Performance

Zstandard is designed for real-time compression scenarios, offering excellent speed while maintaining competitive compression ratios. Understanding its performance characteristics helps you optimize for your use case.

Benchmark Results

Desktop Performance (Core i7-9700K @ 4.9GHz)

Tested on Ubuntu 24.04 using the Silesia compression corpus:

Compressor	Ratio	Compression	Decompression
zstd 1.5.7 -1	2.896	510 MB/s	1550 MB/s
brotli 1.1.0 -1	2.883	290 MB/s	425 MB/s
zlib 1.3.1 -1	2.743	105 MB/s	390 MB/s
zstd 1.5.7 —fast=1	2.439	545 MB/s	1850 MB/s
quicklz 1.5.0 -1	2.238	520 MB/s	750 MB/s
zstd 1.5.7 —fast=4	2.146	665 MB/s	2050 MB/s
lzo1x 2.10 -1	2.106	650 MB/s	780 MB/s
lz4 1.10.0	2.101	675 MB/s	3850 MB/s
snappy 1.2.1	2.089	520 MB/s	1500 MB/s
lzf 3.6 -1	2.077	410 MB/s	820 MB/s

From README.md:42-53

Zstandard at level 1 provides better compression ratio than zlib at level 1, while being nearly 5x faster for compression and 4x faster for decompression.

Server Performance (Core i7-6700K @ 4.0GHz)

Compression Speed vs Ratio characteristics:

Low levels (1-5): 100-500 MB/s compression
Medium levels (6-12): 20-100 MB/s compression
High levels (13-19): 5-20 MB/s compression
Ultra levels (20-22): 1-5 MB/s compression

From README.md:67-79

Decompression Speed Consistency

Key characteristic: Decompression speed is preserved and remains roughly the same at all settings, a property shared by most LZ compression algorithms. From README.md:64-65

Decompression typically runs at 400-800 MB/s regardless of compression level, making Zstandard excellent for scenarios where data is compressed once but decompressed many times.

Memory Usage

Compression Memory by Level

Memory requirements increase with compression level:

Level  1 : Compression Mem =   434 KB
Level  3 : Compression Mem =  1335 KB
Level  5 : Compression Mem =  2367 KB
Level  7 : Compression Mem =  3712 KB
Level  9 : Compression Mem =  6574 KB
Level 10 : Compression Mem = 13086 KB
Level 12 : Compression Mem = 21277 KB

From examples/streaming_memory_usage.c:126-132

Decompression Memory

Decompression memory remains relatively constant:

Decompression Mem = 75 KB (for standard window sizes)

Decompression memory depends on the window size used during compression, not the compression level.

From examples/streaming_memory_usage.c:126

Buffer Size Recommendations

For optimal performance with streaming API:

size_t ZSTD_CStreamInSize(void);   // Returns 128 KB (131,072 bytes)
size_t ZSTD_CStreamOutSize(void);  // Returns 128 KB (131,072 bytes)
size_t ZSTD_DStreamInSize(void);   // Recommended input size
size_t ZSTD_DStreamOutSize(void);  // Recommended output size

From lib/zstd.h:842-843,950-951

Multithreading Performance

Enabling Multithreading

ZSTD_CCtx_setParameter(cctx, ZSTD_c_nbWorkers, nbThreads);

Where nbThreads:

0: Single-threaded mode (default)
1+: Specific number of threads
0 with env: Detected CPU core count

From lib/zstd.h:471-479

Multithreading Trade-offs

Benefits:

Improved compression speed on multi-core systems
Scales well with available cores

Costs:

Increased memory usage
Slight reduction in compression ratio

Multithreading increases memory usage. Each worker thread requires additional memory for its own compression context.

Job Size Control

ZSTD_CCtx_setParameter(cctx, ZSTD_c_jobSize, size);

Controls the size of compression jobs when using multiple threads. Default is 0 (automatic based on compression parameters). From lib/zstd.h:480-484

Long Distance Matching

Long distance matching mode improves compression ratio for files with matches at large distances (up to 128 MB).

Enabling Long Distance Mode

zstd --long file.txt
zstd --long=27 file.txt  # Specify window log

From programs/README.md:199

Performance Impact

Example with tar of multiple clang versions:

Method	Ratio	Compression	Decompression
`zstd -1`	5.065	284.8 MB/s	759.3 MB/s
`zstd -5`	5.826	124.9 MB/s	674.0 MB/s
`zstd -10`	6.504	29.5 MB/s	771.3 MB/s
`zstd -1 --long`	17.426	220.6 MB/s	1638.4 MB/s
`zstd -5 --long`	19.661	165.5 MB/s	1530.6 MB/s
`zstd -10 --long`	21.949	75.6 MB/s	1632.6 MB/s

From programs/README.md:303-310

On files with long distance matches, —long mode can triple compression ratio while doubling decompression speed.

Memory Requirements

Enabling long distance mode:

Sets window size to 128 MB
Increases memory usage for both compressor and decompressor
Default window log is 27 (128 MB)

From programs/README.md:286-287

Dictionary Compression Performance

For small data (e.g., 1KB records), dictionary training provides:

Better compression ratios: Dramatically improved on small data
Faster compression: Compared to no dictionary
Faster decompression: Compared to no dictionary

From README.md:100

Dictionary gains are most effective in the first few KB of data. After that, the algorithm uses previously decoded content.

