Article by Intel says i7-900 only uses max 1066 RAM

Toggle sidebar Toggle sidebar
Mongox

Mongox

Distinguished
Aug 19, 2009
1,124
0
19,460
I ran across this article from a poster in another section of the forums.
http://www.intel.com/support/processors/sb/CS-029913.htm

Clearly says that:
What is the maximum frequency for DDR3 memory when used with Intel® Core™ i7-900 processor series?

These processors support DDR3 memory with a maximum frequency of 1066 MHz. If faster DDR3 memory is used (such as 1333 MHz or 1600 MHz), it will be down-clocked to operate at 1066 MHz.
---------------
What is the maximum frequency for DDR3 memory when used with Intel® Core™ i7-800 processor series and the Intel® Core™ i5-700 processor series?

These processors support DDR3 memory with a maximum frequency of 1333 MHz. If faster DDR3 memory is used (such as 1600 MHz or higher), it will be down-clocked to operate at 1333 MHz.
Click to expand...

So the i5-700 and the i7-800 support 1333 RAM and the i9-900 only supports 1066? This make sense to anyone?

I know there's one statment in the article that doesn't match Intel's policy -
Intel® recommends using memory that adheres to the Jedec memory specification for DDR3 memory which is 1.5 volts, plus or minus 5%. Anything over this voltage can either damage the processor or significantly reduce the processor life span.
Click to expand...
In other places, Intel puts this standard at 1.5V to 1.65V

I'm hoping someone here that knows Intel and their great documentation better can make sense of all this.
 
Sort by date Sort by votes
bilbat

bilbat

Splendid
Sep 21, 2008
4,568
10
23,160
I have been saying this here forever:
http://www.tomshardware.com/forum/265249-30-help-needed-mobo-combo
http://www.tomshardware.com/forum/263505-30-gigabyte-ex58-decreasing-problem
http://www.tomshardware.com/forum/263598-30-ddr3-1600mhz-ex58-ud4p
http://www.tomshardware.com/forum/264746-30-p55m-build-issues
Not only does Intel officially only support 1066 for the tri-channel 1366s:
SLBEQ 3.33 GHz I7-975 D0
SLBCJ 3.20 GHz I7-965 C0
SLBEU 3.20 GHz I7-960 D0
SLBEN 3.06 GHz I7-950 D0
SLBCK 2.93 GHz I7-940 C0
SLBCH 2.66 GHz I7-920 C0
SLBEJ 2.66 GHz I7-920 D0
and 1333 for the dual-channel 1366s:
SLBJG 2.93 GHz I7-870 B1
SLBJJ 2.80 GHz I7-860 N/A
SLBLC 2.66 GHz I5-750 B1
but, there appears to be no real advantage to running the memory faster:
http://www.tomshardware.com/reviews/memory-scaling-i7,2325-5.html
it just:
uses more power
makes your memory hotter and more 'stressed'
makes your CPU hotter and more 'stressed'
makes it harder to achieve long-term stability...

If you are interested in the Intel docs, I've zipped 'em up here:
http://www.mediafire.com/?yzj5ggmyt4g

The only thing that appears to matter to these platforms is latency - and you pay for that, buying the RAM; here's a little tool that I've been working on:
0137b.jpg

It 'looks at' the listed latency at whatever claimed speed, converts it to 'likely latency' at 1066 in the 'adjusted' columns, and then, in the LV (latency value) column, 'weights' the average latency for 'bang for the buck'; notice that the entire 'spread' of average latency is 4.2 counts, and the median value, 8.1 is already reached by the fourth item, the $140 F3-12800CL8T-6GBRM - at that point you are paying $15 more (12%) to get to half the available range, but to get to the other half, the price differential rises to more than double the cost of the base entry (a whopping 108% increase!)
 
Upvote 0 Downvote
Mongox

Mongox

Distinguished
Aug 19, 2009
1,124
0
19,460
But no one wants to buy even 1066 RAM for a high-end Intel computer! It's hard enough convincing them to get 1333's using things like that article by Tom's.

