Compression Performance: 7-Zip, MagicRAR, WinRAR, WinZip

By Manuel Masiero, Achim Roos
published

We're taking a look at the the features, performance, and compression rates of four file archiving and compression tools: the free and open source 7-Zip, the heavyweight WinRAR, the old-school WinZip, and the lesser-known MagicRAR.

Hyper-Threading Comparison: Proprietary And ZIP

7-Zip posts the largest performance gains from Hyper-Threading, reflecting optimizations for parallelization. Going from four to eight threads massively cuts down compression time. 7-Zip is 28 percent faster with its own 7z format, and 33 percent faster with the ubiquitous ZIP format. WinZip also benefits from from having Hyper-Threading enabled, while WinRAR posts smaller performance gains. MagicRAR is single-threaded, and consequently gets no benefit with Hyper-Threading turned on. In fact, it ends up being somewhat slower with the feature enabled.

Manuel Masiero
77 Comments Comment from the forums
  • audi90
    One does not simply buy WinRAR...
    Reply
  • mayankleoboy1
    1. There is no difference between LZMA and LZM2 . Both are the same algorithm. The only difference is LZMA is limited to 2 threads. LZMA2 is much more threaded, but uses double the amount of RAM.

    2. PPMd is strictly for compressing text. It compresses text better than any other algo. But it is limited to 1 core only.

    3. WinRar 4.2 is much better threaded than previous versions.

    4.7z threading depends a lot on the type of file compressed. On large files, it can use 100% of any number of cores. For many small files, it generally uses only 1 complete core.
    Reply
  • mayankleoboy1
    7ZIP is even more impressive when you consider that the LZMA format was designed by one single person. And then the program 7ZIP was also coded by that single person only.

    Maybe contribute a few dollars to Igor Pavlov , the creator of 7Zip ?
    Reply
  • s3anister
    This is an interesting article, I was rather surprised by the overall poor performance of WinRAR in every aspect when compared to 7zip.
    Reply
  • belardo
    For reference, shouldn't the built in ZIP tool in the windows OS?
    Reply
  • ojas
    Hey i had written this in the Haswell preview, but i think Chris missed it, so i'm repeating it here, since it is related.

    Could we have an AES-256 encryption comparison between CPUs and/or archive managers?

    Like without encryption vs with encryption, encryption with and without OpenCL, etc.
    Reply
  • mayankleoboy1
    ^ 7 zip can use the Hardware based Intel AES-NI extensions.
    Reply
  • abbadon_34
    Nothing new here, 7zip > WinRar > WinZip for quite some time. Why the inclusion of MagicRAR is a mystery, maybe a paid (failed) review? I'd be interested in an examine of the Parity/Recovery option of WinRAR and others. While still far behind PAR2 (or even the shady ICE Ecc), it is an important feature in Archiving that deserves more attention.
    Reply
  • PreferLinux
    mayankleoboy11. There is no difference between LZMA and LZM2 . Both are the same algorithm. The only difference is LZMA is limited to 2 threads. LZMA2 is much more threaded, but uses double the amount of RAM.

    2. PPMd is strictly for compressing text. It compresses text better than any other algo. But it is limited to 1 core only.

    3. WinRar 4.2 is much better threaded than previous versions.

    4.7z threading depends a lot on the type of file compressed. On large files, it can use 100% of any number of cores. For many small files, it generally uses only 1 complete core.4. You mean the 7Z format rather than 7-Zip.

    I've seen 7-Zip, using the Zip format, hitting 100% CPU usage when archiving around 1500 – 2000 files, the vast majority of which (like >75%, if not >90%) were tiny, about half under 100 B and the other half between 1 kB and 4 kB. But with the same set of files I did a quick test, and using LZMA2 to 7z it was using 1 and a bit cores (going by my total CPU usage).
    Reply
  • LiviuTM
    Great article.
    Maybe you can add IZArc (http://www.izarc.org/) to the comparison.
    Reply