*Official* NBR Desktop Overclocker's Lounge [laptop owners welcome, too]

I have an Asus Stix 1070 sitting here as well.

 

Good stuff! I'm ok with just GTX 1070 as it's enough for 1080p, even 144Hz if I turn down a couple of settings. At the moment I'm playing BF1, late starter(!), and it's maxed out apart from no AA, no motion blur, Lighting Quality High, and it's mostly stable at 142 fps (the cap I've set), with some drops to about 135fps (just from glancing at fps counter once in a while).

 

https://www.3dmark.com/3dm11/12735674 | http://hwbot.org/submission/3829784_

 

3DMark 11: https://www.3dmark.com/3dm11/12735721 | http://hwbot.org/submission/3829819_


3DMark 11 Extreme: https://www.3dmark.com/3dm11/12735815 | http://hwbot.org/submission/3829851_

 

Nice scores! I can barely get stable at 5.1ghz @ 1.46v. But I can run no problem 5.0ghz @1.35v. Big jump from 5.0-5.1 for me.

 

Thanks, bro. Yes, there is a lot of variance in bin quality with 87ooK. Silicon lottery is getting harder to win, or so it seems. Here is another one.

3DMark Vantage: https://www.3dmark.com/3dmv/5686690 | http://hwbot.org/submission/3829902_

 

This is quite interesting, you wanna try and buy Dual Rank RAM, that's not the same as Dual Channel by the way (we all know we want Dual Channel) - Dual Rank describes the fact there are 2 'sets'/'ranks'/'groups' of RAM chips on each RAM stick. At the following link, F1 2015 gained 30 fps by going Dual Rank.
http://www.tomshardware.com/reviews/corsair-vengeance-lpx-16gb-ddr4-4600-memory,5344-2.html

A lot of RAM bought now in 16GB kits (ie 2x8GB) are Single Rank, but there are a few kits around that are Dual Rank in 2x8GB - I've got one of those kits for instance. Most games are not gonna show 30fps difference at high refresh rates, but Dual Rank is better for gaming in general, and other CPU related tasks. The 32GB (2x16GB) kits out there, they're all Dual Rank from what I've read. Like I said, I've got one of those Dual Rank kits, and I tightened the timings on them ages ago too, given all this it's probably close the most ideal RAM you could get for gaming given it's Dual Rank & CL14: 16GB G.Skill Ripjaws V DDR4-3200 tightened to [email protected]. That's probably a good call considering gaming at 144fps is my thing. Skylake & Kaby Lake (& I guess Coffee Lake?) perform best with a total of 4 Ranks of RAM in the system: ie. 2 sticks of Dual Rank RAM

You can find out if you have Dual or Single Rank by using CPUz:


(Ha, when I was buying my RAM back in 2016 I didn't know about Dual Rank, so I got lucky in my choice!) From what I've read Dual Rank RAM can outperform even a higher overclocked Samsung B-die RAM that's not Dual Rank when it comes to gaming, so Dual Rank is a big deal for gaming from what I've read.

I found out about this stuff recently from reading that Toms Hardware link & I did some brief googling to confirm from other sources, and it seems to hold true, but if you have some links or testing that downplays Dual Rank's importance then I don't mind hearing about it of course.

 

You have to use a non-branded version of CPU-Z or it may not show that. The "special" ROG version does not. I found out about dual rank when I starting trying to overclock the RAM higher on the M18xR2 and P570WM. It does make a difference. The first G.SKILL Ripjaws DDR3 sticks I got were single rank and ran slower than the Corsair Vengeance dual rank DDR3 modules running at the same speed even though the G.SKILL timings were tighter.

 

3DMark Sky Diver: https://www.3dmark.com/sd/5065543 | http://hwbot.org/submission/3830030_

3DMark Fire Strike Ultra: https://www.3dmark.com/fs/15291397 | http://hwbot.org/submission/3830017_

3DMark Time Spy: https://www.3dmark.com/3dm/26035847 | http://hwbot.org/submission/3830010_

 

@bloodhawk - this vBIOS is working good. I figured out what was getting torn up on my W7 OS with the Strix vBIOS. It was the bootloader. When it crashed with a BSOD it was changing the boot priority in the BIOS and screwing everything up. So, now I know to stop and check that after a hard crash, LOL. Very odd, but at least it is simple and I can keep using the better vBIOS.

This big VRM heat sink and fan on the Hybrid cooler does a nice job as long as the fan is put on max. The power components are staying between the high 40s and high 50s with all this benching. Between runs I reset stock clocks and voltage and let it idle down to 135MHz and they almost immediately drop back down to 29°C-32°C. It's awesome never seeing any GPU-Z PerfCap reasons and the core clocks never moving from whatever I set (without K-Boost).

