|This is the talk page for discussing improvements to the DDR3 SDRAM article.|
|WikiProject Computing / Hardware||(Rated B-class, Mid-importance)|
- 1 Latency Table
- 2 How many pins?
- 3 Peacock terms in DDR3 vs. DDR2 pricing
- 4 re-write the Overview?
- 5 Why not QDR
- 6 Increased Latencies
- 7 Speed Record
- 8 wtf ddr3-2000 exist and its not listed.
- 9 Contradict tag
- 10 DDR3 above JEDEC standard ram
- 11 Extensions section
- 12 rank
- 13 Disadvantages compared with DDR2
- 14 it's not 2009 and it's not early 2010 anymore
- 15 Inappropriate tone
- 16 10600 vs. 10660 vs. 10666
- 17 DDR4 SDRAM
- 18 Power Usage
- 19 Damage
- 20 Burst implementation
- 21 Power Consumption
- 22 Maximum DIMM/SO-DIMM capacities?
- DDR4 better not come out for a while, because ddr2 came out not that long ago and now ddr3 is comming out, maybe intel is using the quick introduction to drive amd out of bussienes becuase they cant adopt a new memory that quickly.
Can we get a picture of this? Here's a nice shot but I don't know what the licensing is on it. I think this article would be much more effective with an image. --Anthony5429 18:25, 8 August 2006 (UTC)
JEDEC has finalized the DDR3 specs. I'm far too lazy to melt this with the article.
The Latency Table has both the Memory Clock speed and the Bus IO Speeds Listed. It's not obvious from the table if the Timings are specified based on the Bus IO Clks or the Memory clks. Shouldn't we explicitly state that they are based off the Bus IO Clks? DDR3 has an 8x ratio (effective speed: memory speed) while DDR2 has a 4x ratio to the memory speed, but both have the same 2x ratio to the Bus I/O speed. So DDR3-1600 (Mem: 200MHz, IO 800MHz) with a CAS of X clks has the same actual CAS ns latency as DDR2-800 (Mem: 200MHz, IO: 400MHz) with a CAS of X/2 clks. Calculating based on the Memory Speed will incorrectly lead to a belief that the DDR2 has half the nanosecond latency as the DDR3 module since the memory speeds are the same. —Preceding unsigned comment added by 220.127.116.11 (talk) 17:51, 19 May 2009 (UTC)
How many pins?
Slot-compatible w. DDR2?
Does not even say...
- ~DDR3 IS NOT backward compatable it runs on lower voltage than DDR2 and they have unique patterns and when wrongly inserted it can ruin the DIMM and the motherboard
Peacock terms in DDR3 vs. DDR2 pricing
Please don't hide the important facts: instead of telling us that DDR3 "generally" costs "much" more than DDR2, either show us some price comparisons, or show us a range of percentages that relate DDR3 prices to DDR2 prices. SixSix (talk) 16:52, 25 December 2007 (UTC)
- I agree. Newer products always cost more than the old generations of its family. Saying it "generally costs more" isn't adding any material to the article. —Preceding unsigned comment added by 18.104.22.168 (talk) 19:59, 4 January 2008 (UTC)
- "Newer products always cost more than the old generations of its family" <-- with ram in my experiance they do for a while, then they become about equal and finally the newer technology becomes cheaper (compare the price of DDR and DDR2 nowadays for an example of this last phase). Plugwash (talk) 22:12, 7 December 2009 (UTC)
re-write the Overview?
The overview section needs a rewrite to better accompany the excellent table underneath it. As you don't get a good sense of the sense of the functions of the internal sections of the device and their speeds and how they work together. Nothing too involved, but more educational! MrESaulved! (talk) 04:33, 15 March 2008 (UTC)
Under the "Increased Latencies" subsection of the overview it says, "The JEDEC standard latencies for the newer DDR3 memory are 7-7-7-15." I think that last timing (TRAS) is a typo and should be "7-7-7-20" per the CAS_latency article, right? Trumpetpunk42 (talk) 04:59, 11 April 2008 (UTC)
Why not QDR
Because the memory chips are (still) being accessed twice per tick of the memory clock. (Read the Double_Data_Rate article.) DDR2 and DDR3 basically just speed up the I/O bus clock in relation to the memory chip's clock. Trumpetpunk42 (talk) 04:59, 11 April 2008 (UTC)
Yes, this is correct.