I think I'll stick with recommending the 1333. If I go down to 1066 then I have to worry about whether it's a 800 series or an i5, etc...

And I'll get into all these arguments with other forum posters who swear by 1600 RAM and faster.
 
Upvote 0 Downvote
cyberkuberiah

cyberkuberiah

Distinguished
May 5, 2009
812
0
19,010


wow ... i have bookmarked this and will return to read them myself , and going by your chart , it makes no difference to give more beans for 1600 , or even more ridiculous , the latest ddr3 2000 with fancy coolers . what an informative thread find , the marketing from ddr3 2000 will get me , no , you saved me first !!
 
Upvote 0 Downvote
B

brianz

Distinguished
Jun 5, 2009
52
1
18,545
Thanks bilbat and thanks mongox for the referral over here.

So for an i920/i975 processor on a typical x58 mobo, it sounds like I should just get the lowest latency 1333 MHz DDR3 I can find? The mobo should automatically read the RAM speed/timings, but the processor will "downclock" to 1066 apparently... and the "downclocking" will be automatic I guess? So if the BIOS just reflects the stated RAM speed/timings automatically, and if the processor downclocks to 1066 automatically, I shouldn't have to set anything in BIOS myself?

I just don't like selection of the 1066 DDR3 available... all the good name brands seem to start at 1333 MHz.

Thanks again,
Brian
 
Upvote 0 Downvote
Mongox

Mongox

Distinguished
Aug 19, 2009
1,124
0
19,460
Yes on the auto settings - sometimes!

Often you need to bump the DDR voltage up a little even with 1.5V RAM - almost always with 6 modules and I think you were looking at 12GBs. And sometimes no way to avoid setting the speed also. It depends too much on the motherboard, BIOS and RAM - sometimes even between identical systems. But it's easy to set and won't need adjusting once you get it right.

I think the downclocking to CPU is all in Intel's mind anyway! LOL
 
Upvote 0 Downvote
bilbat

bilbat

Splendid
Sep 21, 2008
4,568
10
23,160
brianz;

So for an i920/i975 processor on a typical x58 mobo, it sounds like I should just get the lowest latency 1333 MHz DDR3 I can find?
Click to expand...

Not quite; the point I'm trying (over and over) to make, is that you want to get the lowest overall latency that you're willing to pay for! Again, the DIMM's latency is a physical entity - it's an amount (actually, various amounts) of time, in nanoseconds or picoseconds, that physical functions (like the amount of time between a row being activated by precharge and deactivated successfully, which is tRAS) involved in reading from or writing to the DIMM take. The RAM doesn't care (nor does the CPU) what frequency it's waiting at (well, there is actually a constraint there - RAM won't likely work at all at a frequency higher than specified), whether it's 24 cycles at 2133, or 12 cycles at 1066, it's still got to wait the whole one and an eighth nanosecond...

I realize that, as a result of the failings of our current educational system, most of the population is 'math-phobic' and innumerate, but this is one of those thing where you really need to look at the math! That's the whole point of the Excel tool above - to do the math for you, so you aren't comparing apples to applesauce! What's faster, 9-9-9-24 @ 2133, or 8-8-8-21 @ 2000? You can't tell without the math!! Here's how it works:

The first thing we're after is relatively how long a memory cycle is at whatever the rated frequency of the RAM is; we don't have to have the number in actual nano/picoseconds, as we're only comparing, so one (second) divided by the nominal frequency will do - thus, for a 2000 DIMM, we can simply use 1/2000, which gives us .0005; next, we want to multiply by the number of cycles that length, so if we're looking at the nine of some 9-9-9-24 2000, it's 9 x that .0005, or .0045... Now, we need the comparative cycle time of our 'target', or comparison frequency, which for the 1366 platform, might as well be the 'rated' 1066; thus, our cycle time at 1066 is 1/1066, or roughly .000938; we divide our 'wait time', .0045, by our slower cycle time, .000938, to get the number of 1066 cycles we need to get the same time: 4.797 - and we have to round that up, to five, as we can't set our BIOS to 'wait' four and eight tenths of a cycle for a column address strobe! The rounding operation points out the only advantage, in physical reality, of running memory faster than Intel's rated speeds; the higher the frequency you clock the RAM at, the 'finer' the 'granularity' of the adjustments get - to accomodate this 'integer rounding'... It should be apparent that, for any random physical time in nanoseconds or picoseconds, you will 'lose' half a cycle, on average, to 'integer rounding, and half a cycle at 2133 is only half as long as half a cycle at 1066! This improvement in 'granularity' hardly compares to what you give up in terms of higher power consumption, higher voltage requirements, more thermal dissipation, and poorer stability issues with increasing memory speed...

