Hi Philip, How does the conclusion be made that ARM can not swallow the current max data transfer rate? I need to build a project that need to process 60MB/s data, so any way to achieve my goal. Use a more powerful CPU or use dsp on the omap?
On 5/25/12, Philip Balister <phi...@balister.org> wrote: > On 05/24/2012 09:46 PM, Page Jack wrote: >> Thanks Ben, >> does e100 use EMIF to transfer sample data between FPGA and ARM? If so >> the >> data rate should be able to improved. >> Anyone have tried to improve the data rate? > > EMIF is basically identical to GPMC. The interface uses DMA to move data > in 2K chunks between the FPGA and memory. This is the largest transfer > possible due to how we connected the address and data lines > > My impression of the current limiting factor is interrupt response time. > There is probably some room for small improvements, but as Ben notes, we > are already collecting data faster than the ARM can swallow it. > > Philip > >> >> Regards >> >> On Thu, May 24, 2012 at 9:01 AM, Ben Hilburn <ben.hilb...@ettus.com> >> wrote: >> >>> The CPU sets up the initial DMA parameters, but from then on, it's pure >>> DMA. No CPU is required. >>> >>> Cheers, >>> Ben >>> ---------------------------- >>> Ben Hilburn <http://goo.gl/5DdZ3> @ Ettus Research, >>> LLC<http://www.ettus.com/> >>> >>> >>> >>> On Wed, May 23, 2012 at 5:55 PM, Page Jack <jack.page...@gmail.com> >>> wrote: >>> >>>> Thanks, does the ARM memory bus use DMA or it eat cpu? >>>> >>>> >>>> On Thu, May 24, 2012 at 5:06 AM, Ben Hilburn >>>> <ben.hilb...@ettus.com>wrote: >>>> >>>>> Page - >>>>> >>>>> The memory bus to the ARM provides 40 MBytes / second. This is used >>>>> for >>>>> streaming samples, as controlled via software. Currently, UHD supports >>>>> 16 >>>>> bit and 8 bit samples for TX & RX. The GPMC can only going to talk to >>>>> one >>>>> slave at a time; the possible slaves are TX, RX, and ethernet. So you >>>>> can >>>>> only be sending TX samples, receiving RX samples, or communicating via >>>>> ethernet. >>>>> >>>>> Thus, doing the math with the numbers above, you can stream: >>>>> 16 bit I, 16 bit Q -- Total: 32-bit samples -- @ 10 MSps >>>>> 8 bit I, 8 bit Q -- Total: 16-bit samples -- @ 20 MSps >>>>> >>>>> What you choose to do with this data is obviously up to you. It is >>>>> very >>>>> easy to try to do more processing than the ARM can handle, in which >>>>> case >>>>> samples will start getting thrown out by UHD. Thus, you can typically >>>>> process between 4 and 8 MHz of baseband bandwidth, depending on your >>>>> application. If you are willing to dig deep into the code to make NEON >>>>> and >>>>> C64 optimizations, you can improve the performance dramatically. >>>>> >>>>> Cheers, >>>>> Ben >>>>> ---------------------------- >>>>> Ben Hilburn <http://goo.gl/5DdZ3> @ Ettus Research, >>>>> LLC<http://www.ettus.com/> >>>>> >>>>> >>>>> >>>>> On Tue, May 22, 2012 at 7:47 PM, Page Jack >>>>> <jack.page...@gmail.com>wrote: >>>>> >>>>>> Hi, >>>>>> I want to know the overo model used in e100 and the largest data >>>>>> transfer rate between fpga and overo in e100. >>>>>> >>>>>> Regards! >>>>>> >>>>>> _______________________________________________ >>>>>> USRP-users mailing list >>>>>> usrp-us...@lists.ettus.com >>>>>> http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com >>>>>> >>>>>> >>>>> >>>> >>> >> >> >> >> _______________________________________________ >> USRP-users mailing list >> usrp-us...@lists.ettus.com >> http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com > _______________________________________________ Discuss-gnuradio mailing list Discuss-gnuradio@gnu.org https://lists.gnu.org/mailman/listinfo/discuss-gnuradio