If you compare DDR2-800MHz with DDR3-1600MHz, yes, the DDR3 one could get lower latency (MOst of time this is not true though). However if you compare DDR2-800MHz with DDR3-800MHz, you will find DDR3 has much bigger latency. It has to be. Period. Then go to the price point of view, DDR3 one is still much more expensive than DDR2 rigth now (Year of 2008)
NOTE: in Jedec spec, DDR3 CAS latency range from 5 to 11, DDR2 Latency range from 2 to 6.
Please look at this article: 
The best available JEDEC standard DDR3 modules are DDR3-1333 at 9-9-9-24 timings... a far cry from the best low latency DDR2-1111 4-4-4-12 timings - not to mention the DDR2-1250 5-5-5-15 modules almost matching DDR3 in speeds with still much lower latencies and I think history will repeat itself. The DDR2 camp may currently be laughing at the feeble latency and performance of the low end DDR3-1066 9-9-9-24 modules, but then later realizing that DDR3-1333 may be the point of parity - and that DDR2's doom may well be sealed by DDR3-1600 and above.
Sorry, its wrong.
Example: DDR2 with 800MHz at 5-5-5-15 means that one cycle (Memory Clock at 100MHz) need 1/100Mhz = 10ns. So multiplied by 5 you need to wait 50ns.
DDR3 with 1600MHz at 7-7-7-15 means that one cycle (Memory Clock at 200MHz) need 1/200MHz = 5ns. Multiplied by 7 equals 35ns.
Please think about [..] In order to increase the speeds of the memory modules with DDR3, it was also necessary to increase the latency of the modules. Latency is the amount of time that it takes for a memory module to process commands in a number of clock or command cycles. The higher the latency, the slower the memory will be at processing a command. [..]! Yeah, you're right, this statemant is totaly wrong. DDR3 has a lower latency than DDR2.
Yeah, it's the same story again as before with DDR/DDR2. All 3 (DDR, DDR2 and DDR3) have roughly the same latency, which is about 7 ns. (I'm not sure about the exact value). There was just recently a nice table in the german C't magazine, where this was clearly shown. (I can take a picture of it if anyone wants it). --Xerces8 (talk) 11:42, 22 April 2008 (UTC)
One more thing, the above calculation is wrong. According to DDR2_SDRAM the memory clock for DDR2-800 is 200 MHz. Also the latencies are given in the I/O Bus Clocks ticks , which is 400 MHz for DDR2-800. A typical CAS of 5 gives 12.5 nanoseconds. Which is the same as the CAS latecny of a typical DDR-400 RAM. Also DD3-1006 CAS7 gives 13ns. Just as DDR3-1333 CAS9. Hmm, strange that DDR3-1600 modules in shops have CAS of 7 and 8, even as low as 6... --Xerces8 (talk) 10:06, 28 April 2008 (UTC)
This section still has ambiguity.
Isn't it the case that a DDR3 1066 CAS8 will have double the latency of a hypothetical DDR2 1066 CAS4? So, in keeping with the idea of doubling up on the memory clock - in order to achieve the same external chip speed, DDR3 latency is doubled? (i.e. chip speeds of 133Mhz and 266Mhz respectively for the same bus clock, to give latencies of 14.4ns and 7.2ns?). Is this where the confusion lies?
This section seems to conflict (superficially) with the info on increased latency for DDR2 in the DDR2 article. Is there some way we can clear this up so the two articles mesh more smoothly?
Laptcd (talk) 09:21, 5 November 2009 (UTC)
I have removed the note:
"There is no corresponding reduction in latency, as that is a feature of the DRAM array and not the interface."
