• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

"future proof" RAM upgrade

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

alternety

New Member
Joined
Mar 14, 2010
I am currently running G-Skill Ripjaws 1600 on a Gigabyte P55a board. I want to increase my RAM size. I have read the very interesting Anand-tech article on Sandy Bridge, about RAM speed vs actual application performance. My current thinking is that I will upgrade my motherboard in 1 to 2 years. Mostly for new interface standards. NOT Win 8 (what are they thinking, and stopping what drugs will make them better?). I am not in the group with a pathological urge to wring every single cycle out of a CPU. I do mild and stable overclocks for 24X7 operation.

I have never tried buying parts with a specific eye toward use with the next generation. This time it seems to me that it may be appropriate for the RAM. There appears a possibility of DDR4 starting to be implemented in the 2 year range. I have no idea if this will be an abrupt (no new boards use DDR3) or a gradual transition (both available for several years).:facepalm:

Generally if you want to expand RAM, you should replace it all with matched sets.

I prefer G.Skill RAM. I want to go to 16GB at this time. I have posted elsewhere and received this recommendation; at least 2400 speed. Get 2X8GB sticks. The rational for 2X8 in the response is that in the near future, enhanced XMP will allow the use of mismatched RAM in the machine. This leaves me the option of easily going to 32GB in the future.

After research (particularly the Anand-Tech article), it would not appear to me that I should pay a premium for very fast RAM. The Ivy Bridge CPUs seem to be a poor choice if you really want to overclock because of an odd choice by Intel on heat coupling within the package. That may change (it is like introducing Vista or Win 8). The same generation of support chip sets seem to have advantages. In two years it may make sense to get an Ivy Bridge chip set and a Sandy Bridge processor or a successor.

So the question. I assume that lower timing/latency for the same frequency is better and worth something. But what speed should I be looking at for real-world applications that will still be suitable for hardware two years from now. All applications, nothing specific. I might game, compress video, fold molecules, whatever. If a guru or two would kindly rub their crystal balls and their HP 11C and give me the basis for selection (or a preferred number) of the most cost effective speed I would really appreciate it.
 
1600 mhz, 8 gig of ram is fine for 99% of general usage these days. It will run you about $50.

If new ram comes out buy new ram then. Ram is so cheap now that any new tech is not worth considering IMHO.

Cross matching ram will always be a bad idea.
 
Thanks for your opinion, but that is kinda not what I asked.

What RAM to buy to work in a MB 2 years from now.
 
But what speed should I be looking at for real-world applications that will still be suitable for hardware two years from now.

4GB-8GB RAM at 1600 Mhz and any latency should be optimal for anything you might be doing. Future-proofing is kind of silly any way. In this industry, technology progresses so fast, and prices change so rapidly, "future-proofing" is hardly possible or even a good idea. Buy for what you need NOW. Worry about other upgrades when you need them. And I predict that any DDR3 RAM you buy today should work in a Mobo 2 years from now, and if it doesn't, there's not much you can do about it. If they implement DDR4, then you'll have to buy new RAM no matter what you buy.
 
4GB-8GB RAM at 1600 Mhz and any latency should be optimal for anything you might be doing. Future-proofing is kind of silly any way. In this industry, technology progresses so fast, and prices change so rapidly, "future-proofing" is hardly possible or even a good idea. Buy for what you need NOW. Worry about other upgrades when you need them. And I predict that any DDR3 RAM you buy today should work in a Mobo 2 years from now, and if it doesn't, there's not much you can do about it. If they implement DDR4, then you'll have to buy new RAM no matter what you buy.
What he said :)

You're probably fine were you are RAM wise, if something better comes out get it then.

I did get some pretty nice OCZ DDR2-1200 ram my wife uses now pretty good time ago that my wife uses, but you don't see it really making a major difference at the moment from what I've seen.
 
Haswell is going to be launching from March-June next year, and bring with it integrated DDR4 support. I'm going to assume the chipsets will be backwards compatible with DDR3. Now to find that article...

EDIT: Looks like I remembered incorrectly, and it will be Haswell server CPUs supporting DDR4, expected to launch in 2014. Couldn't find the exact article I read the info from, but here's another one: http://dtptech.blogspot.com/2012/06/processors-intel-haswell-ep-will.html
 
It looks more like Intel don't want to make anything on DDR4 so fast as they are still making money on DDR3 but IC producers see good profit in DDR4 and they are pushing Intel to add DDR4 support to next gen chips.
For example Samsung and Hynix already have working DDR4 but still in 30nm version while official supposed to be 20nm.
We will see DDR4 probably in next year in servers but since IMC is integrated in cpu and Intel never made it backward compatible then I doubt that we will see the same generation on DDR3 and DDR4 together ( at least for desktops ).
Not that it will be any difference in performance. More like difference between DDR2 and 1st DDR3 where DDR2 were actually faster at the beginning ;) Standard DDR4 will start from 2133 clock on looser timings than we have now DDR3 so we can expect something good in maybe 2 years.
8GB DDR3-1600 will be still good enough for almost everything in next 1-2 years or even more.

