Our team of HPC experts has over two decades of expertise in porting, modernizing, and optimizing large HPC codes. The team utilizes our own specialized performance analysis tools to capture details about the way the application utilizes the features of the system. Our team has worked with HPC codes from industry, government, and academia. EP Analytics improved the performance of a mission-critical DOD HPC code by over 10 times.
“Thanks again for all your help. We are running at 15-45 secs/step right now and scaling 32:64 [cpus] well. It was taking 5 min/step last night!!”
- Dr. Rainald Löhner, DOD FEFLO Developer
In order to improve the performance of a large scale HPC application, one needs to understand the complex interactions between the system, compiler, runtime, and the application. EP Analytics has specialized tools that work on HPC production executables (or binaries) to capture such interactions. These interactions then inform optimization strategies. Our team actively engages with the application developers to implement identified optimizations.
Our performance analysis and modeling expertise and tools are excellent additions to either sides of the HPC System procurement process. Using our specialized HPC tools, we can characterize large HPC applications. These characterizations can be used to develop predictive models that can inform the impact of different hardware design points on applications' performance (e.g., the addition of NVM memory technology). We have provided procurement support for many "household" name semiconductor companies.
We help our customers understand their workloads better. Modern computing architectures expose many performance features that applications can utilize for improved time-to-solution. Quantifying how well the applications leverage such features is the first step towards understanding your workload efficiency and devising strategies to improve the efficiency either through software optimizations or hardware upgrade.
Today’s HPC systems have multiple levels of parallelism, and performance-critical hardware features change each year. Code modernization involves code changes that can utilize multiple levels of parallelism and new hardware features (without resorting to ninja coding that utilizes non-portable architecture-specific code changes). These changes allow the compiler to do the architecture-specific optimizations.
We help our customers ensure continuous integration and continuous delivery of their HPC applications through automated testing on local, cloud-HPC, and government HPC systems.
The HPC experts at EP Analytics have been involved in porting large scale (legacy) HPC applications to new architectures for decades. Our unique understanding of the system and application enables efficient use of the architectural features of the system.