I've been doing research to find how much faster register access is compared to primary memory. I find the ration is about 100 times faster, can that be correct, has it always been about that number of is this number increasing or decreasing?

I read that the microprocessor can access the register file twice every clock cycle and that a primary memory access can take 50 ns. A modern computer microprocessor's clock cycle is abt 4 GHz, which is about 10^6 cycles per second i.e. about 1 cycle per nanosecond (?)

Are my number appromimately correct or am I wrong by factor 10 somewhere?


I think it's a little less than that, but the ratio has been getting higher over time, at least from the 70s onwards.

1 cycle per nanosecond = 1GHz. 4GHz is right on the high end and is four times as fast. However a latency of 50ns is a bit high as an estimate, this article on latency is quoting numbers around 10ns in 2009. If I use that number I get a ratio of 80 times as fast; I suspect real world numbers will be 50-100 depending on configuration.

This is why having lots of cache on-die with the processor makes such a big difference.

  • 1
    \$\begingroup\$ The Column Access Strobe Latency from the article you linked is not the total latency of a memory access. Aside from assuming a row hit, this does not account for the transfer time (typically four DRAM cycles)--not supporting critical word first can simplify the design--and ignores overhead in cache tag checking, the on-chip network, the memory controller, etc. 50ns is probably about right. \$\endgroup\$ – Paul A. Clayton Jul 17 '13 at 12:52

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.