By Kevin Gildea, Vladimir Polin, and Keegan Sheedy, Intel Corporation
Intel® Xeon® 6 Processors Set the Pace
When the market closes, the real work begins. Portfolios shift, scenarios expand, and job schedulers fan out millions of pricing and risk tasks across your grid. Two dials matter every day: how much work you can push through before the opening bell, and how predictably you can do it as conditions change. Throughput without predictability misses SLAs; efficiency without scale misses the window.
When milliseconds matter and billions of trades stream across the financial exchanges, Intel Xeon 6 processors push the frontier of performance forward. In the past year, Intel Xeon 6 processors have set 26 world records on industry-standard STAC benchmarks, showcasing exceptional performance across quantitative finance workloads. Now we show the same strength in QuantLib, the open‑source foundation for real‑world pricing and risk models. Strong throughput for large batch runs, efficient execution for latency and license‑sensitive workflows, and consistent scaling as concurrency increases. This balance translates into faster time to insight and more accurate measures of risk during volatile sessions.
STAC and QuantLib: Industry Benchmarks and Open-Source Reality
Global banks, high-frequency traders and quant firms measure workload performance using both audited benchmarks and open-source frameworks. Intel Xeon 6 processors demonstrate leadership in both, closing the loop between audited industry-standard tests and open-source frameworks.
QuantLib
The QuantLib project is a well-established open-source framework for quantitative finance, with a 25-year history of providing robust tools for financial modeling and risk management. Recently, a benchmark suite has been introduced, providing a valuable resource for conducting evaluations and exploring potential improvements to the library in straightforward ways.
The benchmark in QuantLib v1.39 packages 87 workloads from the library’s quality assurance suite and runs them concurrently, measuring an overall benchmark throughput that a given hardware and software stack can achieve. In comparison to highly optimized benchmarks such as STAC-A2, where algorithms are separated from QA and may fully utilize hardware capabilities, the workloads in this benchmark use QA tests without optimizations. Internal repetition count varies from 1 to 100000 for different tests, and every test includes an initialization and a deinitialization. Additionally, benchmarks contribute differently to the performance score for different launch sizes. External repetition count varies with the benchmark suite launch size. For QuantLib v1.39 sizes are XXS=60, XS=120, S=240, M=480, L=960 or user defined.
STAC
STAC Research is the trusted voice for technology performance in capital markets. STAC improves technology discovery and assessment in the financial services industry through empirical research and rigorous benchmarking. The STAC Benchmark™ Council is a global community of over 500 financial institutions and 70+ technology vendors who collaborate to advance high-performance technologies in capital markets. STAC produces standardized benchmark tests that help financial firms assess how well their technology systems—including trading platforms, risk engines and data analytics stacks—perform under strict audit conditions across quantitative finance workloads.
The Results
Intel and Micron have joined forces to set new industry records on both the STAC-A2™ and STAC-M3™ benchmarks, demonstrating breakthrough performance and efficiency for financial services and capital markets.
On STAC-A2™, the joint solution with Intel Xeon 6 processors and Micron MRDIMMs delivered world-record speed for risk analytics and Monte Carlo simulations, achieving 35.2 milliseconds to market insight, over two times throughput, nearly 10 times faster cold-start, and a 28% gain in energy efficiency compared to previous bests.
For STAC-M3™, Intel and Micron posted record-setting results for high-performance tick database queries and time-series analytics. Powered by Intel Xeon 6 processors, Micron 9550 NVMe SSDs and high-capacity DDR5 memory, our solution completed the compute-intensive 100-user market statistics benchmark 36% faster than ever before, all the while using 62% fewer CPU cores than the previous record holder. These achievements directly benefit banks, hedge funds, trading firms and exchanges — enabling faster, more scalable analytics for algorithmic trading, risk management and regulatory compliance.
Discover how Intel and Micron are reshaping the frontier of financial infrastructure:
- 35.2 Milliseconds to Market Insight with Micron MRDIMM Memory and Intel Xeon 6 Processors.
- A Stacked Solution: Micron SSDs, Intel Xeon 6 Processors & Supermicro Servers Break 25 STAC-M3™ World Records.
