FAQ  •  Register  •  Login

RAM Usage

Moderator: MiRai

<<

mytu

Posts: 2

Joined: Mon Oct 13, 2014 12:46 am

Post Mon Oct 13, 2014 12:51 am

RAM Usage

Hi there,

I just sold my Z87 PC to upgrade to x99..so far I've only bought

Asus Rampage V Extreme,
Corsair 900D,
Corsair 1500i

I want to purchase RAM and CPU next but I don't which RAM to choose,

I'm going for the 5960X but as RAM goes, no idea..I had previously 32 GB and I multibox 25 accounts which is enough but with the new DDR4, is there any different? is wow.exe going to use less RAM in DDR4? and if it will, does the Mhz makes any difference?

example..if I purchase 16GB 3200Mhz DDR4 will perform better than 32GB at 2400Mhz? I mean wow.exe will use less resources on the 3200Mhz?

examples : http://www.overclockers.co.uk/showprodu ... ubcat=2557 VS http://www.overclockers.co.uk/showprodu ... ubcat=2557

They are very expensive and I want to know which one is worth purchasing..

Thanks,
Mihai
<<

bob

User avatar

League of Extraordinary Multiboxers

Posts: 4587

Joined: Sat Feb 15, 2014 11:14 am

Location: In the dining room, with the lead pipe.

Post Mon Oct 13, 2014 4:32 am

Re: RAM Usage

The speed of the RAM (the DDR, MHZ values) will make no difference at all to the amount that the game uses. The two are not connected. If you needed 32GB previously (which you probably do, a rule of thumb is 1GB per game instance - but it is only a "rule of thumb"), you'll still need 32GB. If you could have gotten away with 16GB previously, then you'll still be able to get away with it.

As for how the speeds inter-relate, that is harder.

DDR is the measure of the number of instructions that can be posted per clock cycle. SDR (pre DDR) could read or write once per clock cycle (this being a digital clock which goes from a 0 state to a 1 state and then back to a 0 state). DDR changed this so that a Read or Write could occur on the rise of the clock or the fall of the clock. DDR2 got fancier and doubled the base memory clock rate, so could perform could perform 4 actions (2^2) on the rise/fall, DDR3 quadruples the base clock so can perform 8 actions (2^3) on the rise/fall etc.... Hopefully you get the idea.

MHz is the speed at which the I/O clock is running at, but this number is pumped up again by the marketing people because they can, and that DDR has a data rate which is twice the I/O (remember that DDR passes instructions on the rise and fall of the clock). In reality a DDR4-3200 is only running at 200Mhz base clock, but operates at a 1600MHz I/O clock, and has a data rate of 3200 MT/s (million transfers). Most marketers and sales shop use MHz when it refers to MT/s. What really makes a difference is working out the potential data transfer bandwidth, along with the latency and how that affects it.

DDR4 data transfer rates (MB/s) can be determined by the calculation of (memory clock rate) × 8 (for bus clock multiplier) × 2 (for data rate) × 64 (number of bits transferred) / 8 (number of bits/byte). So a DDR4-3200 module is: 200 x 8 x 2 x 64 / 8 = 25,600 MB/s. So far this is good. More data. DDR3 was topping out around 17,000MB/s. (edit: normal DDR3 was, found some that got to 23,500MB/s at 12-14-14-36, and going by the numbers is effectively faster than DDR4 at this time, although it leaves the potential door open wide for DDR4 to kick ass when it is running at a base clock of 366MHz too).


What does this all mean. Very little unless you include the latency at which memory accesses can occur. Latency is presented in sets of 4 numbers in the form of X-X-X-X. Like 16-18-18-36 for DDR4
As the MT/s (and MHz) has increased so has the memory access latency, which is an effective go slow for all the speed increase in the clock. The Latency is measuring the number of clock cycles that are required before a given instruction will occur. These are generally worst case numbers. The best example is the first number in the set normally refers to CAS Latency or CL. This is the number of cycles that are required before the requested data is available ON THE PINS (i.e. the request has hit the memory chip, and the chip has gone to it's storage banks, retrieved it and it is ready to be read from the pins of the memory module). This can be measured in nanoseconds by t = 1 / f, where f = frequency (or I/O clock). So out DDR4-3200 with 16-18-18-36 timings is waiting 1/1,600,000,000 (remember that MHz is Million Hertz) or 0.000000000625ns * 16, which is 10ns. So will the 3200 set outperform the 2400 set. Yes. Mainly because the latency is almost the same. Normally a lower MT/s (or MHz in marketing) nets a lower latency which ends up negating the clock cycle difference (check out the range of DDR3 modules and you'll see that as you go to lower "MHz" the latency also lowers).


