1

Let's say that we're looking at a DRAM with typical timing parameters (those of interest are listed below) with read and write latencies of ~20-50 nanoseconds. How would one go about changing the timing parameters to cause the read and write latencies of the DRAM to increase? Specifically, let's say that we wanted read and write latencies of ~1 microsecond. The parameters I have to work with are the following (if others are important and not listed, please mention them; any of these which are unimportant or which don't make sense can be ignored):

* tCCD = CAS to CAS command delay (always = half of burst length)
* tRRD = Row active to row active delay
* tRCD = RAW to CAS delay
* tRAS = Row active time
* tRP = Row precharge time
* tRC = Row cycle time
* CL = CAS latency
* WL = Write latency
* tWTR = Write to read delay

The reason I'm asking is that I want to do a small simulation to investigate the effect of varying memory access latencies on program performance for various memory access patterns. My knowledge of memory hardware is woefully limited; given the parameters above, I'm thinking that memory latency for a single access would be something like tRAS + CL to select a row/column, plus WL? I really apologize if these aren't common timing parameters. Thanks in advance!

EDIT:

Come to think of it, if I have a set of timing parameters p1, p2, ..., pN, with know read/write latency X, can I get a new set of parameters p1', p2', ..., pN' for a desired read/write latency Y > X by taking p1' = (Y/X)p1, p2' = (Y/X)p2, ..., pN' = (Y/X)pN ? It seems like I should, since if read/write latencies are some linear combination of underlying DRAM timing parameters, I should be able to simply scale the parameters and get an equivalent scaling in the derived quantities... right?

2
  • I could not possibly answer, but aren't there limitations to most of the bios control to be able to achieve such slowness? Could it possibly be easier to set up , ram working so hard, that the thing your testing would break similar to the ram responding slowly via timing? While it would in no way be consistant, testing long enough might show how slow responces would act?
    – Psycogeek
    Feb 6, 2012 at 19:07
  • @Psycogeek This is really for the purposes of simulation... I have no intention of actually trying to underclock a DRAM device to 1 microsecond latencies. My real difficulty is in understanding how read/write latencies are related to (or scale with) the underlying timing parameters of DRAM devices, such as the ones I list in my post... sorry if this is confusing.
    – Patrick87
    Feb 6, 2012 at 19:26

0

You must log in to answer this question.

Browse other questions tagged .