Lossless data compression software benchmarks / comparisons
Maximum Compression's goal is to show the maximum achievable data compression ratio for
several filetypes (text, executable, jpeg etc). The best programs for every filetype
are compared in a table indicating compression ratios and switches/options used to achieve
that compression (SFC). Every program will only be listed once (with
the switches yielding the highest possible compression ratio for that test/file). I prefer command
line (console) compression programs over GUI ones.
If you are looking for more realistic data like efficiency and (de)compression speed of lossless compression
software have a look at the Multiple file compression test benchmarks (MFC).
The MFC benchmark lists performance comparisons of nearly all known archivers. The MFC test set
consists of dozens of different files types (totaling 510 files, 300+ MB).
If you know of an (experimental) program not listed here or know of a switch combination yielding
better results than the one listed please let me know!. Also general improvement suggestions are welcome,
simply write an entry in the Guestbook (temporary unavailable).
- 28-November-2011 : Added SCM 0.0.1b, XP v5 and Crush 0.01. SCM is a good compressor, but is still buggy (e.g.
crashes on JPG file with -o option) and options given seem to have no effect on compression level.
- 19-November-2011 : Added WinRAR 4.1b3, 7-zip 9.25a, NanoZip 0.09a, BSC 3.0.0, Comprox 0.3.0 and PackJPG 2.5.
Because of further multi-threading optimizations, Nanozip in default mode, makes a nice jump in
compression efficiency and
closes the gap to number one, FreeARC. Comprox crashes on the PDF file when using a blocksize < 5 MB and
crashed on the MFC test when using a blocksize < 17 MB :-(
- 25-June-2011 : Added ZPAQ 2.05, PPMX 0.07, DACT 0.8.42.
- 19-June-2011 : Added WinRAR 4.01, 7-zip 9.22, BCM 0.12, BSC 2.7.0 and UPX 3.07. WinRAR has a small improvement in efficiency as
decompression speed was improved in version 4 of WinRAR.
- 03-November-2010 : The guestbook in down for maintenance (you will still be able to read it, but posting is not possible).
- 23-July-2010 : Moved the location of the guestbook, after a huge spam(bot) attack.
- 6-July-2010 : Added FreeArc 0.666, NanoZip 0.08a, 7-ZIP 9.15, PackJPG 2.4, Etincelle RC2,
YZX 0.04, BSC 2.2.5, UPX 3.05, BCM 0.11 and CSC 3.2a6. PackJPG 2.4 takes 3th place in the
JPEG compression test from WinZip.
As expected, the new versions of both FreeARC and NanoZip keep fighting for the top places in the
most efficient compressor ranking.
- 10-April-2010 : Added Etincelle Beta 4, ULZ 0.02, BSC 1.0.2, FlashZIP 0.099b8 and ZPAQ 1.10.
- 31-December-2009 : Added Stuffit 14 and Squeez 5.63. As in previous Stuffit tests I selected the switches yielding best bit-perfect compression.
BMP compression has significantly improved, Stuffit takes 2nd place on both BMP and JPG test.
An interesting observation : The decompression times are more then twice the compression time for all MFC switch combinations I tested.
- 25-December-2009 : Added WinZIP 14, ZPAQ 1.09 and ZHuff 0.02.
- 23-December-2009 : Added FreeArc 0.60, WinRAR 3.91, 7-ZIP 9.10, NanoZip 0.07a, UPX 3.04, BCM 0.10, Packet 0.91b,
CSC 3.1, RINGS 1.6, LZTurbo 0.95, QPress 0.38b, FlashZIP 0.99b5, BWMonstr 0.02 (MFC todo) and LZP2. FreeArc keeps improving as version 0.60 shows improved compression speed (and in some cases
even improved compression ratio) compared to the 0.51 version. FreeArc again sets a new performance record in the
efficiency ranking!. WinRAR also made a big jump in efficiency as
compression speed was greatly improved by adding better multi-core support (introduced in WinRAR 3.90). It now takes a clear 3th
place in efficiency ranking, behind FreeARC and NanoZIP. LZP2 fails to qualify for listing in the MFC charts as the compressed
filesize produced by LZP2 is too big (> 1/2 org size).
- 01-May-2009 : Added PAQ8px, FreeArc 0.51, Hook 1.4, BMF 2.0 and LZBW1 0.8.
PAQ8px has slightly improved compression ratio compared to PAQ8P, and takes first place in both
SFC benchmark!. This latest PAQ8 is
based on PAQ8P and includes changes to the used models (new PBM, 1-bit BMP, nestModel and modified word,
WAV and JPG model). It is about 30% slower then it predecessor. FreeArc 0.51 is (still) performance
king and takes place 1 through 4 in the efficiency ranking!.
- 18-April-2009 : Added FreeArc 0.50a (April 15 2009), ASH 07, LZTurbo 0.94, LZP2 and M1 0.3b-1.
- 01-April-2009 : Added Squeez 5.61 and retested more old programs on the Q6600 (including PAQ8P, WinACE 2.69).
- 30-March-2009 : Added WinRK 3.1.2 and Slug X. WinRK 3.1.2 has a new Audio codec and better
SMP handling, it takes 2nd place in both the MFC and
SFC benchmark behind PAQ8P.