When you think about it more then standard for DDR3 didn't change in last 3 years. It's still about 1600 but for example I was buying 1600 7-7-7 memory back then and now there is no way to find CL7 kits like that and all are on looser timings. With IC all is looking worse than ~2 years ago when you could buy Elpida or PSC chips with really tight timings and high clocks together.

You won't really see any difference above 2133 for IB except benching. While benching new kits are not as fast as you can read in reviews. In some magical way in all reviews you can find info about best and fastest 2400-3000 kits with top scores which speed in real base on IMC and cpu clock. You can compare 2133 9-11-9 memory that has 2 years now to new 2600 10-13-13 ... good marketing is making miracles. You won't find any older IC in reviews comparison ...
Since we have almost only Hynix and Samsung IC in enthusiast lines I would pick something like RipjawsZ 2133/2400 CL9 over any higher/newer series that are now on market.
 
Last edited:
So the question. I assume that lower timing/latency for the same frequency is better and worth something. But what speed should I be looking at for real-world applications that will still be suitable for hardware two years from now. All applications, nothing specific. I might game, compress video, fold molecules, whatever. If a guru or two would kindly rub their crystal balls and their HP 11C and give me the basis for selection (or a preferred number) of the most cost effective speed I would really appreciate it.
IMHO, Woomack is spot on. I'd go with 2133 and the best timings you can get. In 16GB kits, it looks like this G.Skill set is the tightest on the egg.

Above 2133, you need to be benching to see any difference.
You won't really see any difference above 2133 for IB except benching. While benching new kits are not as fast as you can read in reviews. In some magical way in all reviews you can find info about best and fastest 2400-3000 kits with top scores which speed in real base on IMC and cpu clock. You can compare 2133 9-11-9 memory that has 2 years now to new 2600 10-13-13 ... good marketing is making miracles. You won't find any older IC in reviews comparison ...
Since we have almost only Hynix and Samsung IC in enthusiast lines I would pick something like RipjawsZ 2133/2400 CL9 over any higher/newer series that are now on market.
FWIW, you can compare an older 2133 / 9-11-9-28 kit with a 2666 / 11-13-13-35 kit in our review. I also didn't use my old set of Pis (2400 / 9-11-9-28) because you simply can't find them any more. No sense in comparing something nobody can get their hands on. :salute:
 
Thanks Hokie for that link ( I'm just too lazy today to browse results so I said that fast from my head :p ) . So there is +/-1% in everything ;)
There are also RipjawsZ 4x4GB 2133 CL9-11-10 , 4 sticks but $40 cheaper -> http://www.newegg.com/Product/Product.aspx?Item=N82E16820231503
I'm not sure if getting 2x8GB with plans to upgrade it later is best idea as soon will be new IC on market, other timings etc and problems will start. You already see it on X79 where mixed kits are not really working.
 
Thanks Hokie for that link ( I'm just too lazy today to browse results so I said that fast from my head :p ) . So there is +/-1% in everything ;)
There are also RipjawsZ 4x4GB 2133 CL9-11-10 , 4 sticks but $40 cheaper -> http://www.newegg.com/Product/Product.aspx?Item=N82E16820231503
I'm not sure if getting 2x8GB with plans to upgrade it later is best idea as soon will be new IC on market, other timings etc and problems will start. You already see it on X79 where mixed kits are not really working.
NP. I've said it before and I'll say it a million times hence, if you don't a) bench, b) want to say you have fast RAM or c) just enjoy playing with high speed RAM, > 2133 doesn't really matter. 2133 with tighter timings really is the sweet spot for price and performance IMHO (as long as you don't get 2133 with CL 11 or something silly like that).

Also, OP, this man speaks the truth...if you're sure you're needing 32GB, get it now, all in one kit. Don't count on even being able to get different ICs that run the same speed and timings down the road. The market is fluid and manufacturers are having to use whatever ICs they can get their hands on. Even IF you get a kit with the same speed, timings & voltage, you may not get them to play nicely together.
 
would I be looking for broken ram if I tried to run my 1866 ram at 2133 or so? Right now I'm running 2x2gb of 2200 and 2x4gb of new 1866, running both at 1900 (all four sticks are gskill highish end). Should I try higher, or just be happy where theyre at, which is what I was planning on, but would take faster if safe...
 
I run my 2133 at 2400, and so far so good.
(had to do some memtest to get all the timings without error.. but it worked fine)

I say keep pushing until you hit VDIMM 1.70V and VCCAS 1.35V and go no higher on the voltage, and see what is the highest level of overclock your RAM can sustain, and VIOLA.
 
Personally, I think 1600 or 1833 is the best price/performance. Any faster and you're looking at a only couple % faster for almost 100% more cash. And keep in mind that a 5% percent increase in memory bandwidth will be hardly noticeable in any real-world application.

I'd recommend some 1600 or 1833 DDR3 sticks. I just picked up 2x8GB DDR3-1600 RipjawX's for about $100. Those will definitely keep you until DDR4. I read an article last week somewhere that Micron (I think) is just now starting to sample DDR4. It'll be a loooooong while before they're in mass production.
 
Back