Ron Javor, Product Manager AI, HPE
Chuck Newman, Cloud Solutions Manager, HPE
Latest STAC-A2 report documents records in speed, throughput, and efficiency.
On the historic afternoon of May 6, 1954, Roger Bannister redefined the limits of human achievement by conquering the first 4-minute mile. Just as Bannister’s feat revolutionized running, HPE and NVIDIA have shattered records in the world of financial services with their groundbreaking performance in the STAC-A2 benchmark test.
The STAC-A2 Benchmark
The Strategic Technology Analysis Center (STAC®) recently performed the STAC-A2 Benchmark tests on an HPE ProLiant XL675d Gen 10 Plus Server configured with eight NVIDIA H100 Tensor Core GPUs. Designed by quants and technologists from some of the world’s largest banks, STAC-A2 serves as the gold standard for testing technology stacks used for compute-intensive quantitative finance workloads, involved in price discovery and risk management.
This powerful combination achieved unparalleled results in many performance and efficiency benchmarks, with several results that outperformed all publicly reported solutions to date. Notable achievements include:
- The first sub-10 ms warm time (8.9 ms) in the baseline Greeks benchmark[1]
- A cold time in the baseline Greeks benchmark (38 ms), over 3x faster than any previous benchmark[2]
- The fastest warm[3] (0.51 s) and cold[4] (1.85 s) times in the large Greeks benchmarks
- The most correlated assets[5] (400) and most paths[6] (310,000,000) simulated in 10 minutes
- The best energy efficiency (311,045 options/kWh)[7]
These results highlight the potential of GPU-accelerated computing in addressing the most demanding real-world financial applications, offering speed and energy efficiency in equal measure.
The Winning Combination
At the core of this breakthrough is the HPE Apollo 6500 Gen10 Plus System, comprised of a 6U HPE ProLiant XL675d Gen10 Plus Server with 8 NVIDIA H100 80 GB PCIe accelerators (up to 10 are supported) to form an exceptional platform for High-Performance Computing (HPC) innovation. This solution supports NVIDIA GPUs in either an SXM configuration or in a PCIe Gen 4.0 configuration, the latter of which was used for this benchmark.
Built for the exascale era, the HPE Apollo 6500 Gen10 Plus System accelerates performance with NVIDIA HGX A100 or H100 PCIe GPUs and is certified to run NVIDIA AI Enterprise software for secure, stable, supported production AI and HPC. NVIDIA AI Enterprise is an enterprise software platform that provides development tools, frameworks, and pretrained models to accelerate quantitative finance and data science workloads and streamline the development and deployment of production-ready generative AI, computer vision, speech AI, and more. This HPE and NVIDIA full-stack, purpose-built platform provides enhanced performance with premier GPUs, fast GPU interconnect, high-bandwidth fabric, and configurable GPU topology, providing rock-solid reliability, availability, and serviceability (RAS). For those looking for the additional performance afforded by the NVIDIA H100 GPU in the SXM configuration, HPE also offers the HPE Cray XD670 and HPE Cray XD665 solutions.
The new NVIDIA H100 GPU ushers in a new era of performance. Built for AI, HPC, and data analytics, the H100 GPU accelerates over 4,000 applications and is available everywhere, from the data center to cloud. . NVIDIA AI Enterprise software is included with the H100 PCIe GPU-integrated systems to simplify AI adoption with the highest performance.
For HPC applications, the H100 GPU almost triples the theoretical floating-point operations per second (FLOPS) versus the NVIDIA A100 Tensor Core GPU and adds the Transformer Engine with FP8 precision to achieve even higher performance. These new H100 GPUs feature fourth-generation Tensor Cores and second-generation Multi-Instance GPU (MIG), built-in NVIDIA Confidential Computing, and NVIDIA NVLink technology. With these features, the H100 aims to securely accelerate workloads for every data center from enterprise to Exascale.
This new solution offers the potential for significant savings through full scalability across the entire platform and ecosystem with HPE servers and NVIDIA GPUs, resulting in higher performance and lower operating costs.
Expanding FSI (Financial Services Institutions) Possibilities
Beyond intraday risk valuation (Greeks) and options pricing, this breakthrough in performance opens doors to a range of financial service workloads including:
- Leveraging alternative data and the accelerated compute platform enables traders to harvest petabytes of data to discover alpha signals previously undetectable.
- Fraud detection gains a powerful ally in the form of rapid data processing, enabling real-time identification of potentially fraudulent activities.
- Recommendation engines reach new heights by swiftly analyzing consumer behavior, driving customer satisfaction and loyalty.
