P4P800 memory question

[Darkfalz]

Is memory ever more stable at faster timings by increasing VDIMM (SDRAM
Voltage)? Or is it not going to make one iota of difference?

 

[Paul]

Is memory ever more stable at faster timings by increasing VDIMM (SDRAM
Voltage)? Or is it not going to make one iota of difference?

It makes enough of a difference, that even the manufacturers do it.
Memory chips sold as PC3200 are specified in the datasheet as needing
2.6V, whereas slower ram runs at 2.5V. The manufacturer decided
that a boost of 0.1V was required for reliable PC3200 operation.

Overvolting is OK up to a point, but too much could shorten the life
of the ram. (I'm not aware of any studies that relate overvoltage
to decreased lifetime - electromigration is one possible wearout
mechanism.)

Paul

 

[Darkfalz]

It makes enough of a difference, that even the manufacturers do it.

Memory chips sold as PC3200 are specified in the datasheet as needing
2.6V, whereas slower ram runs at 2.5V. The manufacturer decided
that a boost of 0.1V was required for reliable PC3200 operation.

Overvolting is OK up to a point, but too much could shorten the life
of the ram. (I'm not aware of any studies that relate overvoltage
to decreased lifetime - electromigration is one possible wearout
mechanism.)

Eh I just tried it and even at 2.85v it didn't let me use higher memory
timings. Worth the experiment, I suppose

Actually my ram is okay at CAS 2.5 (it's CAS 3) but I didn't see any actual
performance increase from that, so I just left it at SPD.

 

[Paul]

Eh I just tried it and even at 2.85v it didn't let me use higher memory
timings. Worth the experiment, I suppose

Actually my ram is okay at CAS 2.5 (it's CAS 3) but I didn't see any actual
performance increase from that, so I just left it at SPD.

I think the value might be in allowing you to use a higher frequency.
It would be pretty hard to get a significant improvement in memory
timings by doing that. In terms of memory bandwidth at PC3200 speeds,
clock rate can matter more than setting memory timings from 3-4-4-8
down to 2-2-2-5. At lower frequencies, the memory timings are more
important, but above PC3200, clock rate makes more of a difference.
That is why people are happy to buy PC4000 memory with 3-4-4-8 timings,
because of the 25% boost running at DDR500 instead of DDR400.

The main value of the voltage boost may be coming from the driving
of data on the memory bus outside the memory chip, as opposed to
improving things inside the memory chip. I don't know of an easy way
to separate the two effects, to prove or disprove that as a theory.

It took about 20 minutes with a search engine, but I found a paper
on a DRAM design:

http://www.cisl.columbia.edu/courses/spring-2002/ee6930/papers/00799866.pdf

"The chip is powered from a 2.5-V external supply, which is
regulated internally to 2.1 V for support circuitry, 1.5 V for
the array, and 1.0 V for RWD."

This implies that overvolting can only influence the periphery (I/O)
of the ram chip and not the core behavior. So, overvolting improves
communication to the DDR SDRAM chip, without changing what goes on
inside it. At least that is my interpretation...

Paul