Overclocking memory

[rAvEN^Rd]

I have had the opportunity to tamper with a couple of GF4 cards for the last couple of weeks. This has raised a couple of questions regarding overclocking of memory.

When I overclocked memory on these cards I though that hitting the theoretical max frequency would be a good goal to shoot for. I figured that the memory would have some margins on the seektimes specified but I also assumed that there would be some losses and that the max value would be hard to reach. I turned out I was dead wrong.

I overclocked some Samsung 4,0 ns memory. The theoretical max speed would be 500 MHz but I could push it as far as 530 MHz. Some other 4,0 ns memory would only hit 510 MHz but I was content since I had past the theoretical maxvalue.

I came across two boards with Samsung 3,6 ns memory. The theoretical max frequency is 555 MHz but these boards would go as far as 650 and 660 MHz. I was amazed!

When I got a board with 3,3 ns memory from Hynix I thought I would go past 700 MHz but it turned out I experienced visual anomalies when the card went past 600 MHz which is below the theoretical max.

I conclude that the difference in tolerance from different manufacturers makes the theoretical max frequency pointless since no good conclusions can be drawn from it.

Comments? Can anyone restore my faith? I want to be able to draw mathematical conclusions without being totally run over by uncertianty.

 

[pxc]

There are also other memory timings that affect the max speed. Command rate, burst rates, initial access and CAS rates vary between memory grades/speeds.

For example, 5ns DDR-SDRAM chips rated at 400MHz (200MHz DDR) could run CAS2.0 1T 5-2-2 timings at 366MHz (183MHz DDR), but only CAS2.5 2T 6-3-3 timings at 400MHz (200MHz DDR). Fast DDR memory is often rated like this. Unless you are able to adjust memory settings other than clock speed, you can't really tell what real maximum speed the memory is capable of. You would start off by reading the chip part number, then check the manufacturer's web site for a data sheet that will list speed ratings vs timing.

There is no simple relationship between ns rating and max speed if you don't know the chip timings used.

 

[Gunhead]

First, two 3.3 ns chips [let alone GF4 cards or anything else with several chips] from the same manufacturer, e.g. Hynix, won't be identical. Complex electronic devices, with much logic inside, them chips.

Second, do you know what the nanoseconds figure stands there for? It's not like the PC figure (calculated with some leeway) specifying a mnimum speed the memory will always run at...

Just being my helpful self.

 

[rAvEN^Rd]

Gunhead: I'm well aware that two samples from the same manufacturer will not be identical. That's part of my problem btw since that makes predictions so much harder.

Second, do you know what the nanoseconds figure stands there for? It's not like the PC figure (calculated with some leeway) specifying a mnimum speed the memory will always run at...

Please elaborate on this one. I'm not sure I understand what you mean.