Unfortunately WinRK uses more memory then specified (caused by multi-threading?), when compressing
in High and Maximum mode it even uses (much) more memory then the 800MB allowed for the MFC test. I decided
to list the results anyway as both WinRK and PAQ8 show the worlds best possible compression. To make
it a fair comparison I will rerun PAQ8P in -8 mode. The two mentioned program are so far ahead of
the competition that both single and multiple file ranking will not change.
Update: PAQ8P in -8 mode fails to compress the MFC, it gives an out of memory after about 2 hours
of compression (I have 4 GB installed of which about 3.25 GB are available in WinXP 32-bit...).
- 28-March-2009 : Added WinZIP 12, PIM 2.90, FlashZip 0.94 and retested more old
programs. WinZIP 12.0 now follows the steps of PAQ8 and STUFFIT and includes a
lossless JPG (re)compressing algorithm. The compression ratio of WinZIP 12, using the 'Optimize
for best compression' switch, made a huge jump and is now on par with WinRAR and 7-Zip!
- 21-March-2009 : Added 7-Zip 4.65, QPress 0.35b and retested more old programs on the Q6600.
- 20-March-2009 : Added ZPAQ 1.00, BWMonstr 0.01, FlashZip 0.93b, BCM 0.07 and Ultra 7z Optimizer 0.05.
- 14-March-2009 : Added WinRAR 3.80, ZPAQ 0.09, BWMonstr 0.00, ASH 0.6a, FlashZip 0.93a
and again retested some old programs on the Q6600.
- 08-March-2009 : Added NanoZIP 0.06a, ZPAQ 0.07, BCM 0.05, BIT 0.7, BALZ 1.15, LZPM 0.16,
PPMX 0.05, ARJ 2.85, ARJ32 3.15 and retested some old programs on the Q6600 for
ZPAQ is the latest project of Matt Mahoney. The idea is to develop an open standard for highly
compressed data, sort of like deflate, but with a flexible paq-like algorithm. Once the standard
is finalized, developers can continue to discover new compression algorithms without breaking
archive compatibility with existing decompressors.
- 28-Februari-2009 : After some testing the speedup of single threaded programs seems
to be around 2.6 for the newer programs, the older programs gain a little less from the move to
the Q6600 because at the time they got compiled the compiler didn't know about the nice functions
the current processors have. PKzip seems to suffer the most with a speedup of only 1.5(!), and a
increased decompression time compared to the xp2800. In case of programs like PKzip, AIN I think they
are extremely optimised for the processors and disk buffer sizes which where mainstream 10 years
ago. Because of this those programs will hardly take advantage of the new architecture of the current
processors but only in the increased clock frequency.
So the formula used for the initial time values is: Time_new = 3 + (Time_old/2.6). One other update
is the MFC efficiency rating. This is now based on combined compression + decompression time
instead of just compression time.
- 21-Februari-2009 : Hi all, I'm back ;-) There will be a change to the MFC test, as
the aging AMD 2800xp had some technical problems, I thought it was about
time to upgrade the MFC testing to a new faster PC. I will use my Quadcore Q6600 @ 3.2 GHz for MFC testing
from now on. As the timings of the new Q6600 will of course be a lot faster then the 2800xp ones I need
to do some conversions. As it will take ages to rerun all MFC tests, I will convert the current times to an
approximate new time e.g. (time_new = 3 + (Time_old/3.28)) and add a note it's an approximation. Then I
will benchmark all current top programs in speed/compression/efficiency on the new PC to get the correct
timings. I think this is the way to go.
- 28-September-2008 : Added PAQ8P, 7-Zip 4.60b, FreeARC 0.50a (June 23 2008),
Tornado 0.4a, M1 0.1a, BZP 0.3, NanoZIP 0.04a, Blizzard 0.24b and WinRAR 3.80b5 (MFC to do
for WinRAR and 7-Zip).
PAQ8P manages to squeeze out an additional 12 KB from the BMP file, further increasing the
gap to the number 2 in the SFC
benchmark; newcomer NanoZIP takes 6th place in SFC!. In the
MFC benchmark PAQ8 now takes
a huge lead over WinRK 3.0.3, but WinRK 3.1.2 is on the todo list to be tested. To be continued...
- 27-September-2008 : Added CMM4 0.2b, BALZ 1.13, BMA 1.35b, Packet 0.90b,
FlashZip 0.91b, BIT 0.3, Tornado 0.4a and M99 2.2.1. CMM4 takes 10th place in
the SFC benchmark.
- 17-May-2008 : Added 7-Zip 4.58b, BALZ 1.07, BMA 1.34b, SLUG 1.27b, FreeARC 0.50a
(May 15 2008), Stuffit 12.0, UPX 3.03, QuickLZ 1.40b9 and Packet 0.01.
- 27-April-2008 : Added RINGS 1.5c, CMM4 0.1e, RZM 0.07h, BALZ 1.04, SLUG 1.24, CCM 130c,
FreeARC 0.50a (Apr 6 2008) and finished MFC test of PAQ8O10 and Ocamyd 1.66test1. PAQ8O10 still contains
a bug; after some point in the MFC compression all remaining files get the next error/warning 'Transform
fails at 0, input=xx decoded=xx, skipping..' where xx are different numbers. Compression seems to be
negatively affected by this (+850 KB over the latest stable o8 version). The latest FreeARC 0.50a again
sets a new record in compression efficiency.
