Newsgroups: comp.dsp Message-ID: <1996May7.102101.25391@hgl.signaal.nl> Sender: usenet@hgl.signaal.nl (USENET News System) Nntp-Posting-Host: suns2c22 Organization: Hollandse Signaal apparaten BV X-Newsreader: TIN [version 1.2 PL2] References: <3187DA38.1A9D@trw.com> Date: Tue, 7 May 1996 10:21:01 GMT Lines: 147 Here the latest, corrected figures of Complex FFT preformance for several current available processors, GENERAL purpose DSP's and SPECIAL purpose DSP's. +--------------------------------------------------------------------------------+ | FFT | Am29050 | CRISP | i860 | PowerPC 604 | Pentium #| P6 | | | 40 MHz | 50 MHz | 50 MHz | 150 MHz | 133 MHz | 200 MHz | |================================================================================| | 8 | ? | ? | ? | ? | 20.6 us | ? | | 16 | ? | ? | 6.8 us | ? | 46.2 us | ? | | 32 | ? | ? | 14.0 us | ? | 104.6 us | ? | | 64 | ? | ? | 27.0 us | ? | 238.3 us | ? | | 128 | ? | ? | 70.9 us | ? | 538.8 us | ? | | 256 | 690.0 us | 360.0 us | 149.5 us | ? | 1207.9 us | ? | | 512 | ? | ? | 312.9 us | ? | 2678.9 us | ? | | 1024 | 3740.0 us | ? | 753.1 us | ? | 6012.1 us | ? | | 2048 | ? | ? | 1793.9 us | ? | 13427.5 us | ? | | 4096 | ? | ? | ? | ? | 30130.0 us | ? | | 8192 | ? | ? | ? | ? | 64959.9 us | ? | | 16384 | ? | ? | ? | ? | 139490.4 us | ? | +--------------------------------------------------------------------------------+ +---------------------------------------------------------+ | FFT | DSP56001 | DSP56002 | DSP56301 | DSP96002 | | | 33 MHz | 80 MHz | 66 MHz | 40 MHz | |=========================================================| | 8 | ? | ? | ? | ? | | 16 | ? | ? | ? | ? | | 32 | ? | ? | ? | ? | | 64 | ? | ? | ? | ? | | 128 | ? | ? | ? | ? | | 256 | ? 940.0 us ? | 160.0 us | ? | ? | | 512 | ? | ? | ? | ? | | 1024 | ? 1040.0 us ? | 828.0 us | 692.7 us | 1050.0 us | | 2048 | ? | ? | ? | ? | | 4096 | ? | ? | ? | ? | | 8192 | ? | ? | ? | ? | | 16384 | ? | ? | ? | ? | +---------------------------------------------------------+ +-------------------------------------------------------------------------------------+ | FFT | TMS320C25 | TMS320C30 | C40 | C44 | C50 *| C80/C82 (1of4) | | | 50 MHz | 40 MHz | 60 MHz | 60 MHz | 80 MHz | 50 MHz | |=====================================================================================| | 8 | ? | ? | ? | ? | 4.5 us | ? | | 16 | ? | ? | 8.0 us | ? | 11.5 us | ? | | 32 | ? | ? | ? | ? | 29.6 us | ? | | 64 | ? | ? | 47.5 us | ? | 72.3 us | ? | | 128 | ? | ? | 111.2 us | ? | 171.3 us | ? | | 256 | 1800.0 us | 680.0 us | ? | 396.1 us | 396.2 us | 95.3 us | | 512 | ? | ? | ? | ? | 900.4 us | ? | | 1024 | 15600.0 us | 1970.0 us | 1297.5 us | ? | 2017.4 us | 482.0 us | | 2048 | ? | ? | ? | ? | ? | ? | | 4096 | ? | ? | ? | ? | ? | ? | | 8192 | ? | ? | ? | ? | ? | ? | | 16384 | ? | ? | ? | ? | ? | ? | +-------------------------------------------------------------------------------------+ +-------------------------------------------------------+ | FFT | ADSP2100 *| ADSP2181 *| ADSP21000 | SHARC | | | 16.7 MHz | ? MHz | 33.3 MHz | 40 MHz | |=======================================================| | 8 | ? | ? | ? | 1.5 us | | 16 | ? | ? | ? | 2.2 us | | 32 | ? | ? | ? | 8.8 us | | 64 | ? | ? | ? | 18.4 us | | 128 | ? | ? | ? | 41.8 us | | 256 | 445.4 us | ? | 135.0 us | 80.9 us | | 512 | ? | ? | ? | 195.3 us | | 1024 | ? | 1070.0 us | ? | 384.9 us | | 2048 | ? | ? | ? | 928.6 us | | 4096 | ? | 9200.0 us | ? | 1814.0 us | | 8192 | ? | ? | ? | 4350.1 us | | 16384 | ? | ? | ? | 8388.7 us | +-------------------------------------------------------+ +------------------------------------------------------+ | FFT | TMC2310 | PDSP16510 | LH9124 *| swiFFT *| | | 20 MHz | 40 MHz | 50 MHz | 40 Mhz | |======================================================| | 8 | - | - | 0.3 us | 1.5 us | | 16 | 6.6 us | 0.7 us | 0.3 us | 2.2 us | | 32 | 15.0 us | - | 1.3 us | 3.3 us | | 64 | 27.8 us | 4.1 us | 2.6 us | 5.4 us | | 128 | 66.6 us | - | 7.7 us | 9.5 us | | 256 | ? | 20.4 us | 11.6 us | 16.0 us | | 512 | ? | ? | 33.2 us | 41.9 us | | 1024 | 514.0 us | 97.7 us | 64.6 us | 80.5 us | | 2048 | - | - | 167.6 us | 158.1 us | | 4096 | - | - | 249.8 us | 311.8 us | | 8192 | - | - | 660.2 us | 824.1 us | | 16384 | - | - | 1316.0 us | 1600.0 us | +------------------------------------------------------+ - <=> not available ? <=> unknown (yet!?) # <=> these are figures for unoptimized FFT-code, a fully optimized 1k complex FFT takes about 1250 us on a P5-133 * <=> verified figures For all FFT's I assumed that the data was already in zero wait-state external memory, so no reduction of performance due to slow memory. The read and write cycles (getting data into cache or core) plus any other unavoidable overhead (like bit reversing and windowing) were taken into account. Where available, Radix-4 butterfly figures or figures for combinations of Radix-4/Radix-2 are used. Greatings, Henk P.S. I have some more figures, but I can't the time to verify and evaluate them. P.P.S. Thanks to Mike Flemming for correcting the LH9124 figures, to John Marsh for the swiFFT figures, to George Kechriotis for supplying the i860 figures, to Heiner Schepers for supplying part of the SHARC figures to Ahti Heinla for the Pentium figures and to Vesa Ekholm for his contribution. -- ===============================[ Unclassified ]=============================== H.R.B. Holdijk (Henk) Email: hrb_holdijk@hgl.signaal.nl dc064@hgl.signaal.nl Naval Combat Systems Fax. +31-74-2484044 R&D Radar & Sensors Tel. +31-74-2483999 processing section Hollandse Signaalapparaten B.V. P.O. Box 42 7550 GD, Hengelo ov. The Netherlands ==============================================================================