In this new post, we’ll focus on our latest round of benchmarks with the QuantLib suite, showcasing further performance leadership. Across the QuantLib “Size L” suite, Intel Xeon 6972P processors (96 cores) deliver near-top throughput within 10% of 128-core systems. And at equal core count, Intel Xeon 6980P processors (128 cores) extend the lead, delivering the highest total throughput of all processors tested. These results translate to better autoscaling behavior, greater efficiency, and a greener footprint, all without giving up the throughput needed to succeed.
Figure 1: Normalized relative QuantLib v1.39 performance per vCPU
Figure 2: S-curve, QuantLib v1.39, Intel® Xeon® 6980P processors vs AMD EPYC™ 9755 processors, see notes for numbers
Figure 3: Normalized relative QuantLib v1.39 throughput per server
Note: For optimal performance, it is recommended to utilize vendor-specific compilers. Therefore, we employ the DPC++ compiler for Intel Xeon processor systems and the AOCC compiler for AMD EPYC processor systems.
From STAC to QuantLib, Intel Xeon 6 processors continue to demonstrate leadership across the workloads that financial institutions rely on every day. STAC shows Intel at the top of industry-standard tests, while QuantLib confirms that strength translates directly into the open-source models used for pricing portfolios, running risk overnight, and keeping pace with market volatility. For quant desks and risk teams, that means finishing calculations before market open, hitting intraday SLAs without overprovisioning, and controlling license costs. These results highlight Intel’s dedication to providing exceptional performance for the financial services workloads that matter most.
Notices & Disclaimers
Performance varies by use, configuration and other factors. Performance results are based on testing as of dates shown in configurations and may not reflect all publicly available updates. No product or component can be absolutely secure. Your costs and results may vary.
System Configurations:
- 1-node, 2x Intel(R) Xeon(R) 6972P, 96 cores, HT On, Turbo On, Total Memory 1536GB (24x64GB DDR5 8800 MT/s [8800 MT/s]), BIOS BHSDCRB1.IPC.3544.P80.2506250319, microcode 0x10003e0, 1x I210 Gigabit Network Connection, 1x 953.9G KXG60ZNV1T02 KIOXIA, Ubuntu 24.04 LTS, 6.8.0-71-generic, QuantLib v1.39, Intel(R) oneAPI DPC++/C++ Compiler 2025.2.0 (2025.2.0.20250605). Test by Intel Corporation as of 08/11/25.
- 1-node, 2x Intel(R) Xeon(R) 6980P, 128 cores, HT On, Turbo On, Total Memory 1536GB (24x64GB DDR5 8800 MT/s [8800 MT/s]), BIOS BHSDREL1.IPC.3544.P64.2505160241, microcode 0x10003d0, 1x MT28908 Family [ConnectX-6], 1x I210 Gigabit Network Connection, 1x 894.3G Micron_7450_MTFDKBG960TFR, Ubuntu 24.04 LTS, 6.8.0-71-generic, QuantLib v1.39, Intel(R) oneAPI DPC++/C++ Compiler 2025.2.0 (2025.2.0.20250605). Test by Intel Corporation as of 08/11/25.
- 1-node, 2x AMD EPYC 9755 128-Core Processor, 128 cores, SMT On, Turbo On, Total Memory 1536GB (24x64GB DDR5 6400 MT/s [6000 MT/s]), BIOS 1.3a, microcode 0xb00211e, 2x I350 Gigabit Network Connection, 1x 3.5T KIOXIA KCMYDRUG3T84, Ubuntu 24.04 LTS, 6.8.0-64-generic, QuantLib v1.39, AMD clang version 17.0.6 (CLANG: AOCC_5.0.0-Build#1377 2024_09_24). Test by Intel Corporation as of 08/11/25.