Currently busy testing Stuffit 12.0, Squeez 5.61.
- 22-April-2008 : Added PAQ8O10, BZIP2 1.0.5, Ocamyd 1.66test1,
RINGS 1.4c, CMM4 0.1d, RZM 0.07e, LZTurbo 0.92, BruteCM 0.1d, BALZ 1.03, SBC 0.970r3
and Tornado 0.4. It's now also possible to sort the MFC benchmark on
- 02-March-2008 : Added PAQ8O9, FreeArc 0.50a, QuickLZ 1.40b6, RINGS 1.0, CMM3 07/02/2008,
SR3a, RZM 0.04, LZTurbo 0.9, FlashZIP 0.03, LZXQ 0.4, LZW 0.2, LZRW1 and BruteCM 0.1a.
PAQ8O9 fails to compress the MFC testset. It seems to be caught in an endless loop while trying to
compress a certain file. The MFC results of FreeARC 0.50a are about 4MB worse for all switch
combinations tested compared to FreeARC 0.40. In MFC, the compressed filesize produced by LZRW1
is too big to qualify for listing in the MFC charts.
- 27-January-2008 : Added CCM 1.30, FreeArc 0.40, QuickLZ 1.40b5, RINGS 0.3, WinTurtle 1.60b,
LZPM 0.15, CMM2 13/01/2008, FlashZIP 0.02 and LZXQ 0.1.
- 18-December-2007 : Added 7-Zip 4.57, HOOK 1.3, FreeArc 0.40pre4, CMM2, LPAQ8,
LZPM 0.13, BIT 0.1, Symbra 0.2, LCSSR 0.2, UPX 3.02 and Stuffit 12.0 beta. For Stuffit SFC I selected
the switches yielding best bit-perfect compression, for MFC I disabled PDF recompression as
it appears to be 'pixel-perfect' for most PDF files. It takes 2nd place on both BMP and JPG test.
- 02-December-2007 : Added PAQ8O8, WinACE 2.69, CCM 1.26b, FreeArc 0.40pre3, QuickLZ 1.40,
LZC 0.08, SR3, RINGS 0.2, BCArchive 1.08.7, WinTurtle 1.40b, LPAQ7, LZPM 0.12, HOOK 1.2 and tested
the new Stuffit 12.0 beta (todo SFC WinTurtle).
The latest PAQ8O8 again features an improved JPG model and now takes first place in the
JPEG test. It still keeps first
place in both the SFC
benchmark. FreeARC now takes place 1, 2 and 3 (!) in the most
efficient compressor ranking, it
has very good compression and is still relatively fast.
The tested Stuffit 12.0 is an early beta version, of Stuffit for windows. It includes a lot of changes
to the compression code, the most important being recompression techniques for several new file types
including MP3 and PDF!. Stuffit 12.0 would take 2nd in both the JPG and BMP test, and 1st
in the PDF one (it compresses the file to under 3300 KB!). On the MFC test it shaved of about 3,5 MB
compared to version 11.0.
That's very good news, but there is one thing to note: Some file types are compressed "pixel perfect",
in stead of "bit perfect", in the case of images for example, not a single pixel or ancillary info
(headers, tags, etc) is changed or lost in the compression/decompression process, however the
decompressed size and/or bit sequence may be slightly different than the original after it has
been compressed/decompressed. This is perfectly fine for normal users, but this is a lossless
compression test. It would be unfair towards the lossless "bit perfect" compressors to include these
Stuffit 12.0 result. I'm not sure what to do at the moment. I can retest it after I disable all
the recompression tricks (but compression would be worse) or I can add the results but with a note.
- 14-October-2007 : Added PAQ8O6, Stuffit 12.0 beta, CCM 1.25, PackJPG 2.3, LZTurbo 0.1 and
LPAQ4. Also added a new curiosity to the fun and
curiosity page, look for 'High and low compression with one file?' part.
The latest PAQ8O6 features an improved StateMap (from lpaq1) by Matt Mahoney, an improved JPEG
model (from paq8fthis4) by Jan Ondrus and speed improvements by Enrico Zeidler. It takes first
place in both the SFC
benchmark and MFC
benchmark!!. Congratulations to Matt Mahoney and all people involved in improving the PAQ engine
- 24-September-2007 : Added 7-ZIP 4.55b, WinRAR 3.71, HOOK 1.0, LZC 0.06,
LZPM 0.11, WinTurtle 1.30, lpaq1 v2, CMM 18-09-2007 and RINGS 0.1.
- 26-August-2007 : Added PAQ8O, Stuffit 11.0, THOR 0.96, LZPM 0.10, ASD 0.20, TarsaLZP
21-aug-2007, WinTurtle 1.21 and LZTurbo 0.01. The latest PAQ8 featuring the new PAQ8this2
JPEG model (developed by Jan Ondrus) and a new BMP model (developed by Andreas Morphis), takes
first place in the BMP test and
second place in the JPEG test.
It is identical to PAQ8L in compressing non-JPEG/BMP data.
PAQ8O is the first compressor capable
of compressing the entire SFC
testset to under 9 MB!!. Unfortunately PAQ8O uses more memory then specified in the help
text, so I had to run the MFC
benchmark in -6 mode. It still closed the gap to WinRK to under 500 Kb. In -7 mode it would
have taken first place with an archive size of 64.192.933 bytes.