Wish I had $1300 to blow on a GPU. These are back in stock again.

https://www.evga.com/products/product.aspx?pn=11G-P4-6798-KR

...or, for an extra $200 versus an aftermarket water block...

https://www.evga.com/products/product.aspx?pn=11G-P4-6799-KR

 

Interesting that Windows 7 gets a better score on the hardest test (Blue Room). Seems par for the course and comes as no surprise, although I thought this newest benchmark would have been tilted in favor of Windows 10... at least artificially.

VRMark Orange Room: https://www.3dmark.com/vrpor/216060 | http://hwbot.org/submission/3830107_

VRMark Cyan Room: https://www.3dmark.com/vrpcr/11728 | http://hwbot.org/submission/3830095_

VRMark Blue Room: https://www.3dmark.com/vrpbr/38449 | http://hwbot.org/submission/3830086_

 

That's cool, so you have personal experience backing up the importance of getting Dual Rank RAM, that's good to know. It's a pity that there's not many 16GB (2x8GB) kits around nowadays that have Dual Rank RAM, (that & the sky high RAM prices!) - so, if you're buying RAM it looks like you definitely wanna make sure it's Dual Rank, tip for the day!

 

Would you say dual rank is more important, or less important, than dual channel for performance?

Reason I ask is, I'm receiving the system in my sig in the coming days, and I actually had the seller swap out the original 1x16GB dual rank stick for 2x8GB single rank sticks because I wanted dual channel. This was before I learned about dual rank and its importance.

 

To give you a quick reply before I do any research: I'd say go for the single Dual Rank stick with the plan to buy another 16GB dual rank stick when RAM prices are optimal, then you will have the best of both worlds. But I'll research the performance differences & update this post later, (a few other things to do now).

 

I did consider this route before asking the seller to make the switch, but decided against it after research because Corsair does not sell the 16GB sticks individually, only in sets.

 

You definitely made the wise choice going to 2x8GB for dual channel as opposed to 1x16GB single channel.

I would say dual channel is more important than dual rank or RAM speed. Single channel cuts your bandwidth in half and performance suffers more noticeably than if you had much faster memory with tight timings. Much better to have single rank sticks with a slower speed than a single dual rank stick at a faster speed.

When you get the new machine you can test it and so can @Robbo99999 right now. Run a benchmark like 3DMark 11 or Fire Strike. Make note of the score. Now, pull one of the sticks and repeat the benchmark.

 

After doing that research I talked about, I think Dual Channel is more important than Dual Rank, I couldn't find any specific comparisons for those 2 variables but in BF1 there is a massive difference between Single & Dual Channel (and that's a heavy CPU usage game, so makes sense):


If you know you're not going to be able or want to upgrade to Dual Channel 32GB later when RAM prices are better, then I'd advise pushing to get Dual Channel and Dual Rank - considering a lot of the 16GB (2x8GB) Dual Rank kits are older kits, how about trying to buy some off ebay second hand? Or maybe there are some new kits that do have dual rank in 16GB kits, they might be hard to find, but you might be able to. I'd just try to get the best of both worlds: Dual Channel & Dual Rank.

(Mr. Fox, I think I'll forgo the hassle of pulling one of my RAM sticks for testing - I think they're hidden underneath my monstrous NH-D14 air cooler!)

 

The importance of single and dual rank is news to me as well. And there's no way to determine the rank of the kit you're buying in advance? I was planning to buy the following kit:

http://www.gskill.com/en/product/f4-4133c19d-16gtzr

Should I assume it's single rank because it's a 2x8GB kit? AnandTech has the tighter CL17 variant pinned down as Samsung B-die:

 

I found a review for that RAM, which contains a CPUz screenshot, it's Single Rank. That's the 4133Mhz C19 version.

https://www.hardwareasylum.com/reviews/memory/tridentz_rgb-4133c19/page4.aspx

Perhaps you'll need to do the same for any RAM you're looking at getting - find reviews, find CPUz screenshots; it's a pain in the ass, manufacturers should really state in their specs on the website!

EDIT: found some threads with a list of Dual Rank kits!
http://www.overclock.net/forum/18051-memory/1665209-list-dual-rank-ddr4.html
https://www.reddit.com/r/Amd/comments/62vp2g/clearing_up_any_samsung_bdie_confusion_eg_on/
https://geizhals.eu/?cat=ramddr3&xf=439_dual~5828_DDR4~5830_UDIMM1~5831_DIMM

 

So in terms of speed, the confirmed list tops out at 3400MHz@CL16 (Trident Z) and 3466MHz@CL19 (HyperX Fury), and 8GB modules are a no-go. Single-sided vs double-sided means dies on one or both sides of the PCB, right?

