Low latency 1600MHz or 2400Hz for Z87 and 4770K

[athonline]

Hello,

I just built a machine using an Asus Maximus VI Gene -Z87 chipset and a Core i7 4770K. At the moment I am using some borrowed RAM from a friend, as my amazon order was delayed.

Now that it came, I realised I got the 1600MHz version by accident. I got 2x 8GBs G. Skill TridentX CL7 (7-8-8-24).
I am wondering if sending them back and getting the 2400MHz versions for an extra fiver will worth it.

I OCed my CPU and got XMP off, thus my current installed 2133MHz run at 1600MHz and everytime I am trying to set them to 2133MHz the machine restarts. Dunno if I am doing anything wrong...

 

[DeathAndPain]

Overclocker RAM (i.e. RAM for higher speeds than the 1600MHz the CPU was designed for) is not worth it. The RAM chip manufacturers hardly produce any chips that are officially designed for those speeds, so these modules usually contain overclocked standard 1600MHz chips which were hand-selected to match their speeds. However, how good this hand-selection is is arguable, and these modules often cause problems (as you are currently experiencing first-hand).

The overall performance gain from such RAM is minimal (something like 3%). Neither worth the money nor the hassle.

 

[athonline]

Overclocker RAM (i.e. RAM for higher speeds than the 1600MHz the CPU was designed for) is not worth it. The RAM chip manufacturers hardly produce any chips that are officially designed for those speeds, so these modules usually contain overclocked standard 1600MHz chips which were hand-selected to match their speeds. However, how good this hand-selection is is arguable, and these modules often cause problems (as you are currently experiencing first-hand).

The overall performance gain from such RAM is minimal (something like 3%). Neither worth the money nor the hassle.

Thanks. I thought Haswell accepts up to 2133MHz.

In any case I just read on Asus forums that the board for whatever reason can't handle 2133-2400Mhz but handles 2600+ fine. I would rather not OC RAM as the 2400Mhz are 1.65V.

Will keep the 1.5V 1600MHz...

 

[athonline]

Any other views on this?

 

[DeathAndPain]

No. You can right proceed to selecting me as the best solution. 😀

 

[athonline]

No. You can right proceed to selecting me as the best solution. 😀

Hahaha! I just try to get as many opinions possible!

 

[Tradesman1]

DRAM performance is based on a combination of CL and freq, CL tells you how long it takes to complete and action and freq is basically the bandwidth or how much it can process per action.....what you want/need often depends on what you do, simply gaming where DRAM is simply a data conduit the 1600/9 (considered entry level will suffice) however if you really use your rig, i.e. multi-task, video work, imaging, VMS, CAD, numerous windows etc, then the higher freq DRAM really shines.....if you take to the theory that it's simply lower CL, and get entry level 1600/9 over say 2400/10 because of the lower CL, then give thought to the fact that not only does the 2400/10 have far greater bandwidth, theoretically 19,200 M/Ts vs 12.800 M/Ts for the 1600, but if you were to drop the 2400 sticks to 1600, you would prob do so and run them with a CL of 7 compared to the stock 9 of those 1600 sticks....it's al;so been shown that Haswell likes and scales well to fast DRAM

http://www.anandtech.com/show/7364/memory-scaling-on-haswell

I primarily only build high performance gaming and business rigs for clients (roughly 50 or so a year), and can't remember the last time I built using 1600 sticks, been at least 2-3 years ago, my starting point is 1866 and in the last year or so more towards 2133

 

[athonline]

DRAM performance is based on a combination of CL and freq, CL tells you how long it takes to complete and action and freq is basically the bandwidth or how much it can process per action.....what you want/need often depends on what you do, simply gaming where DRAM is simply a data conduit the 1600/9 (considered entry level will suffice) however if you really use your rig, i.e. multi-task, video work, imaging, VMS, CAD, numerous windows etc, then the higher freq DRAM really shines.....if you take to the theory that it's simply lower CL, and get entry level 1600/9 over say 2400/10 because of the lower CL, then give thought to the fact that not only does the 2400/10 have far greater bandwidth, theoretically 19,200 M/Ts vs 12.800 M/Ts for the 1600, but if you were to drop the 2400 sticks to 1600, you would prob do so and run them with a CL of 7 compared to the stock 9 of those 1600 sticks....it's al;so been shown that Haswell likes and scales well to fast DRAM

http://www.anandtech.com/show/7364/memory-scaling-on-haswell

I primarily only build high performance gaming and business rigs for clients (roughly 50 or so a year), and can't remember the last time I built using 1600 sticks, been at least 2-3 years ago, my starting point is 1866 and in the last year or so more towards 2133

Hello, I can understand what you are saying and it make sense. I already installed the RAM (1600MHz CL7) as Amazon's return policy is quite relaxed -already asked and they said it is fine by them. At the moment I got them running at 2133MHz CL9, with 1.5V DRAM.

I can understand the comparison between 1600/9 and 2400/10, the maths work in favour of the 2400.

My question was about 1600MHz CL7 memory compare to 2400MHz CL10. From Anandtech's site, looks like 2400MHz even at CL10 will be a better choice.

