Does Memory Matter?

[greenjizz]

Hi Guys just about to pull the trigger on a new build and I'm stuck on what memory to buy for overclocking. I was looking at the G.Skill Trident Z series but have no idea what speed I should get or even need for that matter. I was looking at the F4-3000C15D-16GTZB (2x8GB) but wasn't sure if that's good enough for overclocking or not. Should I also be looking at the F4-3600C16D-16GTZ (2x8GB)? Will the 3600mhz give me more wiggle room or better performance?

Specs
i7-6700k
Asus Maximus Hero
Ekwb EK-XLC Predator 240
EVGA Supernova 750 Gold

I'm not looking to break any records but would like a solid overclock.

Thanks so much for the help.

 

[poofyhairguy]

For gaming yes, it matters.

 

[Deders]

Not sure if you'd see much of a difference between 3000 and 3600 in games.

 

[cbrunny]

As I understand it, faster ram might give you one or two extra fps at the high end in games that have a heavy RAM dependency.

 

[lehtv]

I would gladly pay the difference between 2133 and 2666, and even between 2666 and 3000-3200.. but the really high clocked kits are just for bragging rights. Diminishing returns.

 

[Ken g6]

Personally I prefer the 2800 1.2V kits for stability.

 

[ElFenix]

actual answer: no
http://techbuyersguru.com/gaming-ddr4-memory-2133-vs-26663200mhz-8gb-vs-16gb?page=1

 

[DrMrLordX]

Considering the price difference between DDR4-3200 and DDR4-3600 kits of that size that I see on pcpartspicker:

http://pcpartpicker.com/parts/memory/#s=403200,403600&Z=16384002&sort=a10&page=1

I can't tell you that picking the 3200 kit or 3600 kit will be a better or worse decision either way. Are you willing to pay $40 for a slightly higher bin? You might have some more overclocking fun with the "better" memory, or you might be able to OC the 3200 to 3600 or beyond, who knows?

Either way, that's not a lot of money for higher-end DDR4.

 

[KingFatty]

Here is the pertinent citation from the Tech Buyers Guru (TBG) site posted by ElFenix:

But running RAM at 3200MHz taxes this platform so heavily that loose timings are necessary, and the RAM simply does not perform better than low-voltage 2666MHz sticks at tighter timings. We no longer recommend DDR4-2800 or DDR4-3000 despite their potentially lower timings because they require more voltage (1.35V) and a higher strap speed, leading to much higher power use at idle and more difficulty in CPU overclocking.

 

[Head1985]

actual answer: no
http://techbuyersguru.com/gaming-ddr4-memory-2133-vs-26663200mhz-8gb-vs-16gb?page=1

interesting but what site is that?btw testing in 1440P will be GPU limited 99% of time.

 

[Face2Face]

interesting but what site is that?btw testing in 1440P will be GPU limited 99% of time.

It's our own Termie's site. Admittedly, I would've like to have seen Fallout 4 make it's way in the comparison.

 

[know of fence]

People who insist including Zero on every axis are trying to prove a point. Seriously why even bother with the graphs, just replace them with the Voyager snapshot of the pale blue dot. Because from the right cosmic perspective, evidently nothing matters. :biggrin:

That said the most useful info I've seen, is included in this table.

It shows that CAS Latency times Clock Cycle Time equals True latency. The 9-9-9-24 latencies are multiples of this Clock Cycle Time(also known as t_CK), which you get by dividing 2000 by the memory clock.

2000/Clock * CAS = true latency

So for 3200 DDR4 RAM with 16 CAS we get a true latency of 12.3 ns, which is an improvement, over those examples in the table above. An Improvement of both latency and bandwidth. But not a latency improvement over 2666 15-X-Y-Z RAM in the examples on Termie's website, which amounts to about 11.25 ns true latency.
That's why we see the middle columns ahead in half of the games.

 

[Ken g6]

[thread=2452438]Sometimes[/thread], memory really matters. Check out the Ryse benchmark: Faster memory nearly doubles the framerate on the i3 6100.

 

[DrMrLordX]

It should be pointed out, that if you buy the "faster" DDR4-3200 or 3600 DIMMs, that you can always tighten timings or lower clockspeed to achieve whatever results you want. So it's not like you're prohibited from running DDR4-2666 if you think that extra vdimm is limiting your overclock.

But if all you're ever going to do is run DDR4-2666, may as well just buy that RAM, right?

 

[Shaun_Brannen]

People who insist including Zero on every axis are trying to prove a point. Seriously why even bother with the graphs, just replace them with the Voyager snapshot of the pale blue dot. Because from the right cosmic perspective, evidently nothing matters. :biggrin:


That said the most useful info I've seen, is included in this table.


It shows that CAS Latency times Clock Cycle Time equals True latency. The 9-9-9-24 latencies are multiples of this Clock Cycle Time(also known as t_CK), which you get by dividing 2000 by the memory clock.