 

Well, according to the links I gave you the Dual Ranked RAM tops out at 3400Mhz CL16, I didn't see the listing for 3466Mhz RAM (the 3466Mhz RAM was listed only because they required a Typhoon Burner screenshot, so 3466Mhz RAM is not confirmed Dual Ranked from that info). I'd try and get that 3400Mhz CL16 RAM if I were you! 3400Mhz is a good frequency and CL16 nothing to sniff at, but you'll probably be able to tighten those RAM clocks when you install them in your system. I think the Dual Rank nature of those RAM sticks will make up for it 'only' being 3400Mhz and I'm guessing they'll be cheaper than 4000+Mhz sticks.

And to answer your question, it looks like Dual Rank sticks are double sided, but I think I read somewhere that single rank sticks are double sided sometimes too (not sure) - but of course you want Dual Rank, double or single sided is not something to concern yourself with I think.

And your other question, there are some 16GB kits (2x8GB sticks) listed in those Dual Ranked Listings, so looks like you'll be able to get your 16GB total RAM you wanted in Dual Rank.

 

The 3466MHz kit was listed on Geizhals. Specifically, it's the Kingston HX434C19FBK2/32. In any case, the 3400MHz kit should perform better due to its tighter timings.

What worries me is that the pool of data which shows DR > Faster SR from Tom's Hardware is limited. There are games and applications which greatly benefit from high frequencies too:

https://www.techspot.com/article/1171-ddr4-4000-mhz-performance/page3.html



Aesthetics are important to me for my upcoming build, so I'm willing to pay a little extra if I don't compromise on performance doing so. The single-rank 4133MHz Trident Z RGB modules (F4-4133C19D-16GTZR) are 20% more expensive than the dual-rank 3400MHz Trident Z Silver/Red (F4-3400C16D-16GTZ). The Silver/Red would clash with my intended black & purple theme, so I'd consider buying the Ripjaws V variant (F4-3400C16D-16GVK) instead, but that kit would be an import and not significantly cheaper as a result. I have to wonder if dual-rank is important enough on average to lose 733 MHz for gaming.

The video above shows a diagram of a single-sided, dual-rank memory module. I think you're right in assessing that whether the sticks are single or double sided is independent from rank. The reason I brought it up was hoping for an easier way to identify dual-rank modules.

My bad. I read your post whilst commuting, so my reply was hastily written.

 

The problem with that Techspot article is that is says nowhere in it what the timings are for the RAM that is not the 4000Mhz version, so the comparisons are a little invalid because we don't know what they're comparing - of course all other variables being equal then more Mhz = more performance. Yes, there's not a huge amount of info on Dual Rank, but that Toms Hardware stuff is impressive, and it shows benefits in other things apart from gaming too. Hell, the 3200Mhz CL15 Dual Rank Kit beat out the 4000Mhz CL18 Single Rank Kit (and that's even a lower CL rating than the CL19 4133Mhz kit you're talking about). So, in this light you can quantify Dual Rank variable being better than 800Mhz difference - (and importantly I think we can call the CL values "equal" given the corresponding Mhz increase) - so yes I think we could say Dual Rank counts for more than 800Mhz speed difference.

Well if you're gonna bring aesthetics into it, then this complicates matters, and then it's your judgement call about how important aesthetics are to you versus a small'ish to large measurable increase in performance depending on game (F1 2015 showing large gains, and how many others show large gains - we don't know as testing pool very small, but you'd think CPU intensive games stand more chance, perhaps in BF1 too as that one showed large gains from Dual Channel). The 3400Mhz Ripjaws V variant looks like it would be faster than the 4133Mhz variant given the more than 800Mhz performance benefit of Dual Rank that I loosely postulated, and it's a bit cheaper too. Minimal gains perhaps. If you know you have more than enough CPU performance headroom for all your games then you might want to prioritise aesthetics - you can overclock your CPU which is likely to more than offset losses from going single rank. It's your judgement call, but in an ideal world you'd get Dual Rank & the fastest Mhz you could afford/want at the tightest timings.