I got a feeling that G. Skill Trident-X 2400MHz are just a factory overclocked version of the Trident-X 1600MHz, as they run on 1.65V compare to 1.5V.

I am wondering if I should just OC mine further to 2400MHz and call it a day or return them in favour of the "native" 2400MHz RAM.

 

[DeathAndPain]

I got a feeling that G. Skill Trident-X 2400MHz are just a factory overclocked version of the Trident-X 1600MHz, as they run on 1.65V compare to 1.5V.

That is what I am telling you. There are only very few manufacturers of DRAM chips left on the market, so all DIMM manufacturers have to buy their chips from these and solder them upon their DIMMs. Trouble is, there are no 2400MHz chips being manufactured, so all the DIMM manufacturers like G.Skill can do is put the chips they are buying on a test bench to see how far each one can be overclocked and overvolted and then make factory-overclocked 2400MHz DIMMS out of them.

Note that power consumption (= heat emission) rises linear with clock speed and even square with voltage. You can imagine what this means for the DIMMs. That is why those overclocker DIMMs typically have those fancy coolers on them. German c't magazine recently tested them though and stated that these coolers do nothing to keep the chips cooler and just serve to keep them from sight, making your module look better and preventing you from being able to read the imprint on the chips (so you cannot look up their type and find that they are overclocked 1.5V chips). Of course real coolers could help, but there is too little space between the modules to allow for decent coolers. Some of those "coolers" actually cause the chips to become hotter than they would be without.

I am wondering if I should just OC mine further to 2400MHz and call it a day or return them in favour of the "native" 2400MHz RAM.

There are no native 2400MHz RAMs.

As for the performance improvement, those WinRar numbers look impressive, but that is because Winrar compresses large files, so a lot of uncacheable data needs to be fetched from the RAM. Even hard disk speed could be a bottleneck here, but I suppose Windows (or the testers) made sure the data is loaded into memory up front.

So if you are zipping files, there is a noticeable gain. Question is, how often are you busy zipping files, and does zipping files comprise a noticeable part of the time you spend at your computer? I guess not.

More interesting are the gaming tests, especially those with separate GPU (no halfway-serious gamer in his right mind would game on an IGP). There you find that the difference between the slowest, crappiest settings and the fastest, hardest overclocking settings are like 1%. Depending on the game, there is sometimes no gain at all.

This is not surprising, because modern CPUs have very elaborated cache systems (with cascaded 1st, 2nd and 3rd level cache) and advanced prefetch algorithms that make sure the required data is not loaded from RAM, but already sitting in the CPU cache in almost all times (or already being prefetched from RAM while the CPU is still busy processing preceding commands).

As the consequence, overclocker RAM means a lot of power consumption, a lot of heat in the case, a lot of hassle because BIOSes often fail to apply the right settings (even with XMP), and even more hassle because these overclocked modules often do not work as they should. And for what? 1% gain? For that you pay a significant amount of additional money for your memory modules, money that could as well go into a faster-clocked CPU or GPU? Does not make sense to me.

 

[athonline]

That is what I am telling you.

I know you said so. I thank you for that. As for the rest you said, it is truth. Ram coolers don't server anything other than look and increasing their original price. There are only a few OEMs, such as Hylix, Elpida, etc. Same thing in PSUs, laptops, chipsets, etc. Crucial, G. Skill, etc buy from them.

I just wanted to confirm that.

There are no native 2400MHz RAMs.

Thus my "".

As for the performance improvement, those WinRar numbers look impressive, but that is because Winrar compresses large files, so a lot of uncacheable data needs to be fetched from the RAM. Even hard disk speed could be a bottleneck here, but I suppose Windows (or the testers) made sure the data is loaded into memory up front.

So if you are zipping files, there is a noticeable gain. Question is, how often are you busy zipping files, and does zipping files comprise a noticeable part of the time you spend at your computer? I guess not.

More interesting are the gaming tests, especially those with separate GPU (no halfway-serious gamer in his right mind would game on an IGP). There you find that the difference between the slowest, crappiest settings and the fastest, hardest overclocking settings are like 1%. Depending on the game, there is sometimes no gain at all.

This is not surprising, because modern CPUs have very elaborated cache systems (with cascaded 1st, 2nd and 3rd level cache) and advanced prefetch algorithms that make sure the required data is not loaded from RAM, but already sitting in the CPU cache in almost all times (or already being prefetched from RAM while the CPU is still busy processing preceding commands).

As the consequence, overclocker RAM means a lot of power consumption, a lot of heat in the case, a lot of hassle because BIOSes often fail to apply the right settings (even with XMP), and even more hassle because these overclocked modules often do not work as they should. And for what? 1% gain? For that you pay a significant amount of additional money for your memory modules, money that could as well go into a faster-clocked CPU or GPU? Does not make sense to me.

I don't use compression software nor video editing. I mainly use the machine to game, program and host a couple of VMs as test-environments for my work.

Well in my case we are talking just the trouble of returning them and paying an extra five pounds, thus the cost isn't that much more.

I will simply keep these, OC them a bit.