DDR3 Latency 1

[NFI]

Hello,

can somebody explain to me why DDR3 has such high latency figures? I understand that this is symptomatic of it being a new technology and DDR2 suffered the same problem at it's inception, but what is it that causes this?

Predictably, DDR3 latency will drop dramatically as the new technology is developed, but what are the changes manufacturers will make which will allow this?

Many thanks,

Paul

 

[Nelviticus]

Eek! Nasty colours. Anyway ...

DDR3 has higher latency figures than DDR2 because it just does. It's a different technology and higher latency is one of its properties - DDR2 also has higher latencies than the original DDR.

One of DDR3's advantages, however, is that it's capable of much higher clock speeds than DDR2. As higher-clocked DDR3 becomes available it will begin to significantly out-perform lower-latency DDR2.

No doubt manufacturers will also begin to bring the latency down when they find cost-effective ways of doing so but as this has far less impact on performance than upping the clock speed it's not an area where they're likely to focus their research. I very much doubt it will drop 'dramatically'.

Regards

Nelviticus

 

[NFI]

<eek> ?

Honestly, is there any need for that?

Paul

 

[kmcferrin]

Since CAS latency represents the number of cycles to takes to access a given bit of memory, it's important to remember that the cycle time is relative to the clock speed. If you remember, you can run your memory at a lower clock speed and get a lower latency, or you could run it at a higher clock speed but you would have to back off the "memory timings" (latency settings). Memory running at 333 MHz (333 million cycles per second, or 0.0000000030 seconds per cycle) has a longer cycle time than memory running at 400 MHz (400 million cycles per second, or 0.0000000025 seconds per cycle). I get these numbers by dividing 1 second by 333 million and 400 million.

So assuming that the memory chips are in both cases made the same, and that the transistors take the same amount of time to switch in both modules (lets assume 0.0000000090 for the sake of argument) then that 333 MHz memory could manage the change in only 3 cycles (0.0000000030 x 3 = 0.0000000090) but the lowest number of cycles that the 400 MHz memory would require to get to 0.0000000090 would be 4 (0.0000000025 x 4 = 0.0000000100).

So you can see that as clock speed goes up CAS latency will always go up along with it. As time goes by, memory manufacturers are able to improve their production process and manufacture memory modules that are capable of switching faster, which means that CAS latency at a given clock speed will go down. Which usually means you can start cranking the clock up again. Just keep in mind that it's all relative. CL6 on 1333 MHz memory isn't the same thing as CL6 on 800 MHz memory.

 

[NFI]

Good point, well explained

So CAS latency is a feature of transistor manufacture; are the differences between DDR2 and DDR3 p nelviticus) at the silicon level fairly pronounced, then?

Thanks for your time,

Paul

 

[cdogg]

kmcferrin did an excellent job explaining that. Just remember that if you had DDR2 and DDR3 running at the same clock speed (say 800MHz), DDR2 will outperform it every time. In fact, the same was true when DDR2 first came out at 400MHz speeds (DDR1 dominated DDR2 at that speed).

Like it was already said as clock speed climbs out of the reach of DDR2 technology, then DDR3 becomes not only the better solution but the only solution! That's why it's there!


~cdogg
"Insanity: doing the same thing over and over again and expecting different results." - Einstein
[tab][navy]For posting policies, click [/navy]here.