• State
  • Locked Locked
  • Replies 1 reply
  • Subscribers 98 subscribers
  • Views 290 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

Understanding of Cycles consumed in SRIO throughput document and example tput_benchmarking example project

RCReddy
Genius 3575 points

Hi All,

          I am confused with cycles consumed [concept]

1. According to the document [Table 25][c66x Throughput performance guide],DirectIO: - 4096 bytes of packet, Total time taken [No time units mentioned in the .pdf] is around 40067.

whereas in tput_benchmarking example project [MCSDK .21]: - directioIO Nwrite [4096 bytes], average cycles is just 3455 cycles.

2. According to the document [c66x Throughput performance guide], Type 11 : - 4096 bytes of packet, average cycles consumed [88997],

whereas in tput_benchmarking example project [MCSDK .21]: - Type11 [4096 bytes], average cycles is around 4587 Average cycles.

can some Ti member please help me in understanding these two figures [one in .pdf throughput user guide and other one obtained through tput_benchmarking example project].

tput_benchmarking are for loopback versus c66x throughput performance benchmarking[this i assume is measured from board to board].please clarify

i am running this test [loopback] on c6670 board with mcsdk .21 and ccsv5.

Thanks

RC Reddy

  • 0
    TI__Mastermind 30825 points

    The Throughput performance guide is using 8 LSU's to transfer 16 transactions of size 4096B each.  A total of 565248 bits are transferred (including packet overhead) over the measured time.  This is done to get a better average throughput when multiple transfers are being executed in parallel.  This reduces the overhead latency associated with single transfers, to get an overall throughput number.

    I'm not certain of how many packets of size 4096B where transferred in the MCSDK example, but I'd assume it's documented in the MCSDK example.

    I hope this clarifies some things.

    Best Regards,

    Chad