 
Upvote 0 Downvote
Mongox

Mongox

Distinguished
Aug 19, 2009
1,124
0
19,460
OK, let's say I want to recommend good quality RAM to someone without teaching them math skills and explaining how to calculate any ratios. Anything wrong with these selections from G.Skill jump out at you? For some reason, they either don't make or Newegg doesn't carry the same modules in both 2 and 3 module kits so you can't get the slightly better modules available in the 6GB kits in the 4GB kits.

And I include both 1333 and 1600 RAM because many insist on 1600. And there's not much out there in 1066 speed and no one will buy them anyway.

DDR3 1333 4GB (2 x 2GB) F3-10666CL7D-4GBRH 7-7-7-21 1.5V $95
http://www.newegg.com/Product/Product.aspx?Item=N82E16820231276

DDR3 1333 6GB (3 x 2GB) F3-10666CL7T-6GBPK 7-7-7-18 1.5V $140
http://www.newegg.com/Product/Product.aspx?Item=N82E16820231230

DDR3 1600 4GB (2 x 2GB) F3-12800CL7D-4GBRH 7-7-7-24 1.6V $115
http://www.newegg.com/Product/Product.aspx?Item=N82E16820231279

DDR3 1600 6GB (3 x 2GB) F3-12800CL6T-6GBTD 6-7-6-18 1.6V $190
http://www.newegg.com/Product/Product.aspx?Item=N82E16820231272
(Tall heatshields)
 
Upvote 0 Downvote
bilbat

bilbat

Splendid
Sep 21, 2008
4,568
10
23,160
OK, let's say I want to recommend good quality RAM to someone without teaching them math skills and explaining how to calculate any ratios. Anything wrong with these selections from G.Skill jump out at you?
Click to expand...
We've reached an impasse here, at a problem that, in its essence, is a philosophical issue; the best way I can think of to address it is by quoting my favorite cognitive philosopher, Daniel Dennett, who, himself, is quoting British psychologist Richard Gregory [emphasis, mine and Dennet’s]:
One of Darwin's fundamental insights is that design is expensive but copying designs is cheap; that is, making an all new design is very difficult, but redesigning old designs is relatively easy. Few of us could reinvent the wheel, but we don't have to, since we acquired the wheel design (and a huge variety of others) from the cultures we grew up in. We may call this sub-sub-subset of Darwinian creatures Gregorian creatures, since Richard Gregory is to my mind the preeminent theorist of the role of information (or more exactly, what Gregory calls Potential Intelligence) in the creation of Smart Moves (or what Gregory calls Kinetic Intelligence). Gregory observes that a pair of scissors, as a well-designed artifact, is not just a result of intelligence but an endower of intelligence (external potential intelligence), in a very straightforward and intuitive sense: when you give someone a pair of scissors, you enhance their potential to arrive more safely and swiftly at Smart Moves.
Anthropologists have long recognized that the advent of tool use accompanied a major increase in intelligence. Chimpanzees in the wild go after termites by thrusting crudely prepared fishing sticks deep into the termites' underground homes and swiftly drawing up a stickful of termites, which they then strip off the stick into their mouths. This fact takes on further significance when we learn that not all chimpanzees have hit upon this trick; in some chimpanzee "cultures," termites are an unexploited food source. This reminds us that tool use is a two-way sign of intelligence; not only does it require intelligence to recognize and maintain a tool (let alone fabricate one), but a tool confers intelligence on those lucky enough to be given one. The better designed the tool (the more information there is embedded in its fabrication), the more potential intelligence it confers on its user. And among the preeminent tools, Gregory reminds us, are what he calls mind tools: words.
Words and other mind tools give a Gregorian creature an inner environment that permits it to construct ever more subtle move generators and move testers. Skinnerian creatures ask themselves, "What do I do next?" and haven't a clue how to answer until they have taken some hard knocks. Popperian creatures make a big advance by asking themselves, "What should I think about next?" before they ask themselves, "What do I do next?"
Click to expand...
I can hand someone a scissors (say, a pruning shear), and give them an idea of how to use it (or even include a manual that they can immediately discard), but I cannot (opinions of members of the plaintiff's bar notwithstanding...) prevent them from lopping off a finger, or stabbing someone with it - the use of an 'intelligence conveying' tool requires some underlying intelligence of the user; and, in a Darwinian world, misuse 'prunes' the genetic tree! A computer is a marvelously complex (and devious!) tool, containing a long heritage of intelligent design; but, its optimal useage requires substantial investment on behalf of the user - the degree to which it is useful is intimately connected to the effort expended on understanding it...

