HPE Intel and Micron Financial Risk Analytics Performance on STAC A2 Benchmarks Delivering Record Throughput and Space Efficiency with Liquid Cooled Intel Xeon 6 Processors
One of the major operational challenges for financial services institutions is performing computationally intensive risk analytics with lower energy consumption and a smaller data center footprint. In scale quantitative models, the key constraint often changes from processing speed to data throughput. Large scale Monte Carlo simulations rely heavily on memory bandwidth in such tasks like options Greek calculation which is essential to derivatives valuation and regulatory capital requirements. An independent, auditor STAC Research Benchmark result shows how hardware synergy overcomes the physical constraints.
 |
| Cold Runs Across Platforms |
 |
| Energy and Space Comparison Across Generations |
 |
| With MRDIMM vs. Without MRDIMM (Same Xeon 6980P) |
 |
| Generational Comparison |
The test utilized the HPE ProLiant XD230 1U server, which can be detailed at hpe.com, running the server with dual Intel Xeon 6980P processors with Micron 8800 MT/s DDR5 MRDIMMs, available at micron.com. The combined hardware delivered record cold runs of the baseline problem size on the STAC A2 benchmark. This solution was tested using an Intel optimized implementation verified independently by STAC with the same strict methodological principles across industry wide benchmarks.
Compared to the prior generation platform with Intel Xeon Platinum 8592+ processors, the modern system shows significant gain in all aspects. The server generated 2.38x portfolio throughput, achieving 100.8 options/sec, a new industry best. Calculation speed for the baseline problem size in cold run and warm run are 10.42x faster and 1.62x faster, respectively. For large problem size, calculation speed in cold run and warm run were 2.04x and 2.07x faster, respectively. Trading companies can gain time to insight with this accelerated performance, consolidate hardware footprint.
The system also achieved 1.58x energy efficiency improvement, 231,271 options/kWh, and 3.26x improvement on space efficiency, 133.8 options/hr/in^3 for 1U server. These metrics are vital for collocation facilities and enterprise data centers where rack space and power is limited.
This record breaking performance largely depends on the memory subsystem. In the test, 24 x 64 GB Micron 8800 MT/s DDR5 MRDIMMs were used which provided 1.5 TB system memory and had 12 channels per socket. As Monte Carlo risk calculations generate and correlate data continuously during the path generation step, high bandwidth memory (especially multiplexed rankDIMMs) greatly alleviates data transfer bottleneck. With MRDIMMs, the same system with dual Intel Xeon 6980P processors shows 1.08x better portfolio throughput and up to 23% calculation speed increase in large problem size, and up to 1.29x and 1.65x energy and space efficiency increase, respectively, compared with similar servers using standard DDR5 RDIMMs.
This dual processor system runs on dual 128 core Intel Xeon 6980P processors which have 504 MB L3 cache and Intel AVX 512 to enhance the heavy vector computation associated with financial modeling. This solution used STAC A2 Pack for oneAPI Rev R including Intel oneAPI Base Tool Kit and HPC Tool Kit for the software part.
Deploying dual 128 core processors in a 1U server required the innovative thermal management approach. The HPE ProLiant XD230 supports a liquid cooling design for processors via a coolant distribution unit that connects to the data center's central water supply system and fan cooling for peripheral components. This combined cooling architecture enabled the system to achieve desired thermal margin under full load and resulted in 1.23x better energy efficiency compared to an air cooled solution running the same processors, direct benefit for energy efficiency.
For data center designers considering high density financial workloads, the STAC A2 result clearly suggests that a synergy between high bandwidth memory, liquid cooled CPUs and dense 1U chassis design is an effective way to save power and increase computing throughput while maximizing rack space.
