Problems? See this thread at archive.org.
With the introduction of DDR3 seems many have let common sense go out the window. I mean this in regards to what does it mean. DDR3 means lower power consumption, 1.5v vs 1.8v for DDR2. DDR3 is able to achieve higher clock speeds because of the increased I/O x8 vs x4 of DDR2.
That is it. DDR2 CL4 vs DDR3 CL4? Other than voltage/power consumption being lower are identical in performance. Much has been misstated eluding to DDR3 some how having special properties? A cycle is a cycle is a cycle. It was yesterday it will be tomorrow. And of course it is today.
Below is a chart I made showing the amount of time the different RAM speeds take to retrieve the first bit and the amount of time A byte would take. Notice that CL has the largest hit on the first bit? But with the following bits the memory clock is the most important.
Time for first bitTime for a byte (8bits)
Mhz CL ns/cycle total time (ns) 533 4 1.875 7.5 666 4 1.5 6 667 5 1.5 7.5 800 4 1.25 5 800 5 1.25 6.25 1066 5 .938 4.69 1066 7 .938 6.566 1066 8 .938 7.5
Other than 1 bit requests higher speeds are more important it would seem? I used an example of a byte to show how speed can over come CL very fast. I am not a computer scientist so I have no knowledge of what a standard/average request might be? But I doubt 1 bit requests are the average. I read something about 64bit being the minimum for many. That would have the faster crushing slower even with lower CL. For those with limited knowledge of latency and what it means wiki has a good overview on it.
Mhz CL ns/cycle total time (ns) 533 4 1.875 20.625 667 4 1.5 16.5 667 5 1.5 18 800 4 1.25 13.75 800 5 1.25 15 1066 5 .938 11.256 1066 7 .938 13.132 1066 8 .938 14.066
In closing there is no fundamental differences between DDR2 and DDR3 in the way they function. The relationship between clocks and CL is the same.
-
-
-
-
Those are still a lot of confusing facts, let's talk in practical terms, the power save, is it sagnificant? What are we talking about here, 15 minutes added to battery time? An hour?
Also the speed, I'm still confused about that, in gaming terms or whatever, like, playing crysis, if I have 2gb of memory, how much, if any, difference will it make? Are we talking 2 fps, 10 fps?
How can you measure the difference in practical terms? The whole clock rate thing isn't really working for me, I want to play crysis, not fetch the byte with my memory stick
-
-
-
CL only is a penalty for the first bit of data retrieved on a request all following bits only have a one cycle cost. So 800Mhz is 1.25ns and 1066Mhz is .938ns. So for all sequential bits read in that request it is the speed not CL. So lets say for example we use a 1KB request? 1024B x 8 (to get bits)= 8192 bits? OK 800Mhz CL5 is 6.25ns 8191 bits to go? 8,191x1.25ns=10238.75ns + 6.25ns=10,245ns? 1066Mhz CL8 is 7.5ns 8,191x.938 ns=7683.158ns + 7.5=7,690.658ns.
So for a 1KB request 800Mhz CL5 takes 10.2micro sec and 1066Mhz CL8 takes 7.68micro sec. So which is faster? That is the advantage of DDR3 clock speed nothing more. But it is an advantage. And the larger the request the greater the advantage. Remember the advantage with these two goes to 1066 CL8 as early as 1 byte 14/ns vs 15/ns.
Clear advantage to faster clock.
I know these are just theoretical and will not hold up exactly in practice because of other factors but I do think the demonstrated trend will hold up. Also I think the consensus on PC5300 CL4 vs CL5 not making much of a difference further supports the relatively inconsequential importance of CL in the real world? Back to the penalty, it is only the 1st bit of request. and that penalty in the example is less than 3ns? 3ns? Even with stopwatch cannot tell the difference.
Any flaws in my math I would be happy to hear about as well I am no mathematician.
DDR3 will end up over taking DDR2 just as DDR2 over took DDR. It is the exact same situation. Faster clocks higher latency. All concern on that have subsided. One should I think not have any sense of wonder or confusion as all the relevant information is there.
That is my guess! -
-
Wow! Way to go John! You found what I have been waiting for. Not as high as I had hoped. I made speculation off these desktop bandwidth benches, link. But well if anyone bothers to read thru the importance of the memory controller is critical (part of chipset on Intel's). Until tests prove different I am going to assume the relative nature of my links bandwidth will hold up. So I expect the same ~25% improvement over Santa Rosa with 800Mhz. But clearly an improvement for those who had doubts.
For those new to this (have not wasted hour upon hour contemplating the implications) ~25% in no way translates into ~25% improvement in overall system performance. In many circumstance even most it may translate to little to no real world improvement, but well, that said more is better. And that is a dig at those who say "Dual Channel" does not matter anyway ~10% on Intel.
For all you who are new to things that well are esoteric and irrelevant to 99% of the average computer user. But sadly important to a much larger % of NBR (myself included). Having a faster FSB than RAM is not all bad. I was just reviewing benches from John and others I have saved. With Napa 667Mhz FSB and 667Mhz RAM bandwidth 3500MB/s (Wolfraider). John, Santa Rosa 800Mhz FSB and 667Mhz RAM 4000MB/s. Yes different chipset so different memory controller and that can have some affect. But the main affect is the faster FSB improves bandwidth. Desktop guys have known for years.
John what are the timings? That could explain the less than 6000MB/s? And if that is the case (high CL) then the 800Mhz CL4 PNB has could really rock!
I learned last week and I was confused myself. DDR2 1066Mhz is a JEDEC standard. How that could affect Montevina I don't know but if it would support at speed would be another reason not to worry about getting DDR2 at this point. Here is the link to the standard.
Too much time on my hands.
-
-
OK. Here's a result for the DDR2-800 with Montevina in the humble Samsung Q210: 5.41 / 5.43GB/s. The RAM is reported as
So, based on these two results for Montevina + DDR2 and DDR3, DDR3 is winning, by a magnificent 4%.
Perhaps the real victory is Montevina over Santa Rosa. It will be interesting to see the bandwidth for Montevina + DDR2-667.
John -
-
-
-
Can you expand on that. I am not sure at this point what exactly you mean. But I do know it can insofar as the memory controller does certainly have an effect and on Intel's that is a part of the chipset.
-
-
An interesting analogy.
However, there's a crushing / screening plant between the gravel pit and the customers. So although you've got more / bigger / faster trucks you also need to increase the processing capacity. That's been done, but not to the extent that it's not a bottleneck during periods of high demand, because for most of the time it's an under-utilised investment.
John -
Yes I know there are onsite equipment breakdowns and worker injury's some resulting in death. These do prevent the pit from reaching it's theoretical max. But are you buying my concept at all?
-
-
-
DDR2/DDR3 RAM guide
Discussion in 'Hardware Components and Aftermarket Upgrades' started by powerpack, Aug 2, 2008.
