StockFish 7 and time control

A "quick" comparison on the ELO increase when allowing more time in Stockfish 7 self-plays (ratios are tried from 1:1 upto 16:1)

Number of games were 100 in every cases, the error bars are related to 95% confidence interval. The reference timing was: 10/15 (10 min + 15 sec), the games were running on 4 cores with "Ponder Off" and with 2048 MB hash table. The opening book was used at 2 ply-depth to provide the necessary variety.

2022. December 31. 

Opening book development :

The distribution of the positional entries in the actual version(s) of the opening book (2026 June 1., version: Alpha 56) are below. (The primary actual goal is still to increase its strength, that is why the bit stagnated position counts during the last few months). (additional evaluations: 2026-06-05) 

The books are available in polyglot (.bin) format:

https://www.patreon.com/analysis4chess

The Alpha versions of the book are intended to be wide in repertoire, with uniform move distribution and with slow increase in depth. The upcoming Beta versions will be more narrow, with weighted distributions of the moves.

2026. June 6.

Stockfish NNUE (16.1, 17.0, 17.1, 18.0) Performance on Core 2 Quad Q9550 CPU: 

SSE2 vs SSSE3

Abstract

This report compares the performance of SSE2 and SSSE3 builds of Stockfish NNUE versions 16.1, 17.0, 17.1, and 18.0 on an Intel Core 2 Quad Q9550 system. Performance was evaluated using the official Stockfish benchmark (bench command), with each configuration executed 100 times sequentially and the resulting Nodes Per Second (NPS) values were averaged. The objective was to quantify the performance benefit provided by SSSE3 on this older Yorkfield-based processor architecture. The observed gains are plausibly related to the additional SIMD capabilities available when targeting SSSE3, particularly byte-shuffle operations (pshufb), although no instruction-level analysis was performed.

Test Platform

Hardware

• CPU: Intel Core 2 Quad Q9550 @ 2.84 GHz

• Memory: 4 GB DDR2-800 (2×2 GB, dual-channel)

• Measured memory bandwidth: approximately 3.85 GB/s (memtest86+ v7.20)

Software

• Operating system: GNU/Linux (x86-64)

• Compiler: GCC 14.2.0

• Build configuration: Stockfish profile-build (PGO)

Benchmark Methodology

For each Stockfish version, separate SSE2 and SSSE3 binaries were compiled from source using the standard Stockfish build system and architecture-specific targets.

Benchmark execution parameters:

• Command: bench

• Hash: 2048 MB

• Threads: 4

• Depth: 24

Each binary was benchmarked 100 times. Average NPS values are reported below.

The benchmark used the default Stockfish bench position set supplied with each version. As these position sets may differ between releases, cross-version comparisons should be interpreted with caution.

Results

Version    SSE2 NPS    SSSE3 NPS     Gain

 16.1      1.049 M     1.383 M     +31.8%

 17.0      1.036 M     1.343 M     +29.7%

 17.1      1.232 M     1.572 M     +27.6%

 18.0      2.049 M     2.419 M     +18.0%

Discussion

Across all tested versions, the SSSE3 binaries achieved higher NPS than their SSE2 counterparts.

The measured performance advantage ranged from 31.8% to 18.0%, demonstrating that SSSE3 remains beneficial on the Core 2 architecture for modern NNUE-enabled Stockfish releases.

Although the relative advantage decreases in newer versions, the absolute gain remains broadly consistent, amounting to approximately 300k–370k additional nodes per second across the tested releases.

Because benchmark position sets and search behavior differ between Stockfish versions, the reported NPS values should primarily be used for within-version SSE2 versus SSSE3 comparisons rather than direct comparisons between releases.

Conclusion

On an Intel Core 2 Quad Q9550, SSSE3 builds of Stockfish NNUE provide a substantial and consistent performance advantage over SSE2 builds.

The observed improvement ranges from approximately 18% to 32%, depending on the Stockfish version tested. Users of older systems supporting SSSE3 can therefore expect a meaningful increase in benchmark throughput when using SSSE3 binaries.

Build Notes

• Source builds performed with GCC 14.2.0

• Profile-guided optimization enabled via Stockfish profile-build

• No additional optimization flags beyond architecture selection

• SSE2 and SSSE3 binaries compiled separately

• Results represent averages of 100 benchmark executions per configuration

2026. June 6.