.
Uncovering the Utility of Floating Point
Earlier this year, we released our new family of high-performance server and datacenter processors, the AMD EPYC™ 7000 series. The EPYC processor is a highly-scalable CPU that includes a very powerful and capable floating point unit (FPU). The philosophy of the EPYC FPU is to deliver the most easily usable performance. Its unique co-processor architecture enables a high floating point instruction issue rate, great memory bandwidth, with energy efficiency to maintain a high frequency even when loaded. We support all the floating point instructions used by today’s systems, up through AVX2. Running existing codes or even the latest floating point benchmarks, the numbers speak for themselves: EPYC delivers incredible FP performance.
We define utility as usefulness over costs. It’s an idea that underpins computing technology, businesses and overall strategies. We’re working to provide and apply the technologies that let the HPC community do more with much less. When you do the math, we’re delivering an EPYC processer that runs at up to 3X* the performance per dollar than the competition. Whether you’re thinking about the immediate or long-term implications, it’s the utility that makes all the difference. And when you couple the EPYC CPU with the utility of the Radeon™ GPU, you can extend that even further.
Unlocking the Power of AMD at SC17: EPYC + RADEON INSTINCT
Although floating point is key to how we approach the next generation of CPUs, for massively vectorized workloads there is an even better choice: the power of today’s GPUs which far exceed any CPU on highly parallel applications. AMD offers leading compute-GPUs in our Radeon Instinct™ line. You can now pair the best of AMD in high-performance CPU and GPU with EPYC and Radeon Instinct to create a heterogeneous supercomputing solution that tackles real-world applications that floating point thrives in – from fluid dynamics and weather mapping to oil and gas exploration and more.
At SC17, we’re bringing the power of a combined EPYC and Radeon Instinct platform to the show floor with Project 47 (P47). Inventec’s P47 platform provides direct access to four Radeon Instinct GPUs through a single EPYC processor without the need for PCI switches, which removes design barriers and streamlines performance. The AMD-based platform then flexes its
scalability by contributing to 20 1P EPYC processor-based Inventec servers to produce a petaFLOPS of single-precision computing. By supporting both heterogeneous supercomputing systems and memory-bound CPU platforms, EPYC addresses several real-world applications to support safer, more productive operations.
Focusing on Math for the Masses
I found my inspiration in science and electronics early on as a student, and there’s a group of university students joining SC17 to compete in the Student Cluster Competition that we’re thrilled to help find their individual sparks.
AMD, along with Supermicro and Mellanox, is supporting a student team from Northeastern University. Using a system developed around EPYC and Radeon Instinct, the team will square off against international competitors to run a mix of known and unknown HPC codes around the clock over a couple days to test their high-performance skills.
The lessons learned from that competition will be invaluable. With any luck, the next-generation of visionaries will find their moment of math inspiration and use the high-performance technologies of today to define a more promising tomorrow.
*Based on SPECfp®_rate2006 scores published on www.spec.org as of October 25, 2017. 2 x EPYC 7601 CPU ($4,200 per processor at AMD 1ku pricing) in Sugon A620-G30, Ubuntu 17.04, x86 Open64 v22.214.171.124 Compiler Suite, 512 GB PC4-2666V-R memory, running at 2400 1 x 1TB SATA 7200RPM has a peak score of 1850 (base score 1670); versus 2P Xeon Platinum 8180M ($13,011 per processor per ark.intel.com)-based Cisco UCS C240 M5 system with SUSE Linux Enterprise Server 12 SP2, ICC 126.96.36.199, 384GB PC4-2666V-R memory, 1x240GB SATA SSD score of 1830 (base score 1800). SPEC and SPECfp are registered trademarks of the Standard Performance Evaluation Corporation. See www.spec.org for more information. NAP-49