I believe DDR3 latency is better described as doubled over DDR2. i.e. you need twice the speed for the same latency (1600Mhz DDR3 CAS8 will give the same latency as 800Mhz DDR2 CAS4).
I dont want to replace the former note with an outright contradition of what the author was trying to get at, so I've removed the note until we have something less misleading. Hopefully no one will object.
Further, "reduction in latency". Wasn't the intended point that DDR3 gives no corresponding increase in latency? (as compared with DDR2 chips in a similar band)
Laptcd (talk) 17:35, 5 November 2009 (UTC)
The problem lies in the fact of the confusion of latency and timings. The four numbers listed for DDR2 and DDR3(and DDR) memory modules are not latencies. They are the number of cycles (expressed in the lower case 't') of the dimm clocks (tCK expressed in nanoseconds) that must pass for that particular action to be completed. For example:
DDR3 1600 CAS 9 has a lower latency then DDR3 1333 CAS 9 because the tCK (in nanoseconds) of DDR3 1600 is 1.25 and DDR3 1333 is 1.50 hence you get 10.5ns for 1600 and 13.5ns for 1333.
DDR2 667 CAS 4 has a latency of 12ns because the speed of the dimm clock (tCK) is twice that of DDR3 1600 at 3ns.
DDR2 clocks in nanoseconds (tCK): DDR2 400 is 5.00ns DDR2 533 is 3.76ns DDR2 667 is 3.00ns DDR2 800 is 2.50ns DDR2 1066 is 1.876ns
DDR3 clocks in nanoseconds (tCK): DDR3 1066 is 1.876ns DDR3 1333 is 1.50ns DDR3 1600 is 1.25ns DDR3 1866 is 1.07ns DDR3 2000 is 1.00ns
Two common memory speeds and timings DDR2 800 5-5-5-15 has latencies of 12.5-12.5-12.5-37.5 (expressed in nanoseconds) DDR3 1600 9-9-9-24 has latencies of 11.25-11.25-11.25-30
While DDR2 800 is not very far behind DDR3 1600 in latencies in this example, DDR2 800 is close to its limit latency wise. Only CAS 4 modules are available while DDR3 1600 has been available in as low as CAS 6 giving the DDR3 a 2.5ns advantage in latency for every timing which over the complete cycles of read or write allow DDR3 to have a lower overall latency then DDR2 but DDR3 increases bandwidth 200% (12.8GB/s vs 25.60GB/s respectively)! DDR3 2000 9-9-9-24 has 25% more bandwidth then even DDR3 1600 while the timings of DDR3 1600 would need to be 7-7-7-20 for comparative latency. DDR3 2000 CAS 9 compared to DDR2 800 CAS 4 the bandwidth increase is 250% with latencies just slightly lower.
"The world record of 2580 MHz was reached at latency settings of 9-9-9-24 using a Corsair Dominator DDR3 memory module..." - as of July 31st, 2008.
Should speed records be in the article? And what about higher-than-standard products? (1800MHz, 200MHz sticks etc.) PluniAlmoni (talk) 15:59, 1 August 2008 (UTC)
yeah, I don't really think overclocking and speed records are relevant to most people since in that test the entire board was frozen to -20c. No one interested in using a performance system regularly would do that (and only a few be able to) as it would damage a lot of the components on the mainboard after an extended period of time. Also that was using a 1gig stick which is pretty low, and not in dual or even triple channel, which is significantly faster than a single channel even overclocked to 2580. However, if you want to add a short paragraph somewhere in the article, and not another table, that could be nice.
wtf ddr3-2000 exist and its not listed.
- We only include official JEDEC specs... modules above this are considered OC and aren't standard. You may add it if you want, but there has to be a distinction.PluniAlmoni (talk) 23:15, 18 October 2008 (UTC)
Even if they are not official standards, they should be mentioned in the article somewhere, even if only to tell readers that they are not official. I came hear to find out about DDR3-1800, but there is not any info.--22.214.171.124 (talk) 21:01, 22 November 2008 (UTC)
- Have you really read the article and especially the listed specifications? Then you would have read the information about the so-called "specs": "specs"inbetween or above the official ones are home-brewed ones that use the officialy specification only as orientation for naming. --Denniss (talk) 23:55, 22 November 2008 (UTC)
the memory still performs at that speed, does it matter if it isn't official? Computer technology is a very fast moving industry, and this page is outdated if you don't let people update it, I thought that was the purpose of an online encyclopedia edited by everyone. I want to see the specs for the faster DDR3 ram, even if it is listed as non official. And not removed altogether as though it doesn't exist.
