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.

AM2434: CPU reading and writing DDR speed is very slow?

Part Number: AM2434

Hi Experts,

I am asking for my customer urgently. Thanks.

We use for traversal function (one time 4 bytes are int type) for testing, both use cache and CCS optimization level opened to L2 level,

1.Code runs in SRAM, but data would in SRAM and DDR:

SRAM  to SRAM 666MB/s
DDR to SRAM 122MB/s
SRAM to DDR 308MB/s
DDR to DDR 90MB/s

2.Code runs in DDR, and data also DDR:

DDR to DDR 89MB/s.

You can see the test data above, when the data and code both SRAM, it would be It is seven times that of DDR.(666MB/s Vs 89MB/s)

Also CPU reading and writing DDR speed very slow. Because we see the DDR up to support 1600MB/s. Could you give me suggestion here? 

  • Hi Shaoxing,

    According to the AM64x System Level Benchmarking: AM64 System Level Benchmarks - Sitara - Confluence (ti.com), the R5F to DDR data transfer latency is about 100ns per 4 bytes (a 32-bit word), that is, 1000000000/100 = 40MB/s, which is even slower than your measured number (90MB/s). There are two factors which will affect the throughput:

    1. The usage of cache. How do you set up the DDR regions in R5F MPU, if the DDR areas are set to cacheable in MPU, then the code and data will be cached at read/write which will improve the performance greatly.

    2. How big is the chunk size when you transfer the data. When you use the chunk size bigger than 4B, then the performance will increase too.

    Best regards,

    Ming