Intel Core i5 Bus Speed/RAM Speed Confusion

frozeninferno

Distinguished
Sep 22, 2009
14
0
18,510
I'm about to build a computer for the first time, and I decided to go with the i5. On this website, it says I should get memory at around 3/4 of my bus speed. Is this referring to real speed or effective speed?

Also, this is kind of confusing since RAM speed is measured in MHz while the bus speed for the i5 is measured in GT/s, but I read on Wikipedia that 1 GT/s is basically equal to 1 GHz (is this true?).

And lastly, when you see something like "4GB (2 x 2GB) 240-Pin DDR3 SDRAM DDR3 1333 (PC3 10600) Dual Channel Kit Desktop Memory", what does 1333 represent?

I am so lost, please help.

EDIT: How would these rank in terms of performance:
1066 MHz with 7 cas latency
1333 MHz with 9 cas latency
1600 MHz with 8 cas latency
 
I believe the 3/4 reference is to effective speed - with the multiplier. So if your bus has an effective speed of 2000, they're saying look for ~1500 speed RAM. Seems like fair advice, but just a rule of thumb. Consider that you're also looking at "DDR" RAM - Double Data Rate. The true speed of 1600 RAM is 800, then doubled.

Wiki does have some good info - try this page for explanation of why DDR3-1333 = PC3-10600
http://en.wikipedia.org/wiki/DDR2_SDRAM
It's actually about DDR2, but gives the info well.

On this board and elsewhere, we tend to always say MHz even though sometimes not totally accurate. Wiki's explanation seems correct.

The basic rule of CL is that the lower the number, the faster the RAM and the faster the throughput - in a given class of modules. So that:
1066 with CL=7 is always faster than 1066 @ 8
1600 @ 8 is always faster than 1600 @ 9

Once you get into comparing different classes (speeds) of modules, you get different opinions. Because of the package of data that's used by RAM, faster transfer rates (MHz) don't always mean faster throughput. The CL numbers represent delays in the package's arrival.

It's like choosing two roads, both the same length - one has a high speed limit but 8 stop signs. The other has a lower speed limit but less stop signs. Sometimes one road is faster, sometimes the other. And people will argue about which is the best route.

You can figure that 1333 @ 9 is going to be slower than 1066 @ 7. One step of speed advantage, two steps of delay. The delay is more important. Argument ensues when comparing 1333 @ 8 with 1066 @ 7 however.

Basically, pick a speed (MHz) of RAM that suits you, then look for the lowest CL available for a reasonable price.

Soooo... don't forget voltage! Intel's and other bodies say the standard is NO DDR3 above 1.65V. The manufacturer's aren't always following that. Don't consider any DDR3 RAM that wants more than 1.65V - look for straight 1.5 if you can, let's you play with voltage later if you want.
 
Thanks tonnes for the reply, really appreciate it.

But you didn't really touch on my dilemma about the FSB/memory speed. Is there a certain amount of speed I should be looking for when taking into account a bus speed of 2.5 GT/s? If Wikipedia was correct about GT/s = GHz, then should I be looking for... 1875 MHz RAM?

Also, I found a 1600 @ 8 CL RAM kit with good reviews and a decent price (I think), but the voltage is 1.9. How would one go about figuring out what voltages his rig would be able to support?

My build plan if it helps:
Intel i5 w/ Biostar TP55
Radeon HD 4850 512MB
Rosewill 550w PSU
WD Caviar Black 500GB @ 7200 RPM /w 32MB cache
 
Some points:

1) Big difference in price.
2) Don't get a lot more than you need (probably 1333 or 1600)
3) timings are important. Lower is better. the "same" timings on a 1600 are actually faster than on a 1333 (7-7-7-24 for example.)
4) Quality

This is in an X58 system but here's an example of the difference TIMINGS make:
http://hothardware.com/Articles/Triple-Channel-DDR3-with-the-Core-i7/?page=10

The OCZ Platinum (7-7-7-24) at about $82 for 4GB seems a pretty excellent deal:
http://www.ncixus.com/products/30371/OCZ3P16004GK/OCZ%20Technology/

 


Ahhh, a Brit! As I said about the ratio, might be a good rule of thumb. Post a link to the article and I'll read it over. But you can see how quickly it fails, since if you're limited (by reasonability) to a RAM speed of no more than 1600, then should you chunk the FSB speed? The answer is no. Get everything as fast as possible, within reasonable budget contraints. I'd say 1600 RAM is the top end of reasonable.