I'll add it in a new table then.
Feel free to edit the table however you like (I know you will) and if you could find out the cycle time for the last two modules that would be nice. Just don't completely remove the non-standard modules from the article, that is like claiming they don't exist at all.
Intro says latency is not improved over DDR2, but body says latencies have improved. If I've misphrased this and somehow this statement is true, it needs to be clarified because on its face it appears to be contradictory. Shadowjams (talk) 21:54, 17 February 2009 (UTC)
- I'll try to fix it. CAS latency is still around 10 ns, although there are very small (few percent) incremental improvements. The intro is pointing out that there has not been a latency improvement corresponding to the 2× bandwidth improvment. 126.96.36.199 (talk) 00:54, 2 March 2009 (UTC)
DDR3 above JEDEC standard ram
Why does Denniss continue to remove the faster DDR3 Ram table? This page does not belong to him, we all need to make compromises yet he won't. All the data in The table is accurate, real, and is labeled correctly, even as non-JEDEC standard ram. The table is very helpful, if Denniss wants it in another wiki he should make one and move it instead of deleting it completely. —Preceding unsigned comment added by 188.8.131.52 (talk) 10:32, 10 March 2009 (UTC)
- I agree with removal of that table. The main reason is that the specification and ratings for non-standard modules are basically up to the manufacturer to decide (and essentially, they can print whatever they want). By definition, there is no standard to hold them against, and there is inevitably a variation from one manufacturer to the next for any given module. Since we aren't here to catalog products, completing this table is out of the question. The article already notes that non-standard modules exist; I think it is perfectly sufficient to leave it at that, though perhaps with an additional sentence to describe the highest purported speed reached. Ham Pastrami (talk) 05:13, 17 March 2009 (UTC)
Why is it "out of the question"?
If I complete and compile the table what harm does it do having it as part of this article?
Why is it not valuable information?
What are you talking about "they can print whatever they want"? it's testable, you plug the module in, and you can set it to it's specified speeds. The standard to use are the common measurements of RAM, bandwidth and latency. And any variation between manufactures will be specified within these measurements. —Preceding unsigned comment added by 184.108.40.206 (talk) 10:36, 25 October 2009 (UTC)
I found it humorous that the Extensions section was written by Olin Coles, who references a website discussing DDR3 written by...you guessed it, Olin Coles. Not only that, but he uses the wrong acronym (It's called Extreme, not Extended).