Build Command Lines:
- Intel(R) oneAPI DPC++/C++ Compiler 2025.2.0 (2025.2.0.20250605)
- BOOSTROOT=~/benchmarking/boost-1.88 cmake .. -G "Unix Makefiles" -D CMAKE_BUILD_TYPE=Release -DQL_ENABLE_PARALLEL_UNIT_TEST_RUNNER=1 -DCMAKE_CXX_COMPILER=icpx -DCMAKE_CXX_FLAGS="-O3 -ffast-math -D__FAST_MATH__=1 -fimf-use-svml=true:exp,pow,expl,erf -fimf-domain-exclusion=31 -fimf-precision=med -fno-math-errno -fno-finite-math-only -fhonor-nans -Wno-unused-command-line-argument -fno-associative-math -march=graniterapids" ; make -j; cd test-suite/
- AMD clang version 17.0.6 (CLANG: AOCC_5.0.0-Build#1377 2024_09_24):
- BOOSTROOT=~/benchmarking/boost-1.88 cmake .. -G "Unix Makefiles" -D CMAKE_BUILD_TYPE=Release -DQL_ENABLE_PARALLEL_UNIT_TEST_RUNNER=1 -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_CXX_FLAGS="-O3 -zopt -march=znver5 -fveclib=AMDLIBM -lamdlibm -Wno-unused-command-line-argument" ; make -j ; cd test-suite
S-curve data, QuantLib v1.39, Intel® Xeon® 6980P processors vs AMD EPYC™ 9755 processors:
testCalibrationTwoInstrumentSets-0.63,testSabrNormalVolatility-0.63,testCalibrationOneInstrumentSet-0.70,testCachedMarketValue-0.70,testWeightedModifiedBesselFunctions-0.75,testGammaFunction-0.78,testHalton-0.79,testSwaps-0.81,testFlatVolCalibration-0.81,testConsistency-0.82,testTimeDependentInterestRates-0.82,testFactorial-0.83,testDAXCalibration-0.83,testPiecewiseConstantInterpolation-0.84,testLinearInterpolation-0.84,testAndreasenHugePut-0.84,testAndreasenHugeCall-0.85,testCalibration-0.85,testLocalVolatility-0.86,testAndreasenHugeCallPut-0.86,testFlatForwardConsistency-0.86,testModifiedBesselFunctions-0.87,testVanillaEngines-0.87,testMersenneTwisterDiscrepancy-0.87,testMomentBasedGaussianPolynomial-0.87,testAmericanCallPutParity-0.87,testImpliedVol-0.87,testFdmHestonAmerican-0.88,testMonteCarloCalibration-0.89,testIsdaEngine-0.89,testArbitrageFree-0.90,testFdBSSwingOption-0.90,testLocalVolFromHestonModel-0.91,testVPPPricing-0.91,testSwaptionPricing-0.92,testSabrVols-0.92,testConvexMonotoneForwardConsistency-0.92,testGlobalBootstrap-0.92,testBermudanSwaption-0.93,testSpreadedCube-0.93,testMultiStepCoterminalSwapsAndSwaptions-0.93,testFdmHestonBarrierVsBlackScholes-0.94,testCachedValues-0.94,testBarrierPricingViaHestonLocalVol-0.94,testAmericanOption-0.95,testVarianceGamma-0.95,testFdValues-0.95,testFdAmericanGreeks-0.96,testMcEngines-0.96,testCachedHullWhite2-0.96,testImpliedHazardRate-0.96,testMultiStepCmSwapsAndSwaptions-0.97,testCachedHullWhiteFixedReversion-0.97,testCallPutParity-0.97,testAnalyticVsMCPricing-0.97,testAnalyticAndMcVsJumpDiffusion-0.97,testGauss-0.99,testCachedG2Values-0.99,testDAXCalibration-1.00,testCmsSwap-1.03,testResults-1.03,testCouponPricing-1.03,testQdEngineStandardExample-1.03,testExtOUJumpVanillaEngine-1.04,testGammaValues-1.06,testBootstrapRegression-1.09,testBootstrapWithArithmeticAverage-1.11,testMultiDimRegression-1.15,testUp-1.16,testBond-1.16,testCeiling-1.16,testFloor-1.16,testExtOUJumpSwingOption-1.18,testDown-1.18,testClosest-1.19,testBaseBootstrap-1.19,testKlugeExtOUSpreadOption-1.26,testIncrementalStatistics-1.28,testParity-1.32,testNonCentralChiSquaredSumOfNodes-1.36,testSquareRootCLVVanillaPricing-1.38,testSquareRootCLVMappingFunction-1.46,testFdAmerican-1.46,testHestonFokkerPlanckFwdEquation-1.53,testFdBarrierVsCached-1.55,testNonCentralChiSquared-2.55,testConversions-4.01