I'm sure the articles mentioned just above by photonboy talk about how silly things can get when you push RAM speed above all else, there's just not enough benefit. Here's an article from this site.
http://www.tomshardware.com/reviews/memory-scaling-i7,2325.html
Be sure to read the last page, the conclusion, if nothing else.

Ummm... I'll say again, I don't recommend ANY DDR3 RAM that requires more than 1.65V. Regardless of price or rebates or any other factors. For one thing, it means it's an old design - I doubt we'll see much 1.9V DDR3 in a year. Too many returns by users are compounding the costs to manufacturers. Intel's forbidden greater than 1.65V for the i7, I think for the i5 too but not sure. Your motherboard defaults to 1.6V for DRAM, anything asking for 1.5V is fine with that. The MB will go up to 2.3V! but again, don't buy it!

Pick a speed, I'd go with 1600... then look for the lowest CL at that speed you can find. [strike]CL=7 is the best, but pricey. The CL=8 is the best deal, good trade of price/performance.[/strike]
------------
Actually, there's lots of CL=7 RAM for same price as CL=8, so I'd go for the 7s.
http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&N=2010170147%201052129233%201052429371%201052315794&name=4GB%20%282%20x%202GB%29

Remember, lower CL will make your system faster than higher MHz... or GT/s!
 
"what does 1333 represent?"

http://en.wikipedia.org/wiki/DDR3_SDRAM

The short answer:
It means how fast it will operate. It's similar to the Average Read/Write rates on a hard drive. There are also delays called latencies (similar to hard drive seek times). You are also limited by the speed of the MEMORY CONTROLLER which may be on the board or the CPU (I think all modern ones are in the CPU?).

The very long answer (I took a few liberties with the techno-jargon for simplicity):
Everything in the computer is basically about crunching math and moving data. A certain colour on one picture of your screen is represented by a number in memory corresponding to that wavelength. Windows and programs load themselves into RAM from the slower hard drive. When a game is running the CPU is following the instructions copied into the RAM.


Let's say the CPU calculates "5x7" and it's a 2GHz CPU (2,000,000 cyles per second.) On the first cycle the number "5" (in machine code "0000000000000101" I think) is sitting on the input copper wires. On the next cycle it is "moved" into the CPU. The MULTIPLY command (again, represented by the voltages on the input wires) would next appear at the input and move into the CPU a cyle at a time. The number seven comes next. After several cycles the answer "35" is placed outside the CPU on the bus (again, electrical wires). *Note that it is the combination of frequency and the number of clock cycles before the answer is spit out that determine the CPU's performance for a given task. Architecture and clock speed. It's similar with RAM.

RAM:
Your RAM is a whole bunch of transistors. They are basically "buckets" which store the data as electrons. They also "leak" meaning they need to be refreshed. One of the values for RAM refers to "RAS" or "Row Array Strobe" referring to the time it takes to "fill up" the leaking buckets with new electrons.

RAM, like your CPU operates by reference to a CLOCK. Everything happens synchronized to these clocks. Requests made to the RAM consist of placing the ADDRESS for the data on the bus. The RAM has an address decoder to select that specific address. It is then found and placed on the bus to be transferred elsewhere. A picture might be sitting in RAM and your CPU starts requesting it, one byte at a time (but very quicly). So each byte might be requested, the address decoded, the data placed on the bus and eventually placed into the VIDEO RAM (which is basically identical in architecture) after which the Video card reads it in a similar fashion and sends it to the monitor to be drawn on the screen.

In simple terms, the FREQUENCY is how fast your RAM can operate as dictated by the clock. However there are delays (latencies) which will slow down how long it takes the data to be accessed.