THOR 0.96 compresses the complete 301MB MFC-testset in 3.3 sec, that's a compression speed of 91
MB/sec(!). Stuffit fails to correctly decompress the created archives in the MFC test, it reports
'damaged contents' every time. Lzturbo is a high performance lz77 based compressor featuring very
fast parallel compression/decompression, it's specially designed for modern processors and
multicore systems. LZTurbo is extremely memory hungry (level 4 already uses 600 MB). But
more importantly it fails to decompress the archive back to the binary identical input file (extracted
file is 3 bytes larger).
- 12-August-2007 : Added PAQ8M, PackJPG 2.2, LZPM 0.08, TarsaLZP 07-Aug-2007 and
WinTurtle 1.01. The JPEG compressor PackJPG shows improved compression and speed compared to previous
version. The latest PAQ8 also features a better JPEG model (developed by Jan Ondrus). It uses
DCT/IDCT to compute pixel values in adjacent blocks to provide extra context. PAQ8M is identical
to PAQ8L in compression for non JPEG data, it takes #1 place on the
SFC benchmark (method 2) and
widens the gap to #2 (WinRK). Unfortunately it crashes on the MFC testset after compressing about
195 MB. Also WinTurtle fails to decompress the MFC set.
- 05-August-2007 : Added 7-ZIP 4.52b, ALZip 7.0b, UPX 3.01, SR2, PIM 2.10, QuickLZ 1.30b13,
(command line interface) mode which is good news for us testers. PIMPLE2 shows big compression
TarsaLZP 01-Aug-2007, PIMPLE2, BCArchive 1.08.6 and LPAQ1. ALZip now also contains a CLI
improvements over PIMPLE, but is also a lot slower. LPAQ1 is a lite version of PAQ8L,
about 35 times faster, but compressed files are about 10% larger.
- 16-July-2007 : Added 7-ZIP 4.49, WinRAR 3.70, CCM 1.23, FreeARC 0.40, Turtle 0.07, PIM 2.03,
QuickLZ 1.30b10, PackJPG 2.0, LZC 0.05b, LZPM 0.6, TarsaLZP 17-June-2007 and MNzip. The JPG
(re)compressor PackJPG version 2.0 is now made public. It has slightly better compression then
the first demo version, taking 3th place, behind Stuffit and PAQ8L, in
the JPG test. FreeARC 0.40 contains
some multimedia filters, giving it a boost in the MFC test. It takes first place in the
efficiency benchmark. QuickLZ
still takes first spot on the MFC fastest
compression benchmark. The compression speed of 7-Zip seems to get a bit lower with the latest releases.
I suspect 7-Zip is being speed optimised for newer processors, which causes a little backfire on my old
XP2800+ reference PC.
I added a compression curiosities
page. If you have some ideas/tricks to add here, please let me know!.
- 12-May-2007 : Added HOOK 0.9c, PackJPG 1.9a, LZC 0.03, LZPM 0.4, CCMx 1.2.1, THOR 0.95,
TarsaLZP 07-may-2007, UPX 3.00 and retested MFC test for QuickLZ 1.30b3, FreeARC 0.36 and M99 v2.1.
PackJPG is a new JPG (re)compressor taking 3th place, behind Stuffit and PAQ8L, in the JPG test.
- 30-April-2007 : Added CCM 1.2.1, 7-ZIP 4.45b, HOOK 0.9b, M99 2.1, Tornado 0.1,
KZip (14-apr-2007), QUAD 1.12, FreeARC 0.36, PIM 1.50, DGCA 1.10 and the high speed compressors
THOR 0.94a, SLUG 1.1b and QuickLZ 1.30b3. QuickLZ (re)takes first spot on the
fastest compression benchmark.
CCM takes 10th place on the SFC
benchmark (method 2). Tornado, FreeARC and SLUG are new entries. BTW I'm thinking of
making a separate ramdrive test section (for the top-speed programs out there).
- 12-March-2007 : Added PAQ8L, CCM 1.1.6c, UPX 2.93, 7-Zip 4.44b, LZPX(J) 1.2h,
WinRAR 3.70b4 and retested HOOK 0.6c. PAQ8L's main changes are the addition of a DMC model
(like hook), and removing redundant sparse models to improve compression, speed and memory usage.
The latest PAQ8 shows improvement over PAQ8JD on almost all files, but performs slightly worse on
executable files. It keeps it's #1 place on the
SFC benchmark (method 2) and
widens the gap to #2 (WinRK). On the MFC
testset it still has the 2nd spot (behind WinRK). CCM 1.16c also shows some big improvements.
- 26-February-2007 : Added HOOK 0.6c, Bee 0.7.9 (build 0316), Quad 1.07b, QuickLZ 1.20,
Chile 0.5 (fails to compress the MFC testset) and CCM 1.1.1a
- 18-February-2007 : Added Quad 1.04a, QuickLZ 1.20, UPX 2.92, Ocamyd 1.66 final,
HOOK 0.6b, TC 5.2dev2, Chile 0.4 (MFC to do) and an excellent new program called CCM
(version 1.05a2). But there is more to come, I still have some backlog because I'm playing
with my new Dual core, 2 GB computer. :)
- 13-January-2007 : Added PAQ8JD, BZIP2 1.0.4, OCAMYD 1.66rc4, DARK 0.51, QUAD 1.02a,
UDA 3.01, HOOK 0.3 and PPMVC 1.2. The latest PAQ8 version improved again and sets a new record in
the SFC benchmark. On the
MFC testset it performs slightly
worse compared to PAQ8JC, but this is probably caused by the change of harddisk on my testrig. The
new HD is NTFS formatted causing the testfiles to be alphabetically sorted on filename in stead of
extension/filename like on the previous hard drive. When I have some time I will retest PAQ8JD with
'normal' file sorting.
