Understanding Linpack GFlops & Residual Values in IntelBurnTest
When IntelBurnTest executes, it prints a grid of technical figures across each pass: Time (s), Speed (GFlops), Result (Residual), and Residual(norm). While many users simply look for the final "Test Successful" confirmation, understanding what these numbers actually mean provides deeper insights into your processor's performance and silicon health.
In this article, we explain the mathematics behind Linpack matrix factorization, why matching residuals guarantee mathematical precision, and what sudden drops in GFlops signify.
What is the Linpack Residual?
At its core, Intel's Linpack benchmark solves a dense system of linear algebraic equations:
Here, $A$ is a gigantic matrix of random numbers generated according to the chosen memory size, and $b$ is a vector. The algorithm solves for $x$ using high-performance LU factorization.
Once the solution vector $x$ is computed, Linpack multiplies $A$ and $x$ back together to calculate the residual:
Because the exact same initial matrix seed and dimensions are computed on every pass, an electrically stable processor operating without arithmetic bit-flips will produce the exact same floating-point residual number on every single pass (for instance, 3.481943e-02 across all 10 runs).
Crucial Stability Rule:
If the residual number shifts even by the tiniest fraction on any pass (e.g. 3.481943e-02 changes to 3.481949e-02), the CPU made a calculation error. This is a silent computational error that would corrupt file downloads, database records, or 3D renders without crashing Windows.
What Does GFlops Measure?
GFlops stands for Giga Floating-Point Operations Per Second (one billion floating-point operations per second). In IntelBurnTest, GFlops indicates raw computational throughput:
- Higher GFlops = Higher Processing Speed: Overclocking CPU frequency directly increases GFlops.
- Memory Speed Impact: Because Linpack constantly streams dense matrices from RAM, faster memory bandwidth (e.g., DDR5 6400+ vs DDR4 3200) significantly boosts overall GFlops scores.
Diagnosing GFlops Variance Across Runs
Healthy Benchmark Run (Consistent GFlops)
Run 1: 345.2 GFlops
Run 2: 344.8 GFlops
Run 3: 345.1 GFlops
Deviation is under 1%. The CPU is maintaining full boost clocks without throttling.
Throttling Benchmark Run (Dropping GFlops)
Run 1: 345.2 GFlops
Run 2: 310.5 GFlops
Run 3: 265.1 GFlops
Significant drops reveal that VRM power limits or thermal throttling are cutting clock speeds.
How to Maximize Linpack GFlops & Stability
- Close Background Software: Anti-virus scans, browser tabs, and RGB software steal CPU cycles and cause GFlops fluctuations.
- Enable High Performance Power Plan: Prevents Windows from parking cores during matrix initialization.
- Ensure Sufficient Memory Allocation: Testing on Very High or Maximum provides more accurate Linpack GFlops scaling by saturating the processor's AVX execution pipelines.