Tuning for Performance

Compression Speed Priority

For maximum compression speed:

zstd --fast=4 file.txt        # Fastest mode
zstd -1 file.txt              # Fast mode with better ratio

ZSTD_CCtx_setParameter(cctx, ZSTD_c_compressionLevel, 1);

Compression Ratio Priority

For maximum compression ratio:

zstd --ultra -22 file.txt     # Maximum compression
zstd -19 file.txt             # High compression (no --ultra needed)

ZSTD_CCtx_setParameter(cctx, ZSTD_c_compressionLevel, 19);

Balanced Performance

For most use cases:

zstd file.txt                 # Default level 3
zstd -5 file.txt              # Good balance

Advanced Performance Parameters

Window Log

Controls maximum back-reference distance:

ZSTD_CCtx_setParameter(cctx, ZSTD_c_windowLog, value);

Larger values: More compression, more memory
Range: 10 to 31 (64-bit) or 30 (32-bit)

From lib/zstd.h:368-375

Hash Log

Controls probe table size:

ZSTD_CCtx_setParameter(cctx, ZSTD_c_hashLog, value);

Larger values: Better compression for fast strategies, more memory
Range: 6 to 30

From lib/zstd.h:376-381

Strategy Selection

ZSTD_CCtx_setParameter(cctx, ZSTD_c_strategy, ZSTD_fast);

Available strategies (from fast to strong):

ZSTD_fast: Fastest
ZSTD_dfast: Fast with better ratio
ZSTD_greedy: Balanced
ZSTD_lazy, ZSTD_lazy2: Good ratio
ZSTD_btlazy2: Better ratio
ZSTD_btopt: Strong compression
ZSTD_btultra, ZSTD_btultra2: Maximum compression

From lib/zstd.h:336-347

Benchmarking

The CLI includes a built-in benchmark mode:

# Benchmark default level
zstd -b file.txt

# Benchmark range of levels
zstd -b1 -e9 file.txt

# Benchmark with specific time per level
zstd -b5 -i5 file.txt  # 5 seconds per test

From programs/README.md:250-252 Benchmark measures:

Compression ratio
Compression speed (MB/s)
Decompression speed (MB/s)

Platform-Specific Optimizations

BMI2 Instructions

Zstandard can leverage BMI2 CPU instructions for better performance:

Automatic detection: Enabled by default on supported platforms
Build-time: STATIC_BMI2=1 or DYNAMIC_BMI2=1
Performance gain: Especially on decoder side

From lib/README.md:167-183

SIMD Optimizations

The library includes SIMD optimizations for:

Huffman decoding
Sequence processing
Memory operations

These are automatically enabled based on compiler and platform detection.

Get Started

Core Concepts

Guides

Integration

Benchmark Results

Desktop Performance (Core i7-9700K @ 4.9GHz)

Server Performance (Core i7-6700K @ 4.0GHz)

Decompression Speed Consistency

Memory Usage

Compression Memory by Level

Decompression Memory

Buffer Size Recommendations

Multithreading Performance

Enabling Multithreading

Multithreading Trade-offs

Job Size Control

Long Distance Matching

Enabling Long Distance Mode

Performance Impact

Memory Requirements

Dictionary Compression Performance

Tuning for Performance

Compression Speed Priority

Compression Ratio Priority

Balanced Performance

Advanced Performance Parameters

Window Log

Hash Log

Strategy Selection

Benchmarking

Platform-Specific Optimizations

BMI2 Instructions

SIMD Optimizations

Build docs developers (and LLMs) love

Get Started

Core Concepts

Guides

Integration

​Benchmark Results

​Desktop Performance (Core i7-9700K @ 4.9GHz)

​Server Performance (Core i7-6700K @ 4.0GHz)

​Decompression Speed Consistency

​Memory Usage

​Compression Memory by Level

​Decompression Memory

​Buffer Size Recommendations

​Multithreading Performance

​Enabling Multithreading

​Multithreading Trade-offs

​Job Size Control

​Long Distance Matching

​Enabling Long Distance Mode

​Performance Impact

​Memory Requirements

​Dictionary Compression Performance

​Tuning for Performance

​Compression Speed Priority

​Compression Ratio Priority

​Balanced Performance

​Advanced Performance Parameters

​Window Log

​Hash Log

​Strategy Selection

​Benchmarking

​Platform-Specific Optimizations

​BMI2 Instructions

​SIMD Optimizations

Build docs developers (and LLMs) love

Benchmark Results

Desktop Performance (Core i7-9700K @ 4.9GHz)

Server Performance (Core i7-6700K @ 4.0GHz)

Decompression Speed Consistency

Memory Usage

Compression Memory by Level

Decompression Memory

Buffer Size Recommendations

Multithreading Performance

Enabling Multithreading

Multithreading Trade-offs

Job Size Control

Long Distance Matching

Enabling Long Distance Mode

Performance Impact

Memory Requirements

Dictionary Compression Performance

Tuning for Performance

Compression Speed Priority

Compression Ratio Priority

Balanced Performance

Advanced Performance Parameters

Window Log

Hash Log

Strategy Selection

Benchmarking

Platform-Specific Optimizations

BMI2 Instructions

SIMD Optimizations