Important note:
If you buy RAM that is faster than you can currently use (say your memory controller is 1066MHz) you will be limited to 1066MHz. The difference between the 1333MHz and 1600MHz memory will be due to any latencies because in this case they are both operating at 1066MHz.

With the cost or RAM it's good to buy slightly FASTER than you need. If I had a 1066MHz Memory Controller I would not buy 1066MHz for example. I'd get 1333MHz or even 1600MHz not only for overclocking (the memory controller not the RAM) but maybe reliability too. I don't want to get too much into reliability but it is a good idea to do a little research and read some reviews. (always run Memtest for at least 12 hours after installing your RAM)


RAM and Video games:

There's ALWAYS a weakest link depending on the scenario. Loading Windows or loading a game is limited by the hard drive usually. Compressing video or data is often limited by the CPU. Games are often limited by the CPU or the Graphics card (with the hard drive too for loading times). RAM, too in certain scenarios can be the weakest link. Let's say you have a game on your hard drive. It's loaded into your RAM. Your CPU starts reading and executing the instructions that are in your RAM. If the CPU is executing the instructions FASTER than it can get them from your RAM then the game will run slower. Let's say there are NO other weak links and we're running a game. Ideally, we'd like to run a game at Maximum Quality @ 60Hz with no components running at 100% at any time. Let's say WITHOUT a RAM limitation our CPU is the weakest link and at least one core is at 100% which causes the average Frame Rate to be 40FPS. But, if our RAM was too slow the game might run at, say, 35FPS and our CPU is only at 90%. This is why ATI talks so much about "Fusion" or balance. Although we see a big difference in the generic "theoretical" RAM tests it's a lot harder in practice for the RAM to be the slow part. It does happen though and the FASTER the CPU, the more important faster RAM becomes as they talk to each other. This is why you need to think carefully about whether you will overclock. From my understanding, some of the annoying, sudden slowdowns in some games are a result of the RAM being too slow. I don't know how often this happens. You might for example, be running the game at a HIGH framerate of over 40FPS, but you turn and the screen is jerky! Huh, you say. That COULD be slow RAM (I have a feeling my I'd see a difference if my X2-4800+ CPU could be paired with 800MHz DDR2 instead of my 400MHz DDR).


 
As far as RAM speed goes, it isn't that hard. There are two main components to RAM speed - bandwidth and latency. Very roughly, bandwidth should scale perfectly with frequency, and latency should scale as CAS divided by frequency.

Therefore:
1600MHz RAM (regardless of CAS) should have roughly 1.5x the bandwidth of 1066 MHz RAM (1600/1066=1.5).
1600 CAS 8 should have the same latency as 1066 CAS 5.33 (8/1600=5.33/1066). This means that 1066 CAS 6 will always be slower than 1600 CAS 8, despite the lower CAS number. 1333 needs CAS 6.665 or better to match 1600 CAS 8.

This is also why CAS numbers are higher for higher frequencies - CAS values are in terms of clock cycles, not absolutes. The true latency can be found by dividing the CAS latency by the clock rate. For 1600MHz RAM, the clock rate is 800MHz (since it is dual data rate). If the CAS latency is 8 clock cycles, the true latency is 8/(800MHz) = 10 nanoseconds. This is identical to DDR2-800 (400Mhz true clock) CAS 4 and DDR-400 CAS 2. DDR3-2000 CAS 10 is also the same true latency. The bandwidth will be vastly different however, so don't think that DDR-400 CAS 2 and DDR3-2000 CAS 10 are identical :)
 

blink.gif
?


Gotcha. So would I be correct to assume that in my situation, the memory speed will bottleneck much earlier than the CPU?


Woops, sorry, completely forgot about your suggestion. Got lost in all my research and whatnot before I had this strange inkling feeling not to go with RAM above 1.5V, but didn't know why or where I read it. Funny it turned out to be in this actual thread...
animesweat.gif



Is this really true? I heard it mainly depends on what types of applications you're running. Or are you referring to the same "true" latency brought up by cjl?