 

Admittedly, that is sketchy, and odd that they wouldn't include the timings in a review. They did mentioned that the tested modules have 19-21-21-41 timings @ 4GHz though, and based on the opening statement, it sounds like they used the same kit all along, simply adjusting its multiplier. That could very well mean automatic timings, but they had the following to say about their findings in regards to the matter:

I don't doubt Tom's Hardware as a source, nor do I doubt Mr. Fox's own findings. There's definitely a significant performance difference between single and dual-rank under the right conditions, as established. When it comes to memory and gaming, however, it seems as though we're mostly seeing significant differences in outliner, memory-sensitive titles, and Tom's Hardware has only provided one such title, F1 2015. The test of Metro Last Light Redux shows that it is not one of those outliners, because performance is very close regardless of memory used (rank & speed). Game performance is my first priority, and so far I've come across more game benchmarks that show a higher framerate from using faster memory, but I realise there's been a much greater focus on speed than rank. However, Tom's Hardware isn't alone in testing single vs dual, as ComputerBase also has an article where, from what I can tell, single-rank 3200 MHz outperforms dual-rank 2666 Mhz in the games tested.

https://translate.google.no/transla...ore-i-ryzen-ddr4-ram-benchmark/3/&prev=search

This might indeed be true for several tasks, though the benchmarks performed are still too limited to establish that dual-rank is always better when there's a great frequency discrepancy, especially in gaming where only one out of two titles tested showed a big improvement. I'm not able to rest assured that the debate of single vs. dual-rank can completely overshadow the frequency debate when there are gains to be had there as well.

Yeah, sorry. I didn't mean to make this about my plans for a future build. I brought up aesthetics as a reason to why I can more easily justify a 20% price increase for faster RAM, assuming it won't have worse performance in most games than slower, cheaper dual-rank modules. The interpolation here is killing me, as I'd much rather know for sure how many games are greatly affected by single vs. dual-rank, compared to sub-3000 Mhz vs. 4+ GHz.

 

LOL, yeah, it's a right minefield we find ourselves in, it seems there's not enough information to reliably form conclusions!

Regarding your 3200Mhz vs 2666Mhz remark, maybe there' s a minimum Mhz that needs to be achieved and then beyond that Dual Rank plays a greater role - old law always decreed that you needed 3200Mhz DDR4 for an optimum gaming experience and Mhz increases above that were less significant. In fact at that Toms Hardware link they say they saw an increased benefit of Dual Rank when they overclocked their CPU, the differences in F1 2015 re Dual Rank really weren't that pronounced when the CPU was at stock. There seems to be a lot of variables & circumstances coming into play here, and I don't understand how they all fit together, I'm not sure many people in the tech world do.

You should choose whichever RAM you want, but I'm hedging my bets on the highest Mhz (at least 3200Mhz) Dual Rank kit you can find that has the tightest timings, and then I'd install that kit & tighten the timings even further - all of the timings all the way up to tRFC; this is my intuition on how to get the best performance based on everything I've seen & experienced, for what it's worth!

EDIT: @Convel ,Just to add. I have the F1 2015 game myself, and I get 179 fps average in the benchmark, which is significantly more than the 156 fps figure that Toms Hardware achieved. Toms Hardware were rocking an overclocked 7700K and GTX 1080, I have an overclocked 6700K and GTX 1070 - so my system is doing something right, perhaps my Dual Rank RAM with very tight timings on CL and other timings are helping even more, my CPU is also at 4.7Ghz on the core & ring. I'm not CPU/Platform limited in that game at all, the GPU is always at 100% load during that benchmark. I think a lot of variables add up, small gains on lots of different variables could equal significantly increased overall performance in some titles perhaps.

 

I think it's not as complicated as all that. Dual rank is better, but if you take two kits running at the same speed and timings, the difference between dual and single rank will be likely be more measurable than meaningful. By that I mean, it might show minor improvement on a memory benchmark and nowhere else. Speed and multi-channel are the most important things affecting performance. Any differences between single and dual rank are not huge, even though one is better than the other. I would rather have single rank 4000 CL18 than dual rank 3200 CL16. Requiring the system to function in dual channel mode (or quad channel where applicable) regardless of memory speeds or timings is just an unspoken basic expectation that goes without saying.

 

Well F1 2015 was more than measurable, it was a 30 fps difference between Dual and Single Rank, and the Single Rank kit was a faster Mhz too. Now that title may be an exception, but even the Dual Rank kit beat the faster Mhz kit at most non-gaming stuff shown on Toms Hardware. Also, you yourself said that Dual Rank was more important than somewhat tighter timings in your own actual experience. I'm not sold on 4000 CL18 being better than dual rank 3200 CL16, although thinking about it CL18 on 4000Mhz is a lot tighter than 3200 CL16, CL14 on 3200Mhz RAM is more comparable in tightness. Ha, it all comes down to how tight you like it and how fast you like to go, and Convel likes the pretty ones anyway! Arghh, too much information, sleep time for me now!