- 30-December-2006 : Added 7-ZIP 4.44a1, Ocamyd 1.66rc3, TC 5.1 dev7, WinHKI 1.74 and Quad 1.01a.
- 08-December-2006 : Added PAQ8JC, WinRAR 3.62, RZIP 2.1 and UPX 2.91. I tested
several new PAQ8 family compressors (PAQ8JA, PAQ8JB, PAQ8JC and PAQ8hp6), of which PAQ8JC turned
out to be the best. It outperforms PAQ8i with 62 KB on the SFC testset, claiming a solid
first place on the SFC benchmark
method #2. What makes this result even more spectacular is the fact PAQ8JC doesn't use an external
dictionary (for that reason it looses to PAQ8i on the TXT and DICT test, but wins significantly on all others)!!.
PAQ8JC takes 1st place in the HLP, DOC and DICT test and five 2nd places, taking 2nd place in SFC benchmark
method #1. On the MFC testset it closes
the gap to WinRK to under 2 MB.
BTW The problem with RZIP 2.1, not able to correctly decompress files is solved by forcing rzip to
open all files as binary. Special thanks to Frank Teo for providing the solution!.
- 02-December-2006 : Added Stuffit 9.5. On windows platforms StuffIt, now owned by SmithMicro,
is mainly known for it's high jpeg (re)compression. This latest version will probably not change that
as compression is about the same as previous tested version. It still takes first place in the
The problem with RZIP 2.1, not able to correctly decompress files seems to be caused by cygwin1.dll in
combination with some AMD processors.
- 26-November-2006 : Added WinZIP 11.0 (final), SQUEEZ 5.5, QuickLZ 1.10, TC 5.1dev6 and RZIP 2.1.
WinZIP 11.0 final performs the same as the last tested beta version, SQUEEZ 5.5 has identical
compression to version 5.20. Both QuickLZ 1.10 and TC 5.1dev6 shows better compression compared
to previous versions for most filetypes. RZIP 2.1 is still not able to correctly decompress the
MFC testset (possibly caused by compile switches?).
- 21-October-2006 : Added WinZIP 11.0 beta, BigCrunch 0.4a1 and AMG 2.2. The latest WinZIP
includes BZ2 and .RAR file support (extract only), and offers some new features like a "Best
Compression" mode, letting WinZip decide which compression method to use for each file to improve
compression. It now also has a special algorithm to compress .wav files!. Research done by
Guillermo Gabrielli shows the audio compressor used for this is WavPack (using Lossless, High compression
mode). On the MFC test, WinZIP 11 in optimize for best compression mode,
now compresses on par with Microsofts CABARC format (in LZX:21 mode). The PPMd compression mode of WinZIP 11
is also improved compared to previous v10.0 (experiments suggest they changed the PPMd parameters from 50MB
order 8, to 64MB order 16).
One site update this time: The rankings resulting from the SFC and MFC tests are now numbered. With the
growing number of programs tested, it will clarify the overall listings. An arrival in the 'top10' or
'top20' or 'top30' is now visible in one glance (thanks to RePack for the suggestion).
- 15-October-2006 : Added TC 5.1dev5, DARK 0.50c, KZIP 11-Oct-2006 and UPX 2.90. The latest UPX now
includes LZMA algorithm support for most of the 32-bit and 64-bit file formats, greatly improving
compression (note: UPX - shorthand for Ultimate Packer for eXecutables - is not lossless and
added to EXE and DLL test for reference purposes only). The latest TC shows higher compression
levels in both in SFC and MFC. DARK now successfully decompresses the MFC testset.
BTW I get lots of questions about KGB and Infima and why they are not added to the site. In short
KGB is a PAQ6 clone (it's PAQ6v2 with a GUI). Infima is one big joke, the program is a combination of all
the top command line archivers the 'author' could get his hands on, without crediting the original authors.
Those compressors are disguised as plugins by compressing them with UHarc using password protection. It's a
sad piece of 'work' and for obvious reasons will not be added to the site...
- 08-October-2006 : Added QuickLZ v0.1, WinHKI 1.73, TC 5.1dev3 and DARK 0.50a. Although it kept
the same version number, QuickLZ v0.1 has undergone significant speed improvements (data is now
being compressed and written to disk in 1 Mbyte chunks). QuickLZ is the fastest command line
compression program ever tested on maximum compression. It takes first place in the
MFC compression speed test. It beats
the previous number 1 (THOR in fast mode) with 0.6 seconds. QuickLZ compresses the complete 300+ MByte
testset in 3.7 seconds (versus 4.3 seconds for THOR). Compression ratio of THOR is however (much) better.
DARK 0.50a has improved compression, but fails to correctly decompress the MFC test-set.