😱 My eyes are opened. Beautiful explanation, much appreciated!

@photonboy: Thanks heaps for the info! It was a very good read, but I was really just referring to whether it represented real or effective clock speed.
 
LOL, an American wouldn't say "tonnes" - unless he's a poor speller.

Ah, did you read the last page of this review by Tom's Hardware:
http://www.tomshardware.com/reviews/memory-scaling-i7,2325-11.html
Knowing all of these results, it is obvious that highest speed DDR3 memory only makes sense for serious enthusiasts, or for those with unlimited budgets. Everyone else should focus on mainstream clock speeds of DDR3-1066 or DDR3-1333, and go for a trustworthy brand and the quickest timings their budget allows.

Always aim at lower CLs.
Consider my 3 choices so far on DDR2 RAM for my new system - yes, we all make mistakes. First I got cheap 1066 that needed 2.2V. No good, my MB maxed out at 2.1V - the 1.8V is standard on DDR2 like 1.5V is standard on DDR3. So that RAM gets replaced by the cheapest 1.8V 1066 RAM. Oops, I'm not happy with a CL=7 when I could pay a little more and get CL=5. So, I finally get RAM I'm satisfied with, 1066 w/CL=5.

Now... this 1066 RAM also has a CL=4 when used as 800 RAM. So that's what I'm doing. I wanted the 1066 but I knew that the lower CL was more important to throughput - in ANY application - than the higher speed. So I'm satisfied - I can run my RAM at 1066, but I won't.

I like the 1333 @ 7 choice. And G.Skill is a solid manufacturer. But, the 1600 @ 7 is nice too, only $7 more! And... it might even run as 1333 @ 6 - but have to ask em over at the G.Skill forum about that.

Any of those you'll be happy with.

handsonram.jpg

I think I'll start adding this pic to all my posts! I love the giant RAM! Or is it tiny hands?
 
Hahaha, I'm Canadian actually, great observation though, threw me off for a sec.
laugh.gif


So basically, definitely the 1333 @ 7 over the 1600 @ 8, but the 1600 @ 7 if I'm willing to spend the extra cash.

The 1333 is at 1.5v though, while the 1600 is at 1.65v... Does this matter? Should I take this into account over the higher clock speed?


"RAM Installation for Toddlers", gotta start em' young!
 
Ah, I knew it! Some of the manuals take the "English" version too seriously and mention "coloured" slots too!

1.65V is within the standard. Might find you want to cheat a little and run it at 1.7V if you decide to overclock it but that would be OK.

Now, you have to make a decision! Be sure to come back and post how things work out when you get the system up and running.
 

Not true.

1066 CL5 has a latency of 9.38 nanoseconds. 800 CL4 has a latency of 10 nanoseconds. You are running with a higher latency than stock by running at the 800CL4 setting.

Alternatively, look at it this way: if low CL is all that matters, why would DDR3-1600 perform better than DDR1?

FrozenInferno: Between those 3, I'd go with the 1600C7 - it's only $5 more, and has the best specs of the bunch.

Here's some proof of my above statements:

By this (synthetic) benchmark, you can see that memory bandwidth is almost entirely determined by the RAM speed - there's a slight difference between the different latencies (because my above statement is an oversimplification), but it's not a large difference. You can see that in bandwidth, memory speed is king:

evreadsa.jpg


In writes, the scaling is perfect - regardless of CAS, the bandwidth is identical for a given speed:

evwritesa.jpg


Here are latency results. Note that 1600 C8 matches 1333 C6, and 1066 C5 can only manage to match 1333 C8. This fairly closely matches the calculations above:

evlatencya.jpg


In some tests (especially compression benchmarks), the bandwidth is almost all that matters:

winrara.jpg

 
See, this is always good for an argument!

Fact is, none of makes much difference by any argument or calculation.

From Tom's Hardware - Core i7 Memory Scaling: From DDR3-800 to DDR3-1600
http://www.tomshardware.com/reviews/memory-scaling-i7,2325.html

Upgrading from DDR3-800 to DDR3-1600 (with low latencies) will result in a 72.3% increase in gross memory bandwidth. Unfortunately, most applications don’t take advantage of it.

