Penguin
Note: You are viewing an old revision of this page. View the current version.

Lower = faster = better

RAM is organized into rows and columns, and is accessed by electrical signals called strobes, which are sent along rows to the columns; when data is needed, the CPU activates the RAS (Row Access Strobe) line to specify the row where data is to be found (high bits), then, after a short time, the CAS, or Column Access Strobe, to specify the column (low bits). After that, the data goes to the output line and to its destination on the next clock tick. In other words, the Column Address Strobe dictates how many clocks the memory waits before sending data on. All registers should be full, or errors will result, which means a longer wait to make sure, and slower operation. The shorter the cycle length, the faster the machine runs, at the expense of stability and data.

Linked with CAS are RAS and RAS-to-CAS, usually set to 2 or 3 with SDRAM Cycle Length, although you may be able to set them independently, and preferably in the reverse order to the above. Numbers on the chip looking like 3-2-2 refer to CAS, RAS-to-CAS and RAS, respectively. Running the chips at higher than rated speeds will mean dropping a CAS/RAS level.

Anyhow, with PC100 SDRAM, the first transfer takes about 50 ns, and the remaining three inside one cycle, assuming burst mode is active and they are in the same column. If not, the extra time is determined by CAS Latency, or the ratio between column access time and clock cycle time, derived from dividing the former by the clock frequency, and rounding up to the next whole number.

To get the maximum theoretical speed of any memory, divide 1000 by the access time. For example, 7ns = 1000/7 = 143 MHz. The combination of RAS and CAS specifies a particular RAM location in a particular RAM chip, where they intersect. Unfortunately, a lot of time is taken up with transferring these values rather than data. Rather than have separate pins providing power and data for both, each pin does double duty, serving rows or columns according to whether the RAS or the CAS pin is being asserted (that is, receiving current).

Your system will operate most efficiently when the RAS and CAS timings are optimized, but you lose stability as speed is gained. With page mode, any column of DRAMs in a row can be accessed any number of times within a short period; since the row is already specified, only the CAS needs to be applied on subsequent memory accesses, making things quicker.

RAS and CAS are measured in nanoseconds; the lower the value, the faster the RAM can be accessed, so the T state delay is similar to wait states. The RAS access time is actually the speed rating marked on the chip; CAS access time is around 50% less. Generally, choose the same speed for DRAM reading and writing, with as few wait states as possible. Burst cycles work the same way as they do for SRAM, consisting of four figures, with the first being larger because that's where the address is read; the remaining figures indicate the clock cycles for the reading of data. They might look like this on the screen:

x222/x333

The first set would be for EDO and the second for Fast Page Mode RAM. The 430HX chipset can use lower figures than the VX. The idea is to keep the figures as low as possible, consistent with your machine working properly. Note that EDO is only faster when being read from; writes take place at the same speed as FPM RAM.