- 01-October-2006 : Added TC 5.1dev1, LZPX(J) 1.2g, PIM 1.25 and QuickLZ v0.1.
QuickLZ is a new program written in 80486 compatible assembler with very high (de)compression speed. It
scores very well on the MFC speed
comparison, but it could do even better, as actual compression/decompression takes place in
(less then) half the times listed. The program flushes is buffers entirely before returning to the
command prompt, loosing valuable seconds. BTW for MFC, I tested a special version with the 100 Mb
- 17-September-2006 : Added 7-ZIP 4.43b, WinRAR 3.61, Dark 0.47b, QC 0.050 and LZPX(J) 1.2f.
The latest LZPX(J) compresses slightly better and is about 2 times faster on the MFC test set. The author
of QC added some switches to improve compression.
- 03-September-2006 : Added Dark 0.47, 7-ZIP 4.43a5, PIM 1.10, CA-ZIP 3.4, KZIP 26-08-2006
and BBB ver 1. The new program BBB (Big Block BWT) is a file compressor capable of using very large
BWT blocks (80% of available memory) and has a PAQ-like order 0 model back end (but no context mixing).
It's specially suited for compressing huge, homogeneous text files where its large block size is an
- 27-August-2006 : Added PAQ8i (MFC to do), Dark 0.46, QC 0.033, BMA 1.33, PIM 1.06,
PPMT 0.1, LZ 1.0, HuffComp 1.3, HIP b1, VuZip 1.8 and Secura Archiver 1.7. The latest versions of QC,
Dark and BMA show slightly improved compression. PPMT, LZ, HuffComp and HIP all fail on the MFC testset.
The latest PAQ8 features a new PGM model and a small bugfix in the BMP model (which saved about 700
bytes on the BMP test). PAQ8I is still ranked #1 in SFC, method
2 and #2 in SFC method 1.
- 17-August-2006 : Added Dark 0.41, QC 0.023, 7-ZIP 4.43 alpha 4, WinRAR 3.60 final, PIM 1.05
and UPX 2.02 (note: UPX is not lossless and added to EXE and DLL test for reference purposes only).
The latest version of Dark (full name 'Dark Archon') has better compression in both
SFC and MFC caused by implementation
of fragmentation of binary data and solid mode. More important, DARK seems to be almost stable now. QC
shows big improvements in compression ratio on both the SFC and MFC test. WinRAR shows improved
compression speed compared to version 3.51
- 08-August-2006 : Added Dark 0.36b, TarsaLZP 05-Aug-2006, WinHKI 1.72 and two new programs
called QC 0.015 and PIM 1.0a4 (QC seems to be the successor of QAZAR and PIM the successor of PIMPLE).
Dark, like previous version crashes on the MFC testset. TarsaLZP shows good improvement, both in MFC and
SFC (mainly due to hand tuned ini-settings for SFC benchmark - thanks Piotr).
- 01-August-2006 : Added TC 5.0 dev11, LZPX(J) 1.2e and TarsaLZP (which still hasn't got
proper version numbering). All three programs show improved compression.
- 23-July-2006 : Added TC 5.0 dev10, WinRAR 3.60b8, 7-ZIP 4.43 alpha 3, TarsaLZP
and UDA 3.00. Also made some small layout changes to the site. UDA 3.00 is a modification of PAQ8H
(with optimizations for speed) and for that reason it's only added to MFC.
- 13-July-2006 : Added TC 5.0 dev8, WinRAR 3.60b7 and WinACE 2.65. The latest TC fails to correctly
decompress the MFC test set (output size is three times the original input size). Update: The author of TC
released a new recompiled version (which uses the same 'dev8' name) solving the problems.
- 10-July-2006 : Added TC 5.0 dev7, WinHKI 1.71 and a new program called DARK 0.32b.
TC 5.0 dev7 shows big improvements in both SFC and MFC test. DARK crashes on the MFC testset.
- 03-July-2006 : Added TC 5.0 dev5, ShipInBottle 1.0 beta 17, WinRAR 3.60b6,
7-ZIP 4.43 alpha 2 and TarsaLZP. TC 5.0 dev5 shows improved compression for SFC tests, but
slightly worse compression for MFC. 7-Zip 4.43 has slightly improved compression in the MFC test.
- 21-June-2006 : Added THOR 0.93a, WinRAR 3.60b5, TC 5.0 dev4, LZPX(J) 1.2a,
KZIP 02-June-2006, UPX 2.01, ASD 0.1.5, Shindlet and WinHKI 1.69. THOR has a new 'ultra'
switch significantly improving compression on most files. TC 5.0 dev4 also has better
compression (compared to dev1). LZPXJ 1.2a wins a couple of bytes, mainly in the EXE/DLL
- 27-May-2006 : Added KZIP 25-May-2006, WinRAR 3.60b4, TC 5.0 dev1 and Ocamyd 1.65
final. The latest KZip version has slightly improved compression. Ocamyd has improved
(de)compression speed compared to the latest release candidate. TC shows greatly improved compression.
- 21-May-2006 : Added OCAMYD 1.65 RC1, ALZip 6.32, LZPX(J) 1.1d and PIMPLE 1.43b.
- 13-May-2006 : Added OCAMYD 1.65dev25, 7-ZIP 4.42, ZPack, PPMd J rev.1 and the
bugfixed version of PPMVC 1.1.