- Process automation benefits from accelerated task execution, reducing manual intervention and boosting operational efficiency.
In the realm of speech recognition, this dynamic duo processes spoken language with unmatched speed, paving the way for advanced applications.
Collaborative Excellence
The close collaboration between HPE and NVIDIA drives new solutions to meet the needs of the financial services industry.
After all, it takes more than just a high-end GPU to make everything fast – getting data into the GPU requires high-speed networking and a fast bus for the server. When you combine HPE systems with NVIDIA GPUs, you’ve created a great solution that can operate effectively at record-breaking speeds.
NVIDIA-Certified
HPE ProLiant and Apollo 6500 Gen10 are NVIDIA-Certified Systems. This means that the systems have been comprehensively tested by HPE engineers to help ensure a faster, more reliable, and predictable deployment. NVIDIA-Certified HPE systems bring together HPE servers and NVIDIA GPUs in optimized configurations that are validated for performance, manageability, security, and scalability and are backed by enterprise-grade support from HPE and NVIDIA.
End-to-End AI Software Stack
HPE and NVIDIA are also collaborating to offer customers of all sizes a complete, end-to-end AI software stack to create custom, accurate, and trustworthy machine learning models. This comprehensive platform speeds up time to value for AI model developers, brings faster time to market by dramatically increasing productivity, enhances data science collaboration, and reduces costs.
At the core of the HPE AI software portfolio are HPE Machine Learning Data Management (MLDM) software and the HPE Machine Learning Development Environment (MLDE). These solutions are built on leading open-source tools: Pachyderm and Determined AI, respectively, and are certified with NVIDIA AI Enterprise.
HPE MLDM software is a flexible, scalable data pipelining and data versioning tool typically used for data preparation. The HPE MLDE is a secure, cost-effective, distributed training solution with hyperparameter optimization and experiment tracking. These software solutions are offered with a reference architecture that encompasses all stages of the AI workflow, from data preparation, model development, and training to deployment and monitoring.
Infrastructure-as-a-Service with HPE GreenLake
To meet modern demands, HPE GreenLake offers an automated, flexible, secure cloud experience for running and supporting AI/ML workloads. It not only accelerates operationalizing AI/ML, but also minimizes risks in data movement and seamlessly scales from development to production, expediting the time to value of data science workflows and ML projects.
HPE GreenLake empowers financial services organizations to drive business outcomes, achieve faster time to value, and reduce operational costs. This as-a-service delivery model combines cloud-like speed and agility with on-prem infrastructure, enabling you to focus on innovation and value delivery.
HPE GreenLake is uniquely positioned to accelerate FSI workloads, including:
- HPC Quantitative Financial applications
- Data Processing & Analytics
- Generative AI, including large language models with various Generative Pre-trained Transformer (GPT) models
These workloads enable firms to drive multiple complex business needs with converged HPC+AI solutions.
Break the Tape
The close partnership between HPE and NVIDIA has produced a solution that not only sets new benchmark records in financial services, but also promises to drive digital transformation and innovation across the industry. With unrivaled performance and efficiency, this dynamic duo is poised to revolutionize the financial services landscape, from HPC quantitative financial applications and data processing & analytics to generative AI encompassing large language models that are taking the AI world by storm. You can learn more about how HPE and NVIDIA’s collaboration can set new records in performance from the edge to the cloud.
Want to dive into the performance details? The full report is available to STAC members and guests (with registration): https://www.STACresearch.com/NVDA230721
Ron Javor is Master Product Manager at HPE, overseeing the Cray AI Supercomputing portfolio. Ron joined HPE in 2021 with over 20 years of product management and engineering experience spanning semiconductors and systems at Motorola, Intel, Broadcom, and SanDisk. Ron holds an MSEE from Texas A&M University and an MBA from Arizona State University.
Chuck Newman is a Performance Engineer with over 40 years at HPE, primarily supporting ISVs, customers, and Industry Marketing in HPC. The Financial Services Industry has been a primary focus domain for Chuck for more than fifteen years.
[1] STAC-A2.β2.GREEKS.TIME.WARM
[2] STAC-A2.β2.GREEKS.TIME.COLD
[3] STAC-A2.β2.GREEKS.10-100k-1260TIME.WARM
[4] STAC-A2.β2.GREEKS.10-100k-1260TIME.COLD
[5] STAC-A2.β2.GREEKS.MAX_ASSETS
[6] STAC-A2.β2.GREEKS.MAX_PATHS
[7] STAC-A2.β2.HPORTFOLIO.ENERG_EFF
HPE_Alliances
HPE Alliance Partners
