Sign in
Search imagedsp
Subscribe to imagedsp
imagedsp by Keywords
Discussion Groups
Discussion Groups | Image Signal Processing | TI DM642 DSP vs. Pentium
Technical Discussions related to Image Processing (image coding, compression, digital effects, mpeg, etc)
TI DM642 DSP vs. Pentium - Lisa - Nov 6 17:48:00 2003
|
Hi all, I am new to the group, and would like to ask a question. I assume you all are familar with TI's new DSP, the "digital media" DM642. It is a member of the new C64x series DSPs, with clock speeds up to 720MHz. My question is, how do you think this DSP will compare in machine vision image processing applications to a "typical" desktop PC in the 2 to 3 GHz range. I know I can't just compare clock speeds, and a lot depends on how well the software is written, but in general, would this DSP be slightly slower, about the same, or slightly faster (or something else) ignoring any IO and memory bottlenecks. Thanks |
|
|
(You need to be a member of imagedsp -- send a blank email to imagedsp-subscribe@yahoogroups.com )
Re: TI DM642 DSP vs. Pentium - Jeff Brower - Nov 8 4:45:00 2003
|
Lisa- > Hi all, I am new to the group, and would like to ask a question. > I assume you all are familar with TI's new DSP, the "digital media" > DM642. It is a member of the new C64x series DSPs, with clock speeds > up to 720MHz. > > My question is, how do you think this DSP will compare in machine > vision image processing applications to a "typical" desktop PC in the > 2 to 3 GHz range. > > I know I can't just compare clock speeds, and a lot depends on how > well the software is written, but in general, would this DSP be > slightly slower, about the same, or slightly faster (or something > else) ignoring any IO and memory bottlenecks. To ignore I/O and memory bottlenecks makes the issue purely "benchmark". You should be able to compare some benchmarks for typical algorithms published by Intel and TI and get some idea. But the real question is "what can you build". Issues of chip size, power consumption, I/O interfaces and throughput are critical. In our product line, the huge heatsink and fan make a 2.4 GHz Pentium impossible. Jeff Brower system engineer Signalogic |
(You need to be a member of imagedsp -- send a blank email to imagedsp-subscribe@yahoogroups.com )
Re: TI DM642 DSP vs. Pentium - Jeff Brower - Nov 9 0:51:00 2003
|
Lisa- > True, but what I am trying to get a handle on is that > TI is big on their "BDTI benchmark" number, and claim > that you have to compare apples to apples, and that > even thought the DSP does not run as fast, it is > better suited for this stuff than a multipurpose > processor that runs faster. > > Your point about the heatsink is a good one for > example, where the Pentium would melt down without > one, the DSP is only a couple watts. I suggest to find TI and Intel benchmarks for something like 1024 x 1024 2D FFT for a 16-bit/pixel image, and convert numbers given in cycles into time using the maximum clock speeds. A usable 2D FFT would require the processor to access offchip memory (SDRAM or DRAM) which I think is very fair -- I would not use a TI benchmark where all data can be stored in onchip SRAM. I think the group would be very interested in what you can find. -Jeff P.S. I posted to the C6x group also, where there is often discussion about DSP + image/video. > --- Jeff Brower <> wrote: > > Lisa- > > > > > Hi all, I am new to the group, and would like to > > ask a question. > > > I assume you all are familar with TI's new DSP, > > the "digital media" > > > DM642. It is a member of the new C64x series DSPs, > > with clock speeds > > > up to 720MHz. > > > > > > My question is, how do you think this DSP will > > compare in machine > > > vision image processing applications to a > > "typical" desktop PC in the > > > 2 to 3 GHz range. > > > > > > I know I can't just compare clock speeds, and a > > lot depends on how > > > well the software is written, but in general, > > would this DSP be > > > slightly slower, about the same, or slightly > > faster (or something > > > else) ignoring any IO and memory bottlenecks. > > > > To ignore I/O and memory bottlenecks makes the issue > > purely "benchmark". You should > > be able to compare some benchmarks for typical > > algorithms published by Intel and TI > > and get some idea. > > > > But the real question is "what can you build". > > Issues of chip size, power > > consumption, I/O interfaces and throughput are > > critical. In our product line, the > > huge heatsink and fan make a 2.4 GHz Pentium > > impossible. > > > > Jeff Brower > > system engineer > > Signalogic |
(You need to be a member of imagedsp -- send a blank email to imagedsp-subscribe@yahoogroups.com )