Do all my memory modules have to run at the same speed?

tundrawolf86

Honorable
Oct 2, 2013
20
0
10,510
Hi, I'm going to be first time builder. The motherboard I'm going to buy has 4 memory modules.

I'm going to start with a single, cheapie 8 gb stick, to preserve the option to go up to 32 gb.

In the future, if I buy a second, more expensive, 8 gb stick, and overclock it, does that create problems with the old stick running slower?

Does OCing memory provide any noticeable (user experience) performance increase?
 
Solution
Yes, they all need to be running at the same speed.

I recommend purchasing a 16 GiB set of two sticks (2x8GiB) to start with. This will grant you 16GiB running in dual-channel mode of operation. Later on you can add an identical set for a total of 32 GiB. You may have to do some tweaking to get all 4 modules running nicely together, but this is rarely needed on DDR3.
Yes, they all need to be running at the same speed.

I recommend purchasing a 16 GiB set of two sticks (2x8GiB) to start with. This will grant you 16GiB running in dual-channel mode of operation. Later on you can add an identical set for a total of 32 GiB. You may have to do some tweaking to get all 4 modules running nicely together, but this is rarely needed on DDR3.
 
Solution

tundrawolf86

Honorable
Oct 2, 2013
20
0
10,510
Thanks very much for your reply. Now, if I may ask, what does "dual channel" mean, and why is that good?

Also, is all RAM overclockable? What is the "clock" referring to? What are the benefits of not buying the cheapest RAM?
 


Modern microprocessors contain memory controllers which have multiple fully independent memory channels.

Each memory channel consists of a DRAM controller and an IO bus.

Each memory module that you install typically has one or two ranks on it. Each rank is a combination of individual DRAM chips that, when their data pins are concatenated, form a 64 bit data bus. If a memory module has four or eight chips on only one side of the printed circuit board, it contains only a single rank. If a memory module has chips on both sides of the printed circuit board it contains at least two ranks.

Electrically speaking, there's very little difference between two single rank DIMMs installed in two DIMM slots on the same channel, and one installed dual-rank DIMM.

The IO bus joins all of the ranks across all of the DIMM slots on a channel together in parallel to a common DRAM channel controller. At most one rank per channel can be active at a time.

When more than one memory channel is present and populated it becomes possible to have multiple ranks active at the same time, as long as they are on separate channels.

Thus, a platform with a dual-channel memory controller has twice the peak bandwidth as a similar platform with a single-channel memory controller or a dual-channel memory controller with only one populated channel. The same logic applies to high end platforms with three or four memory channels.

You may have seen the term DDR3-1600 thrown around, this is one of the most common DDR3 standards which offers 1.6 gigabits of data transfer per IO pin per second. A 64 bit data bus (discussed above) would offer 1600*64 = 102.4 gigabits of data IO per second, or 12.8 Gigabytes per second. This is also known as PC3-12800. With two IO buses in parallel (dual channel) the bandwidth rises to 25.6 Gigabytes per second, and 51.2 Gigabytes per second when four IO buses are used in quad channel configuration.

So, dual-channel, triple-channel, and quad-channel platforms offer a lot more bandwidth than single-channel platforms.

The other benefit is that multi-channel configurations allow for large datasets to be received faster through the use of memory address interleaving. That's a bit much to explain here though.

As for overclocking SDRAM, the answer is "yes, but not quite". SDRAM can be overclocked, but it typically doesn't have the same amount of overhead as the CPU does. It can also get very unstable very quickly and doesn't necessarily yield a large increase in performance.

Overclocking SDRAM is most useful when its coupled with a CPU overclock. In the days of yore PCs lacked fine grained control of separate core and dram multipliers. Overclocking a CPU necessitated overclocking the DRAM as well.

The clock refers to the DRAM IO bus reference clock. The DRAM IO bus makes two IO transfers per clock cycle, one on each of the rising and falling edges. This is known as "double data rate" or DDR. The above mentioned DDR3-1600 makes 1600 transfers per second over an 800Mhz IO reference clock. The IO reference clock is also used to determine the memory module's core clock rate, which for DDR3 is 1/4 the IO reference clock. DDR3-1600 has a 200Mhz core clock. For DDR2 the ratio is 1/2, and for DDR the ratio is 1/1. This is why DDR2-1066 exhibits lower latency than DDR3-1066, the core clock is twice as fast.