EDIT: and my final thought before putting my head down, I'm still a Dual Rank fan - maybe I'm biased!

 

Yeah, I am still a dual rank fan myself. No reason to even think about buying the single rank stuff unless there is a massive difference in price or something like that. I think it would require a lot of testing of different scenarios to identify how much better and how often it makes a difference. There will always be a random one-off with a glaring exception to the norm like F1 2015. And, I am not sure how much of the expensive high end enthusiast RAM is even made in single rank. If there are seldom any examples of it to speak of, then it might not even be relevant, LOL.

 

Yeah, for example the large 16GB RAM sticks are all Dual Rank as far as I know. It does seem like there is conflicting evidence showing how important or unimportant Dual Rank is. My general hunch is that often it is touted that above 3200Mhz minimal gains are seen, so my intuition tells me to throw in another positive performance variable into the mix to try to make some gains, ie Dual Rank & tightening timings. Ok, I probably won't say much more on this topic otherwise we'll have to re-label this thread "Dual Rank Quandries", not that it's off topic to overclocking & performance but we would have to rename the darn thread!

 

After all of this, I can certainly see more of an incentive to go 2x16GB That's a lot of mullah in today's market though, and most people don't spare memory a second thought.

 

Well, here is my Aida64 Memory Test.
Single rank

 

Lots & lots of bandwidth, latency not so great, but I think that's to be expected of that architecture, because it has a mesh a bit like the Ryzen CPU (rather than the ring of 6700K/7700K/8700K), so they suffer from increased memory latency.

 

I just ran it, didn't do any thing in the way of tinkering and adjusting and lower timings or cl. Plus i think the amount of cores makes a lot of difference in latency. Speculations of course.

 

Fire Strike Extreme: https://www.3dmark.com/fs/15305561 | http://hwbot.org/submission/3831480_

 

A little bit of tweaking...

 

1 more...
All at 4.8 Ghz

 

Yes, but you have four modules, so you get four memory banks just like those with two dual-rank sticks. Difference being you obviously have quad-channel memory on X299, and your single-ranks clock higher

...though I think two modules is considered optimal for stability when overclocking? The ROG X Apex only has two memory slots.

 

I knew it was going to be someone with a negative reply as usual. Did you actually read what was said first? I mean from the start?


Heres a better one @Robbo99999 51.8NS

 

No negativity intended, and I know your knowledge on the matter exceeds mine. My post was intended as a pat on the back, in the sense that you take the win in several memory aspects. A sum-up of sorts.

 

Then you have my Apologies sir!

These pictures were more about the latency in the first picture. 66ns which is down to 51.8ns.

 

If I had a nickel for every time I misjudged the intention of a post... Often, the poster has to take some or all of the blame for being unclear, and this instance is no different. This remains one of the friendliest and most welcoming threads on the forum, so don't worry about it.

An impressive feat, giving you read speed bump of 15 535 MB/s too! Did you start off with XMP timings before finding better ones?

 

Memory is not my strong suit. LOL, but I was able to use a memory profile to start with and then go from there. My memory is XMP 3600, but not using an xmp profile.

 

That's a big decrease in latency, well done! That's probably the weakest point of that CPU architecture, so I think you do right to lower the latency as much as possible. The latency of my Skylake system is 'only' 41ns:

It's a lot easier to hit lower latencies on 6700K/7700K though - it's that mesh type design of your many core CPU that increases the latency (like the Ryzen), which is one reason why they're not the best CPUs for gaming, but of course they're better at productivity & the raw CPU power is many times greater.

 

Looks like it barely beats my M18XR1.

 

lol, yeah! But bandwidths are different of course between my desktop & your laptop (which is DDR3!), just like how your quad channel desktop has massive bandwidth - we try to maximise the performance of all our components for max performance. But you're right to lower the latency of your desktop RAM, I think it's the weakest point of that architecture like I said in my last post, and you have loads of bandwidth already with that quad channel setup, so I'm guessing you'd see most gains by focussing on reducing latency.

 

@Robbo99999
Yes, I hear what you are saying, that's for sure. Would not have even pressed the issue if it were not for reading the exchange between you and brother Fox. Thought to myself...yes, I think I can do better after I quite being lazy. So to you sir I say, Thank you!

 

Well that's alright! You reduced latency by a huge amount, I think that could pay some dividends.

 

Wow, was not expecting this ram to run 4k....

 

Not quite stable though... 4242Mhz