Now there is probably a really good formula for working out how to apply the latency values to the data rates, but I don't know them, and it really depends on a whole lot of other stuff. Generally speaking your CPU doesn't normally ask for 25GB of data at a time, so the actual real world experience is way different to this cock and bull.


Does this help you at all? Maybe.

In the end, you want a good compromise of GB of RAM, MHZ, and latency.
In order you want: many GB (although not much point paying for GB you don't need), low latency, many MHZ and all for a reasonable price. You don't normally get a choice on the DDRx side of the fence as this will be limited by the slots on your motherboard.

What are those numbers? Well without your budget, hard to say, but for the most part I don't think you'd notice any difference between any of the sets, regardless of MHz/latency (like really, we are talking nanoseconds here). What you should be looking for in preference is a set of RAM that are on your motherboard compatibility list. If you can't find any, go for a known brand, like Corsair/Kingston/Mushkin/GSkill (some people have their preferences, I'm not claiming an exhaustive list here).
Once you've picked something, I suggest a 2400, 2666 or 2800 set (this seems to be the standard range), with the lowest latency you can afford. Sometimes the high "MHz" sets are touchy about working at full tilt.

BTW, the Crucial Ballistix is missing the final value for the latency (it is always 4 numbers) so you'd need to look it up.


edit: fixed up some misinformation
Last edited by bob on Thu Oct 16, 2014 4:22 pm, edited 5 times in total.
<<

mytu

Posts: 2

Joined: Mon Oct 13, 2014 12:46 am

Post Mon Oct 13, 2014 6:32 am

Re: RAM Usage

Thank you so much for the effort you put in the post.
Was more information than I was expected but I'm very grateful.
I had a look and it seems that the higher the MHz, the higher the latency.

The 2800+ are over £500 which I can't afford atm so the only options are 2666 or 2400..they all seems to have 15-15-15-35/36 while these http://www.overclockers.co.uk/showprodu ... =MY-022-AD are 13-13-13-31 but 2133MHz.

What do you think of those? They seem to be quite decent but never owned anything aDATA..

or should I go for a 2400 - 2666 MHz but 15-15-15-35 latency?
<<

bob

User avatar

League of Extraordinary Multiboxers

Posts: 4587

Joined: Sat Feb 15, 2014 11:14 am

Location: In the dining room, with the lead pipe.

Post Mon Oct 13, 2014 7:57 am

Re: RAM Usage

As MHZ goes to a higher number, the latency will too, and you will generally see no overall difference in the total bandwidth i.e. GB of data that can move across the memory.

So whether you get lower MHZ with lower latency or higher MHZ with higher latency is up to you. I doubt you would be able to tell the difference (Ok this is not strictly true in certain circumstances - lower latency in video editing is better, but then that will generally only matter to you if that is your day job and not a hobby).

Comes down to, buy what fits your pocket.
Also most X99 motherboards have 8 ram slots so you could get 2x16GB kits if that is better pricewise.

Also as for A-DATA, I have used their RAM in the past when doing system building. It is generally OK, especially if you can work out which chips are underlying it (Samsung or Hynix usually). Most RAM failures occur within the first 2 years, so as long as you can get a warranty to cover it, no problem.


Also, another rule of thumb is that one jump in the MHZ rating is about the same as 1us in latency, so a 2133 kit at 13-13-13-35 will be about the same as a 2400 kit at 14-14-14-36 and about the same as a 2666 kit at 15-15-15-37 (I'm making these numbers up a little - the first one in the series is considered the most important one and is normally the one you need to look at, although the others do have an impact). Kits with those kinds of numbers will end up pushing almost the same bandwidth of data (Gb/s) across them.

Return to Multiboxing Hardware

Who is online

Users browsing this forum: No registered users and 8 guests