And this says nothing of the reason we can 'build' computers - the whole idea of DIY systems is the fact that you can acquire a highly 'suboptimized' toy, where all the disparate parts 'talk' to each other in a highly integrated fashion; but, it will not 'fall together' of its own accord. This is the fault of a bunch of cognitive disabilities common to humankind in general, and their exploiters. For every class of common cognitive error, there is also a class of people exploiting that generic error to their own profit! (How else to explain the existence of the Hummer H3, or 'Smiling Bob's' Enzyte? :pt1cable: ) The class exploiting computer-related cognitive disabilities includes, of course, the parts manufacturers, as well as a sizeable contingent of the so-called computer press.

I am going to draw here on a previous post regarding computer literacy:

More than half the recent products of this educational system have an attention span that does not extend beyond texting two lines of grammar-free nonsense full of ‘texting contractions’, which means they have NO frustration tolerance. They’ve never been asked to ‘stretch’ their minds, to tackle something beyond their immediate grasp. Everything in their path has been so ‘dumbed-down’ that they assume any endeavor requiring more than an hour is somehow intrinsically faulty. I spent more than a hundred hours doing ‘due diligence’ research for my current workstation build before ever ordering part one! These kids literally can’t imagine such an undertaking, much less believe that it might be necessary. They don’t ‘get’ the underlying math, and they petulantly “want what they want” – the fastest parts advertised, with no comprehension of what they’re paying for, or getting, or what may be required to make it function. The computer media contribute to this problem. Their articles (to suit their confirmed knowledge about their readership) are too short to contain adequate technical discussion to expose potential problems (which their average reader doesn’t want to have to try to read, anyway…); take a ‘cheer-leader’ approach [see below] to make their advertisers happy; and often present a computer ‘build’ in less than two pages, making it appear as if getting a dozen or two disparate parts to function together was merely a matter of assembly. The underlying problem here is that, if the parts themselves were technically and electronically ‘dumbed-down’, to the point where anything would work with anything (which is certainly, in most cases, possible – I hardly ever answer a question about getting 667 RAM to work), there would be NO ‘high-performance’ systems – no real REASON to ‘roll your own’. You certainly can’t BUY the parts cheaper than Dell, when they order in ten or hundred thousand lots. Such a ‘dumbing-down’ of underlying specifications would eliminate the X48 MCH, PC-9600 RAM, and Q9660 CPUs – and what would remain as the point?