2000/Clock * CAS = true latency

So for 3200 DDR4 RAM with 16 CAS we get a true latency of 12.3 ns, which is an improvement, over those examples in the table above. An Improvement of both latency and bandwidth. But not a latency improvement over 2666 15-X-Y-Z RAM in the examples on Termie's website, which amounts to about 11.25 ns true latency.
That's why we see the middle columns ahead in half of the games.

Where the axis starts is really a subjective question... mostly dealing with the data. If a numerically small change is actually significant one from a real-world perspective, zoom in on that crap. If it's meaningless, make it appear meaningless.

 

[myocardia]

actual answer: no
http://techbuyersguru.com/gaming-ddr4-memory-2133-vs-26663200mhz-8gb-vs-16gb?page=1

That's only because four channels of DDR4-2133 already has nearly as much bandwidth as two channels of DDR4-4266 would have.

 

[ElFenix]

That's only because four channels of DDR4-2133 already has nearly as much bandwidth as two channels of DDR4-4266 would have.

The 6700 results show the same lack of scaling on dual channel.

They tested single channel memory back in 2013 using one 4gb stick vs 2x4, and it didn't matter then.
http://techbuyersguru.com/does-memory-matter-4gb-versus-8gb-versus-16gb-gaming

Skylake may be better able to use the bandwidth than haswell, though.

It's our own Termie's site. Admittedly, I would've like to have seen Fallout 4 make it's way in the comparison.

That I did not know when I found it from Reddit.to termie

 

[know of fence]

That's only because four channels of DDR4-2133 already has nearly as much bandwidth as two channels of DDR4-4266 would have.

Good catch, we see only latency improvements in those graphs, because bandwidth is more than abundant. But bandwidth rarely creates a limit even in dual channel mode, mostly with encode and zipping and similar applications as shown in RAM scaling tests of yore.

The 6700 results show the same lack of scaling on dual channel.

These scale better than quad channel for both min (3.1%) and avg. (2.3%) improvement. Goes to show that bandwidth is more significant than true latency also.

 

[myocardia]

Skylake may be better able to use the bandwidth than haswell, though.

Yeah, for whatever reason, Skylake can take advantage of higher bandwidth RAM, in a few of the newest games. Fallout 4 is the one where it seems to make the most difference, IIRC.

 

[Termie]

Hey all, I won't comment on the results for obvious reasons, but I wanted to say I appreciate the feedback.

As for game selection, one of the reasons Fallout 4 wasn't included is that's it's pretty buggy, which is a problem when trying to track pretty minor performance variations. And Ryse was included specifically to see if Eurogamer's prior findings would play out again on a quad-core system. No dice there.

 

[nerp]

Doesn't matter.

 

[tweakboy]

Who cares about benchmark,, see if you can play games now. and OCing your RAM aint gonna do sh*T for you ,,,gl

 

[know of fence]

[thread=2452438]Sometimes[/thread], memory really matters. Check out the Ryse benchmark: Faster memory nearly doubles the framerate on the i3 6100.

Am I wrong to assume that a 25% increase in RAM bandwidth should not be able to create performance increases higher than 25%? Which makes the 5 - 15% bigger FPS numbers throughout Eurogamer tests quite a huge effect.

Who knows what happed there with Ryse, maybe it's frame pacing or something like that, frames get perhaps thrown out or delayed when they miss the right time window.

 

[Headfoot]

actual answer: no
http://techbuyersguru.com/gaming-ddr4-memory-2133-vs-26663200mhz-8gb-vs-16gb?page=1

That is incomplete.

The game that gets hit the hardest by RAM is Fallout 4 which wasn't benched. From Techspot's review: http://www.techspot.com/review/1089-fallout-4-benchmarks/page6.html. I see Termie responded on this too. Don't take it the wrong way, thanks for doing the benching you do. Fallout 4 is simply a valuable datapoint for this discussion.

Further, Eurogamer showed that dual core Skylake gets hammered by slower ram in some games whereas quads dont get dinged as much if at all.

http://www.eurogamer.net/articles/digitalfoundry-2015-intel-core-i3-6100-review

The correct answer is, it depends.

If you're using a recent i3, especially Skylake, then get faster RAM because its not very much to step up from 2133 to even 3000. The second part is if you plan on playing a lot of Fallout 4, get faster RAM no matter your platform.

Additionally, I don't think its safe to assume a quad in dual core mode is equivalent to an actual dual core here. I trust testing with the actual dual core over simulated at least for memory/cache dependent tests like this. Shutting off cores in BIOS/UEFI may not shut off cache or memory resources that might be shared. I don't know enough of the low level details but I think its plausible enough to control it out by using actual dual cores

Anectdotally, I moved from a 4.5 ghz 2500k @ DDR3 1600 to 5820k @ 4.4 w/ DDR4 3200 and my fallout 4 frame rates were dramatically smoother with less drops on the same r9 290

 

[Face2Face]

Anectdotally, I moved from a 4.5 ghz 2500k @ DDR3 1600 to 5820k @ 4.4 w/ DDR4 3200 and my fallout 4 frame rates were dramatically smoother with less drops on the same r9 290

Nice memory overclock! Uncore?