012_pcmark_mem.png


The memory industry, which typically focuses on offering faster (and more expensive) memory products, must find these results rather annoying—there is simply very little benefit to fast DDR3 memory on a fast processor such as the Intel Core i7-975. More than ever, premium RAM can be compared to a high-end sports car: it may provide better performance, but the benefit in everyday life is often very limited.

DDR3 memory kits capable of running beyond DDR3-2000 speeds at increased voltage are the best you can get. However, we limited the testing to DDR3-1600 speed, and stayed at a voltage level of 1.65 V for this article, which represents a reasonable maximum for most users (hitting 2133 in our i7-975 review required a screaming 1.75 V QPI). Our testing included all other selectable RAM speeds, all the way down to DDR3-800, trying both tight and relaxed latencies for each of the RAM clock speed settings.

High End Memory Is Not Worth The Money

The results are obvious: going from one memory speed to the next, e.g. from DDR3-1066 to 1333, does not provide major benefits. Even the replacement of slow DDR3-800 RAM by DDR3-1600 memory will mostly yield disappointing results. While the performance advantage is measurable, it is never noticeable.

Exceptions, however, do exist. Compressing files with WinRAR is significantly quicker on fast, low latency DDR3-1600 RAM. Some applications, such as games, can at least take minor advantage of the upgraded memory horsepower.

What If?

In this light, we decided to add a few more benchmark results at overclocked processor speeds. We decided to accelerate the CPU by one clock speed increment, which reflects exactly what happens if you decide to purchase a faster processor instead of high performance memory. As you will see, a higher CPU clock speed typically provides better performance in most of the benchmarks—but not in all of them.

Memory Recommendation

Knowing all of these results, it is obvious that highest speed DDR3 memory only makes sense for serious enthusiasts, or for those with unlimited budgets. Everyone else should focus on mainstream clock speeds of DDR3-1066 or DDR3-1333, and go for a trustworthy brand and the quickest timings their budget allows.
 
Agreed :)

Very few applications actually take advantage, so the difference is likely to be a few percent at best in normal applications. I have 1600MHz Dominator because I use a lot of extremely memory heavy apps that do scale with memory speed (Matlab especially likes memory bandwidth, and I'm not sure about solidworks, though Solidworks will quite happily chew through memory 6 or 8 gigs at a time). These are hardly normal usage patterns though, so it really is a very small difference for most users. Still, given the choices, I'd spend the 5 extra bucks and get the 1600C7. It isn't worth much more, but $5 really isn't much of a price premium.
 
I'm playing around right now with the settings back to 1066 @5.

But ya know, it just "feels" faster at 800 @4 - little things like how long the POST takes to find the drives. Doesn't make sense I know.

Regardless, either is stable and doesn't affect any of the tests, like Prime95, I run now and then to test changes to my CPU OC.

goldenfinger.jpg

I'm glad my RAM isn't giving me the Golden Finger
 
Set everything up like last week or the week before, forgot to come back, heh.
animesweat.gif


Regardless, everything went pretty smoothly and the computer runs GREAT, including the RAM, so for my first build ever, not bad at all. Thanks "tonnes" for the help guys, you're the best!

dance.gif
 
RAM is faster in TWO ways:

1) the raw frequency (ie. 1600MHz vs 1333MHz)
2) the timings (i.e. 7-7-7-24 vs 9-9-9-24)

It's possible to have slightly higher frequency RAM be slower overall if the timings are sufficiently faster.

In general, any RAM is likely sufficient unless you're overclocking your CPU and then you need to investigate further.

In general, 1066MHz is fast enough for most people, and very few need more than 1600MHz. 1600MHz currently is quite cheap anyway.

I personally purchased RAM of these specifications:
DDR3 1600MHz @ 7-7-7-24

I've tested my 1155 system with the CPU (i7-860) overclocked to 3.8GHz and the RAM was never bottlenecked at any point.