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.

TMS570LS3137 performance drop when executing from external memory

Context : CCS Version 6.0.1.00040 under windows 7/64 - TMDS570LS31HDK

Hello,

When executing some code hosted in the external SDRAM from  TMDS570LS31HDK we see a performance drop from a factor 8.3 compared to an execution of the same code in internal flash.

Is this performance drop normal, or do we need to optimize our EMIF settings ?


Is this performance drop only due to EMIF or could it be better if using another type of RAM memory (MRAM instead of SDRAM for example ?)


Thanks for your support,

Regards,

Christophe

  • Hi Christophe,

      Running code from SDRAM is expected to take many more cycles than running from internal memory. Normally fetching one 32-bit instruction from internal RAM takes just one cycle. Fetching a 32-bit instruction from flash memory is also 1 cycle if the code is sequential since the flash controller is built with an internal prefetch buffer to prefetch the subsequent sequential instructions in advance. The flash and SRAM are tightly coupled to the CPU for fast access. 

      Running code from EMIF means the CPU needs to fetch the data via its Level 2 AXI interface and then through the on-chip interconnect before reaching to the EMIF interface. This can take some cycles to send the command to the EMIF. In addition, the EMIF interface is only 16-bit. If you have a 32-bit instruction or data to read then it needs to take two 16-bit EMIF accesses to complete. 

      In my opinion running code from SDRAM is not suitable for LS31x device if you are performance critical. SDRAM is more suitable for data storage. In our LC4357 device with the on-chip cache, running code from SDRAM will have minimal impact to the performance since the code will be cached. As long as you have a cache hit, the access to the instructions and data will be one cycle. 

  • Hi Charles,

    You recommend to use the LC4357 device instead of the LS3137 on the previous case. I have some questions about them:
    - Are these two devices built using the same technology?
    - The 4357 device is not yet ACTIVE, can you explain me why? Is it still under qualification? I read on another thread that you expect it to become active on September, is it still correct?

    Regards,

    MCV
  • Hi Mario,
    I just replied you in another thread.

    Yes, both LC4357 and LS3137 are built off of the same 65nm technology.
    Yes, LC4357 is expected to become active by September. It is still under qualification.