I used to subscribe to PC Magazine, as they were fairly technical, and basically honest in their reviewing. This was back during the days when the physical retrieval of a data stream from a hard drive was so slow, that there was much to be gained by doing LZW compression in hardware, on the way in and out. One day I received an issue in which they reviewed a drive controller card which they were never able to make work, although they tried it in several systems, and had access to a level of tech support from the manufacturer ( who, since, I am glad to report, has gone out of business) that no user could ever get – and they still wound up giving it an eight or nine rating, out of a possible ten, BASED ON THE FEATURES LISTED ON THE BOX!!! That very day, I sent them my cancellation notice!

Half the population, BY DEFINITION, is BELOW AVERAGE! People at the twenty-fifth percentile of measured cognitive functionality will consistently estimate themselves to be at approximately the sixty-fifth percentile... “You can lead a horse to water, but you CAN”T teach him calculus!” If you have minimal, or no, math skills; if you never did well in (or avoided completely) science courses – you probably SHOULDN’T be attempting to build a computer… The PC idiots have promoted the idea that everything (and everyone) is equal; that an PET scan (based on particle physics, and incredibly complex computer geometric analysis), Christian Science (based on faith that some possible omnipotent being will take responsibility to heal you), and homeopathic medicine (based on consuming poisonous ‘antidotes’ at dilutions beyond one molecule per MORE than the number of atoms in the WHOLE KNOWN UNIVERSE), are equally valid – are simply a difference of opinion. Well, I’m here to say that your opinion that your pile of parts SHOULD work together to somehow ‘become’ a computer will NOT make it function!
Click to expand...

“You can lead a horse to water, but you CAN”T teach him calculus!” I'm not trying to get anyone to extract a second derivative, here, or integrate an 'under the curve RMS value; I'm pointing out a simple mathematical basis underlying RAM selection that should be comprehensible to any (actually educated) eighth grader, and, as an alternative to doing the math, I've posted an 'intelligence conveying' 'tool' (the spreadsheet itself) that will 'do the math'... I cannot convey some 'magical', intuitive way to comparatively interpret DIMM specifications - because there isn't one!

... still editing!
 
Upvote 0 Downvote
B

brianz

Distinguished
Jun 5, 2009
52
1
18,545


The issue for me and presumably others isn't the math, it's the technical concepts behind it. As a former derivative trader with a masters in finance - which includes advanced calculus and statistics - I'm fine with simple algebra. It's the technical stuff that's challenging. I may as well be telling you why any 8th grader should be able to grasp why you lose money rolling oil contracts with the forward curve in contango.

From what I understand, all else equal, higher RAM speed is better than lower RAM speed. Furthermore, lower latency is better than higher latency. You can get equal performance from 1000 MHz RAM with a latency of 4 as with 2000 MHz RAM and a latency of 8. Beyond that, I'm not too confident I understand much. RAM is very technical and jargon-heavy for non-engineers.
 
Upvote 0 Downvote
Mongox

Mongox

Distinguished
Aug 19, 2009
1,124
0
19,460
You've got it right! Heck, I design a lot of spreadsheets myself, but I think I should be able to eye-ball the specs on the RAM to choose it. Only 11 hours of calculus myself along with an Economics degree!

I'm gonna leave the engineering side to bilbat!
 
Upvote 0 Downvote
bilbat

bilbat

Splendid
Sep 21, 2008
4,568
10
23,160
RAM is very technical and jargon-heavy for non-engineers
Click to expand...
For engineers, too! Trust me!! ;) One of the reasons I do this is because I learn rafts of stuff, nearly every day, that just fascinate the hell out of me - e.g., yesterday while exploring someone's issues with a P55 that refuses to run MemSet to allow us to read his RAM sub-timings, I discovered that, although I had always believed that EPP and XMP settings tables in DIMMs were just different names for the same thing, EPP is an nVidia/AMD spec 'aimed' specifically at the nForce 5, 6, and 7 chipsets, and XMP is the Intel equivalent - and the register arrangement may be different, which would explain a lot of heretofore mysterious problems; in addition, I am always bitching about the dearth of AMD documentation, and, during the same search session, I stumbled across an AMD developer's site, that has, like, a thousand or so documents I gotta read! Already spent two hours just downloading - now, gots to agonize over whether to install one of the dev packages, as it requires a driver with a 'ring zero' component - and I avoid unnecessary ring zeros like the plague!

