Platform and Nature of Decompression Test                  (see notes below for explanations)		Digital Cinema (24fps) 2048x857,  240 frames, W9x7, Cblk={32,32}, 10.176 Mbits/frame    =5.8 bits/pixel (full XYZ) =244 Mbits/s @ 24fps	Digital Cinema (48 fps) 2048x857, 240 frames, W9x7, Cblk={32,32}, 15.088 Mbits/frame  =2.9 bits/pixel (full XYZ) =244 Mbits/s @ 48fps	Big Image (lossless) 13,333x13,333 (full RGB)  RPCL, W5x3, Clayers=8, Clevels=9, Cblk={64,64}, Cprecincts={256,256}	Big Image Tiled (2bpp) 13,333x13,333 (full RGB)  1024x1024 tiles, RPCL, W9x7, Clayers=5, Clevels=6, Cblk={64,64}, Cprecincts={256,256}	4CIF Video (2bpp) 704x576 (4:2:0 YCbCr), 1200 frames, W9x7, Clevels=5, Cblk={64,64}	4 CIF Video (0.5 bpp) 704x576 (4:2:0 YCbCr), 4800 frames, W9x7, Clevels=5, Cblk={64,64}
Core 2 Duo™ 2.4GHz x86 on WOW64                            [All results reported in frames/second]	v5.2.6 (expand)	4.36	6.33	0.0370	0.135	62.50	142.86
	v6.0 (expand)	4.92	7.84	0.0377	0.136	63.83	144.14
	vs6.0 (expand)	4.93	7.92	0.0377	0.136	65.93	162.16
	v5.2.6 (render)			0.0368	0.134
	v6.0 (render)			0.0373	0.142
	vs6.0 (render)			0.0373	0.146
	v6.0 (vex_fast)	6.00	10.90			67.85	179.44
	vs6.0 (vex_fast)	5.98	10.87			70.20	193.09
Core 2 Duo™ 2.4GHz x64 on WinXP64                          [All results reported in frames/second]	v5.2.6 (expand)	4.43	6.72	0.0360	0.131	61.86	144.14
	v6.0 (expand)	4.98	8.28	0.0365	0.132	63.16	147.24
	vs6.0 (expand)	7.04	10.91	0.0532	0.181	90.23	207.79
	v5.2.6 (render)			0.0357	0.131
	v6.0 (render)			0.0360	0.137
	vs6.0 (render)			0.0522	0.197
	v6.0 (vex_fast)	5.76	10.42			66.84	184.06
	vs6.0 (vex_fast)	8.43	15.30			101.59	295.38
Core 2 Duo™ 2.4GHz    best vs6.0 speed improvement over          best v5_2_6 speed	expand (x86 or x64)	1.59	1.62	1.44	1.45	1.44	1.44
	fastest possible (x86 or x64)	1.90	2.28	1.44	1.45	1.63	2.05
Core 2 Duo™ 2.4GHz    best vs6.0 speed improvement over          best v6.0 speed	expand (x86 or x64)	1.41	1.32	1.41	1.33	1.41	1.41
	fastest possible (x86 or x64)	1.41	1.40	1.41	1.39	1.50	1.60
Dual Quad-Core Xeon™  2.66GHz x64 on WinXP64         [All results reported in frames/second]
	v6.0 (vex_fast)	24.89	44.14
	vs6.0 (vex_fast)	34.59	58.18
Dual Quad-Core Xeon™  2.66GHz on x86-64 Mandriva Linux (Spring 2007)            [All results reported in frames/second]
	v6.0 (vex_fast)	24.24	43.17
	vs6.0 (vex_fast)	35.09	60.15
Notes:	1) "vs6.0" refers to the separately licensed Kakadu "speed-pack" -- this is for sure the fastest software implementation of JPEG2000.
	2) "expand" means "kdu_expand" for images and "kdu_v_expand -overlapped_frames" (v5.2.6) or "kdu_v_expand -overlapped_frames -in_memory 1" (v6.0 / vs6.0) for video
	3) "render" means "kdu_render -cpu" for images; it measures the throughput achievable using `kdu_region_decompressor', which is at the heart of all high level Kakadu display rendering tools.  Note that the special "-cpu" option is new to v6.0; we ported the option to v5.2.6 just to obtain a comparison for rendering applications.  You will note that the render option is noticeably faster for large tiled images than "kdu_expand".  This is because "kdu_expand" always processes a complete horizontal row of tiles concurrently, which can cause cache thrashing if an entire large image is being decompressed at once.  By contrast, Kakadu's rendering objects operate on smaller numbers of tiles, writing their results to a random access display buffer.  You will also notice that rendering of tiled imagery has been accelerated in version 6.0.
	4) "vex_fast" means "kdu_vex_fast -display" -- this is a new application to v6.0 and is also the fastest way to render video.
	5) Speeds on x86 MAC OS platforms are pretty much identical to those quoted for WinXP64 -- exactly the same speedup options are available on 64-bit Intel MAC's as on 64-bit Windows boxes, for all variants.
	6) Speeds on 64-bit Linux platforms are pretty much identical to those quoted for WinXP64 -- exactly the same speedup options are available for X86-64 Linux, 64-bit Windows, and 64-bit MAC OSx86 binaries.  In fact, the results quoted above for the digital cinema case suggest that 64-bit Linux may be slightly faster in some circumstances.
	7) Speeds on 32-bit Linux platforms should be a little bit lower (up to about 10% slower) than those for 32-bit (x86) Windows platforms, because the 32-bit block decoder acceleration code is not available.  64-bit Linux does not suffer from this disadvantage, as mentioned above.