11 Comments

1. Chris Madison   •  

   We ran the benchmarks against a modern version of ImageMagick (6.4.3) and got dramatically different results. In most cases the run times are similar. We did notice the benchmark script forces ImageMagick to cache the pixels to disk with these lines:

   MAGICK_MAP_LIMIT=4000
   MAGICK_MEMORY_LIMIT=4000
   MAGICK_AREA_LIMIT=200000
   export MAGICK_MAP_LIMIT MAGICK_MEMORY_LIMIT MAGICK_AREA_LIMIT

   With GraphicsMagick working in memory and ImageMagick on disk we would expect a significant slowdown with ImageMagick. Instead comment out the export statement to get a proper comparison or correct the values (use 4000mb instead of 4000).

   ImageMagick in some cases runs a bit slower than GraphicsMagick mainly because it supports additional features such as full alpha support, most algorithms permit channel selection (i.e. blur only red channel), support for CMYKA images, and ImageMagick is thread-safe whereas GraphicsMagick is not.

   The GM developers are working on getting OpenMP into GM but OpenMP support has been included in ImageMagick for some time now.

   Finally, ImageMagick supports hundreds of features that GraphicsMagick does not. See http://www.imagemagick.org/Usage for details.
2. allspaw   •     Author

   Thanks for the info, Chris. We don’t use any other features other than identify and convert, so the extra features aren’t much of a motivator for us.

   We’ll be sure to try and test the new version of ImageMagick.
3. Bob Friesenhahn   •  

   Hi, this is a response to Chris Madison’s post. I am the principle maintainer of GraphicsMagick. The benchmark was correct for the ImageMagick release which was benchmarked but ImageMagick has changed its syntax since then. The benchmark has since been updated to use the new ImageMagick syntax. Subsequent testing shows that ImageMagick has responded to these benchmark results by eliminating many of the performance issues uncovered by the benchmark.

   It is true that ImageMagick has been releasing OpenMP features earlier than GraphicsMagick but this is because of the difference in development model. When GM 1.3 is released, most algorithms will be OpenMP accelerated, well tuned, and tested with several different OpenMP-capable compilers on a wide variety of platforms. Only a few algorithms remain to be re-implemented.

   The link to the ImageMagick usage page is appreciated, but the implication is that GraphicsMagick does not support most of those features. In fact, GraphicsMagick does support most of the features described.
4. Bob Friesenhahn   •  

   As a followup to the first and last posts, the GraphicsMagick benchmark report URL is now http://www.graphicsmagick.org/benchmarks.html. Since my last post, GraphicsMagick 1.3 has been released and the command-line performance just blows latest ImageMagick out of the water (sorry Chris! :-). The typical speed boost is perhaps 2X to 3X but in one case we observed at 272X speed difference.

   When ImageMagick is executed by hand from the command line it does not feel so bad, but if you run it repeatedly in a script, it runs quite poorly.

   We will soon be releasing GraphicsMagick 1.3.1 which further boosts the dramatic speed advantage that GraphicsMagick has over ImageMagick.

   Bob
5. Ja Hui   •  

   We’re running on a Pentium 3.0GHZ and the timing results were pretty close in most cases where GM was slightly faster for about half the operations and IM was slightly faster for the remaining half. We looped the PerlMagick Sharpen(‘5×2’) method on a large image 20 times and the average ImageMagick time was 16.9 seconds with 22.1 for GraphicsMagick. For Implode(0.3), GraphicsMagick was slightly faster at 12.4 seconds with ImageMagick coming in at 15 seconds.
6. Bob Friesenhahn   •  

   Ja Hui,

   What versions of IM and GM are you using? Also, are they built to support OpenMP? It seems that many/most people using GM are using a very antique version released back in 2004. A huge amount of work has been done since that time.

   See http://www.graphicsmagick.org/OpenMP.html to see what kind of speedups we are seeing with OpenMP.

   Bob
7. Anton   •  

   Cool!

8. Pingback: 再跟 Flickr 学习网站运维经验

9. Pingback: CBCのBlog » 再跟 Flickr 学习网站运维经验

10. Pingback: 再跟 Flickr 学习网站运维经验 – GcogeBlog - 上海酷鸽信息科技有限公司官方博客

11. Pingback: 再跟 Flickr 学习网站运维经验 | web开发