I added a link to the XMP section under SPD timings, corrected the acronym, and replaced Olin's link with Intel's datasheet. I think we can all agree that Intel is the better source here. —Preceding unsigned comment added by 220.127.116.11 (talk) 23:16, 29 May 2009 (UTC)
can anyone explain what exactly "rank" is? the DIMM population requirements of the intel xeon datasheet give the impression that lower rank is better but don't really explain what it is. Plugwash (talk) 00:25, 8 December 2009 (UTC)
Disadvantages compared with DDR2
it's not 2009 and it's not early 2010 anymore
"Although DDR3 was launched in 2007, DDR3 sales are not expected to overtake DDR2 until the end of 2009, or possibly early 2010" —Preceding unsigned comment added by Quelrod (talk • contribs) 06:22, 23 May 2010 (UTC)
I feel this article is biased and authoritarian. For example "family of technologies" should be "group of devices" and the talk about damaging the DIMM or God forbid the motherboard too while trying to forcibly install DDR3 memories into old motherboads is annoying. Somebody fix it. Teemu Ruskeepää (talk) 16:29, 20 July 2010 (UTC)
- The tone of the two instances you have pointed out seem appropriate for an encyclopedic article. Also, the reference to the potential of damaging the equipment is accurate. No changes should be required. Coopman86 (talk) 08:31, 2 August 2010 (UTC)
10600 vs. 10660 vs. 10666
This article is out of date on DDR4. As best I can see, the new article is accurate so far, so this article's mention of DDR4 should probably be reviewed and brought into line with it. FT2 (Talk | email) 00:47, 26 April 2011 (UTC)
A user posted this same mistake else where that because of the lower voltage, ther eis less power consumed. This is simply not true: http://www.ocztechnology.com/drivers/DDR3_faq.pdf I think an expert in the feild (not a nine year old in their mom's basement) should check these facts. In this pdf it explain vyer simply that it is running at a fast clock speed, which negates the benefitt of the lower voltage but overall uses more power. The chips are not resistors! It is not a basic p=v*i or any of the similiar relationships. For example, DDR3 has several power saving features DDR2 doesn't. Flipping the state of a bit is also not the same as power through a simple resitor, etc. I think the citation of an article from a maker of DDR3 is a better source than an OC mag article. Another Article: http://www.xbitlabs.com/articles/memory/display/ddr3_13.html Desktop memory: According to kingston, 2Gb of DDR2 Pc8500 1066Mhz memory uses 1.584 watts per module
Kingston has a 2Gb DDR3 1066MHz memory card which uses 1.890 watts.
- Please check your DDR3 module, that's actually an 18-chip ECC module. It's also very unreliable to compare rebrander-RAM due to the multiple chip and module types they use. --Denniss (talk) 22:45, 1 March 2012 (UTC)
I would like to verify that my understanding of this is correct. Taken at face value, the formula for calculating bandwidth (MT/s) implies that you can improve transfer rate merely by slowing the internal memory clock relative to the bus. So to understand how doing so actually results in improvement (e.g. over DDR2), is it correct to say this slowdown is inversely matched by an increase in parallelism? By my calculation, each DDR3 memory access would effectively (pre)fetch 64*8=512 bits, as opposed to DDR2 where only 64*4=256 bits would be fetched. Is this the case? Ham Pastrami (talk) 19:58, 19 July 2013 (UTC)
The section on "Power Consumption" should be deleted. It is irrelevant and meaningless. Let's look at the three sentences in the current version:
"Power consumption of individual SDRAM chips (or, by extension, DIMMs) varies based on many factors, including speed, type of usage, voltage, etc." Well, yes, that is true of DD3 SDRAM, the topic of this page. Or any memory. Or any integrated circuit. Or any semiconductor, for that matter. I don't see why it needs to be mentioned specifically for DD3 SDRAM.
"Dell's Power Advisor calculates that 4 GB ECC DDR1333 RDIMMs use about 4 W each." That tool tells you nothing about DD3 SDRAM. It estimates power consumption depending on different system configurations, and you can't isolate the power used by memory.
"By contrast, a more modern mainstream desktop-oriented part 8 GB, DDR3/1600 DIMM, is rated at 2.58 W, despite being significantly faster." I don't know why it is "a more modern mainstream desktop-oriented part," but it has no bearing on the discussion. You cannot compare the estimate of the Dell tool with the datasheet of a DIMM. — Preceding unsigned comment added by KZyGfthAK1bp (talk • contribs) 07:47, 13 November 2013 (UTC)
Maximum DIMM/SO-DIMM capacities?
Currently, 8GB seems to be the maximum capacity available on the market, for both 240-pin DIMMs and 204-pin SO-DIMMs. Does anyone know whether higher density modules will be available in the future? Particularly I'm wondering whether a few years from now it will be possible to buy 16GB SO-DIMMs. Considering there aren't currently any 16GB DIMMs even though the DIMMs have twice the number of chips as the SO-DIMMs, I'm afraid the answer may be "no". The answer here says 256GB is the theoretical maximum. But that doesn't really say what the practical maximum will be. --Aij (talk) 05:51, 8 December 2013 (UTC)