RE:Maximizing performance with the DES example

[vitorbal]

@Vitor–

Have a look at page 15 of the FrontPanel User’s Manual. There are some measurements taken from PipeTest that will help guide you.

DESTester was certainly not written for speed. It would be much faster to stream 8 MB at a time through a BlockRAM-based FIFO and allow the DES module to fill a similar FIFO on the other end which then stores the result in SDRAM. When the SDRAM is full, use PipeOuts to transfer the results back to the PC.

USB is a fast bus when used for bulk data. But like any bus, a performance penalty is associated with smaller transfers.

The Spartan-3 FPGA has a fair bit of BlockRAM, but the SDRAM is definitely helpful to keep transfer sizes nice and big to get better performance.

Thank you for the suggestion,
I had already taken in consideration using SDRAM, and that will be my next step.
Regardless, going back to the 20mb example I used, considering the time taken on pipes already takes in account the time that it takes to setup the pipe connections, what other factors do you think are responsible for the remainder time of 661 milliseconds? This shouldn’t all be processing time, seeing as acording to the calculations I presented, the processing time for a 20mb file should be around 1/40 of a second = 0,025 seconds. Any ideas?

//Vitor

[okSupport]

It’s hard to say from this perspective and not looking at your code intently to see how your measurements were made.

My suggestion would be to revisit your measurements and make sure you’ve timed things correctly. I suspect there may be a flaw in your measurements somewhere – as is usually the case when something ends up not being equal to the sum of its parts.