Far and away, the best, most explanatory document I've ever seen, that 'gets it all together' in one place:
http://www.mediafire.com/?dnhrjhrxgyw
...but it's a hundred and fourteen pages! You don't really need to read all of it to extract the gist - the things to take away are: the first chapter explains, pretty much, what the latencies are, and the physical reason for them - the details are not important; the second chapter is the 'meat' - it explains DDR2 and DDR3 physical organization, JEDEC requirements, and actual access methods; chapter three covers processor cache operation, organization, and 'snooping', and is more relevant to CPU selection than memory; and the rest is aimed at helping programmers optimize memory access, and so is a waste of time for most people, though the last chapter is kind of interesting in a SciFi sort of way - talks about possible future technologies...

It's the technical stuff that's challenging. I may as well be telling you why any 8th grader should be able to grasp why you lose money rolling oil contracts with the forward curve in contango.
Click to expand...
:lol: If I haven't made these issues simple enough for an eighth grader, that's my fault :cry: (I am reminded of a David Deutsch lecture on the quantum mutiverse, where he says "this means there are an infinity of possible, slightly different 'me's, delivering this same' but slightly different, lecture, to an infinity of slightly different students, in the quantum multiverse, and, hopefully, some of them are explaining this better...!!") To rehash the concepts:

1 - the latencies, though expressed in counts by our SPD 'reader' software, are actual, physically inviolate time constants, imposed as a sort of 'speed limit' by the elctronics involved...

&

2 - the only way to compare memory (and especially 'bang for the buck') is to use simple (nothing more complex than long division) math to 'normalize' the RAM to some standard value, as we are not 'evolutionarily' equipped to know what has the shortest (fastest) physical latency period between, say, 9-9-9-24 @ 2133, and 8-8-8-21 @ 2000 directly...

You can get equal performance from 1000 MHz RAM with a latency of 4 as with 2000 MHz RAM and a latency of 8.
Click to expand...

That is exactly the concept, but, unfortunately, will only work exactly if you happen to be comparing 1066 & 2133 - hence, the little bit of 'dividing'...

BTW - brianz - we gotta talk; I'm trying my damndest to comprehend the more arcane complexities of the derivative markets, with respect to computer modeling, and could use some pointers to similar 'basic' expositions; obviously, a large number of investment bankers needed the same about eighteen months ago! :heink:
 
Upvote 0 Downvote
B

brianz

Distinguished
Jun 5, 2009
52
1
18,545
Thanks for the link to the Memory PDF - great intro. I am definitely going to get a lot out of reading those first few sections.

So if budget isn't an issue, what's the best memory you'd recommend for a 12 GB set-up on an X58 mobo with an i900 series CPU? I would be looking for a 1333 MHz low latency configuration. The G.Skill 1333 MHz 7-7-7-18 seems to make sense. No reason to get the 1600 MHz for i900 according to the Intel docs, since I don't plan to overclock, right?

Would you recommend G.Skill over Corsair for any reason other than value? I am willing to pay up for quality and low latency but obviously I don't want to pay for features that don't matter, such as 2000 MHz speed.

As for derivatives, they vary significantly in structure and thus price modeling. Some are options, in which case I would recommend that you read about Black-Scholes. That's the classic pricing model off of which all options and option-based derivatives are priced. It makes lots of absurd assumptions like all economic models though, most notably the normal distribution of price changes... as we now know, returns have a "fat tailed" distribution. For credit derivatives, the most important are credit default swaps (CDS). I'd recommend learning about CDS pricing - it's all simply a matter of imputing the market's probability of default based on the credit spread, and then pricing insurance off of that. CDOs and CLOs are pretty stupid instruments that shouldn't have been invented and hopefully are going away so I wouldn't waste much time on them. They are still impacting bank balance sheets though.
 
Upvote 0 Downvote
You must log in or register to reply here.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom