CF Data Transfer Bandwidth

Hi Sandor,

I have not performed any speed tests but have looked at the Verilog & SOPC Builder timing used to implement that component. It is a very simple interface, with setup/hold/wait-state delays in accordance with the CF specification, using true IDE mode (http://www.sandisk.com/pdf/oem/cf-manual-10.7.pdf). Thus I would not expect the interface to go much slower than CF would natively allow.

However, I have been doing some recent work with CF and have learned a few things:

First, as you mention, there are differences in cards. I have a large 'ultra' card in my digital camera that allows me more continuous 'raw' (6megabyte) shots than a conventional card (the camera can take a few frames a second, buffer to RAM, and then writes to CF...since CF is the slowest part of this eventually it becomes the bottleneck and I have to wait before taking another exposure. I haven't look at the specs of these faster/newer cards but the difference is visible to me.

Second (and perhaps more important), I think performance depends on how efficient the host side software is in its access to the cards, especially if a file system is used. From reading this (http://www.sandisk.com/pdf/oem/appnotecfhostv1.0.pdf) document recently it seems that traditional IDE rules apply: if the file system becomes fragmented, performance goes down. If repeated small file accesses are perfrormed (rather than large files), performance goes down, etc.

If I get to doing any throughput tests (I'm actively working with CF a bit this week) I'll try to post something further.

--- Quote Start ---

originally posted by jesse@Feb 16 2005, 12:03 PM

i have not performed any speed tests but have looked at the verilog & sopc builder timing used to implement that component. it is a very simple interface, with setup/hold/wait-state delays in accordance with the cf specification, using true ide mode (http://www.sandisk.com/pdf/oem/cf-manual-10.7.pdf). thus i would not expect the interface to go much slower than cf would natively allow.

--- Quote End ---

Thanks, Jesse. I was involved in doing some evaluation on the CF component and the read data transfer rate was in the neighbourhood of 100k/second, very low. Now, this was a very simple test and may not have been done under optimal circumstances, so I just wanted to know if it's worth putting more effort into getting the test right to try to crank up the speed. I had a look at the verilog code for the CF but I'm a "VHDL man" so all I could see was that it does true IDE mode. You just confirmed my suspicion that it should be able to run much faster.

--- Quote Start ---

originally posted by jesse@Feb 16 2005, 12:03 PM

if i get to doing any throughput tests (i'm actively working with cf a bit this week) i'll try to post something further.

--- Quote End ---

If it's not too much trouble, that would be greatly appreciated.

Thanks again.

Sandor

Hello Sandor

If I remeber correctly, I tried doing a raw data read bandwidth test, meaning, I was just reading sectors as fast aspossible, using the microtronix CF component. The results I got were around 6 Mbit/sec. This , however is only raw data read, with no file system. An MP3 design working from the CF can work with files of up to 160kbps, but I think the bottleneck is not the CF card.

Regards

Nir