- 08-May-2006 : Added OCAMYD 1.65dev23, 7-ZIP 4.40b, WinRAR 3.60b3 and PPMVC 1.1.
PPMVC has slightly improved compression, but is very buggy. It crashes on many different
switch settings ( e.g. -o12 for the PDF file).
- 30-April-2006 : Added ALZip 6.31, WinHKI 1.68 and UPX 2.00.
- 24-April-2006 : Added SQUEEZ 5.20, OCAMYD 1.65dev21, LZPX(J) 1.1b and
PIMPLE 1.43b (SFC to do). PIMPLE 1.43b has decreased MFC compression compared to previous version.
LZPX(J) 1.1b added some switches resulting in improved compression on especially the SFC testset.
OCAMYD 1.65dev21 again shows improved compression; where will it end :-)
- 18-April-2006 : Added WinRAR 3.60b2, LZPX(J) 1.1a. LZPX(J) 1.1 solves the
bug found in version 1.0s and has improved executable compression.
- 16-April-2006 : Added 7-ZIP 4.39b, OCAMYD 1.65dev20 and LZPX(J) 1.0s. Both OCAMYD
and LZPX(J) show improved compression, but LZPX(J) isn't able to correctly decompress the MFC.tar
file (decompressed file size is smaller then original file).
- 07-April-2006 : Added Durilca 0.5, PIMPLE 1.42, OCAMYD 1.65
dev16, UPX 1.95, WinHKI 1.67 and tested Infima Archiver. Durilca (light) now uses the new PPMd-J engine and shows
improved compression on all tests (both SFC and MFC) and takes 4th place in SFC method #2.
It's also the highest ranked non 'PAQ-like-engined' compressor in MFC. OCAMYD has somewhat worse
compression compared to the dev9 version.
Infima seems to be a (buggy) jumble of all available top command line archivers (some of them
even lossy!), the 'author' could get his hands on (without crediting the original authors). Those
compressors are disguised as plugins by compressing them with UHarc using password protection. More
over, I can't get it to run correctly on WinXP with SP2 (out of memory error when I try to start up
the program). Some additional info can be found in the guestbook.
For obvious reasons Infima will not be added to the site...
- 03-April-2006 : Added WinRAR 3.60b1, PIMPLE 1.41, aPLib 0.43
and Coder 1.1. The latest WinRAR shows improved compression speed. In normal mode
decompression speed was also significantly improved.
- 27-March-2006 : Added PAQ8H, PIMPLE 1.40b, DACT 0.8.41, OCAMYD 1.65
dev9, BriefLZ 1.04 and OrangeArchiver 1.05a. PAQ8H, based on PAQ8G with some improvements to
the model, (re)takes 1st place in SFC best overall compression program method #2. It also
shows some improvement in the MFC test, but the gap to WinRK is still quite big. OCAMYD shows
improved compression on most testfiles. PIMPLE 1.40 has also some very nice compression
improvements. OrangeArchiver is a newly tested archiver with moderate compression.
- 19-March-2006 : Added Pimple 1.20, THOR 0.9a and 7-ZIP 4.37b. The new fast-mode of THOR 0.9a
sets a new speed record. Compression of the complete MFC testset takes just 4.33 sec, that's a
compression speed of 69 MB/sec(!).
- 12-March-2006 : A big update this time (including the new number one and two program). Added
programs are: WinRK 3.0.3, PAQ8G, Quark 0.95r, WinACE 2.61, Pimple 1.1b5, WinHKI 1.66, File2Pack 2.0,
THOR 0.81a, UPX 1.94, AI 1.1 and 7-ZIP 4.36b. WinRK 3.0.3 has improved compression (and speed) taking
1st place in MFC and SFC
test, both method #1 and #2. The gap between WinRK and the number two (PAQ8) in the MFC test has
increased to over 3 MB. Well done Malcolm!.
PAQ8G (basically PAQ8F with external dictionaries) takes 2nd place on all mentioned 'best
overall compression program' charts (MFC, SFC method #1 and #2). The gap to WinRK in SFC is much
smaller because of the (embedded) JPG detection feature in PAQ8 which can be used on at least two files
of the SFC set. The latest Quark 0.95r has significant improved compression in both SFC and MFC.
PIMPLE 1.1b5 has better executable compression. THOR 0.8a has improved decompression speed. AI is an old
program (2000) with surprisingly good compression for some filetypes.
- 03-March-2006 : Added PAQ8F, EAZEL 1.0 and THOR 0.7a. PAQ8F has a more memory efficient
context model, a new indirect context model and a more robust detector for x86 preprocessing to
improve compression compared to PAQ8A. It also doesn't use an English dictionary like PAQ8E did.
PAQ8F takes 1st place in SFC test, both method #1 and #2.
On the MFC test it scores slightly worse then PAQ8A/E (probably
because PAQ8F uses a different file-order when compressing files).
The new THOR 0.7a has a max compression switch yielding improved compression in both SFC and MFC.
EAZEL 1.0 is a new compression program with compression results similar to zip/gzip when using it's
own EAZ format.
- 28-February-2006 : Added PAQ8E, 7-ZIP 4.34a1, BMA 1.32, PIMPLE 1.1b3 (b2 for SFC),
KGB 1.1 (MFC) and WinHKI 1.65. PAQ8E has improved compression compared to PAQ8A for some files with
text files due to the addition of WRT / dictionaries, and is also a bit faster. PAQ8E takes 1st place
in SFC test, method #2, 2nd place in method #1 and also a
2nd place in the MFC test. The new BMA 1.32 and PIMPLE 1.1
show improved compression in both SFC and MFC. KGB 1.1 is basically PAQ6v2 with a GUI and some other extensions.
- 19-February-2006 : Added PAQAR 4.5, PAQ8D (MFC) and PIMPLE 1.0b12.
PAQAR 4.5, probably the last version based on the PAQ6 core, takes 3th place on
SFC testset and positions itself just before
PAsQDa 4.4 on the MFC set. As PAQAR now includes
the WRT from Przemyslaw Skibinski's PAsQDa the programs have become basically the same.
For that reason I removed PAsQDa from the SFC charts.
PAQ8D, is an improved version of PAQ8A with a text filter / dictionaries and better file type
detection routines. It takes 2nd place in the MFC set. I decide not to include the
program in the SFC results as it contains a dictionary specifically optimised for this
testset, which makes comparing results to other programs unfair (it would have taken first place).
- 16-February-2006 : Added THOR 0.6a, Fast PPMII (PPMd) rev J and Monstrous PPMII (PPMonstr)
rev J ; The latest THOR (already the fastest compressor tested) is again faster and now finishes the complete
MFC test in just under 6 seconds (51 MB/s). In addition to the
better speed it also has small improvements in compression on most files. The latest version of PPMonstr
shows huge compression improvements on both SFC and
MFC testset. It climbs to 9th place on best overall compression program
method #2 and 6th place method #1. PPMd rev J also compresses a bit better then previous version.
I started adding decompression times to the MFC test, if all times are available I will add an
extra sort-option for it. I have to say I'm already very disappointed by the amount of compressors /
archivers that fail to correctly decompress the created archive... :( If possible please do some
tests with the mentioned programs / settings on large datasets. I'm very curious about the (crash)
- 11-February-2006 : Added PAQ8b (MFC), 7-ZIP 4.33b, PIMPLE 1.0b11 and THOR 0.5a. PAQ8b has
improved compression for some files with text files due to the addition of WRT / dictionaries, and
is a bit faster then PAQ8a (because of Intel 8 compile). On MFC test it scores slightly worse. THOR
0.5a has again improved (de)compression speed over previous version (now 44.6 MB/s).
- 29-January-2006 : Added PAQ8a, 7-ZIP 4.33a2, WinHKI 1.61, LZPX(J) 1.0f, QazaR 0.0 pre5,
PIMPLE 1.0b9 (MFC) and THOR 0.4a. PAQ8a (an experimental pre-release of PAQ8) is an improved and
bugfixed version of PAQ7. PAQ8a takes 1st place in SFC test,
method #2 and 2nd place in method #1. The bugfix (some non jpg-files where coded with a
specific JPG algorithm) also caused a huge positive effect on the MFC
test, were it jumped to 2nd place. And that without making use of a dictionary. Good job indeed!.
THOR 0.4a is now the fastest compressor tested; it finishes the complete
MFC test in just under 7 seconds (44 MB/s).
LZPX(J) 1.0f now has a switch setting to improve compression for 16/24/32-bit data
(like WAV or 24-bit BMP). Also QazaR added some switches to improve compression.
- 26-January-2006 : Added THOR 0.3a, QazaR 0.0 pre4d and 7-ZIP 4.33a1 (SFC results).
THOR 0.3a and QazaR both have better compression while THOR, written in Delphi(!), is also
faster then previous version; it now clocks a compression speed of almost 43 MB/s in
- 24-January-2006 : Added PAQ7plus 1.19, PIMPLE 1.0b8, 7-ZIP 4.33a1, THOR 0.1a. This latest
PAQ7plus 1.19 version has slightly improved compression both in MFC and SFC. Pimple now
uses SSE, improving compression on all filetypes. The latest 7-Zip version is again faster, but
at the expense of higher memory usage (SFC tests still to be done). THOR is a new program with
very high (de)compression speed; in the MFC test it clocks a compression speed of almost 39
MB/s (!). That's about the same speed as LZOP but with much better compression in the MFC
test (in SFC it scores a bit worse).
- 16-January-2006 : Added test data for PAQ7plus 1.11, LZPXJ 1.0e and Windows XP
build-in compression. PAQ7plus combines the models and mixer from PAQ7 (includes .jpg and .bmp models for
enhanced image compression) with the English dictionary and .exe/.dll preprocessor, arithmetic coder and
state table from PAsQDa. PAQ7plus now takes 1st place in SFC test,
method #2 and 2nd place in method #1!.
- 07-January-2006 : Added PAsQDa 4.4, BEE 0.7.9, Pimple 1.0b7, and LZPX(J) 1.0d.
Both PAsQDa and LZPX shows some improvement on the MFC test, minor changes on the SFC test.
LZPX also shows a nice speed improvement. Update 10 jan 2006: Found some better switches for
BEE (thanks to Melchiorre Caruso).
- Updates before 2006 : January 2003 - December 2005
©2003-2011 MaximumCompression (lossless data compression software benchmarks)