Overclock.net banner

Dual Channel vs Quad Channel and Timings - Confused

5K views 6 replies 4 participants last post by  Blameless 
#1 ·
I am reading up on DDR4 Ram as I prepare to once again try to push mine to the limits of my system and I am getting myself confused about some basic things. On G.Skill's site under Trident Series, I can select 32 Gig (8 Gig X 4) Dual Channel or Quad Channel, 3200 MHz tested speed, 14-14-14-34-2N and get the following part numbers.

Dual Channel - F4-3200C14Q-GTZ
Quad Channel - F4-3200C14Q-GTZSW

If I understand G.Skill's part numbering scheme, the SW at the end of the Quad Channel part number indicates the color scheme of the DIMMs and nothing more. Take that away and they are the same part number. If I select them and Compare, I see the image below.



The specs are exactly the same except one says Dual Channel and one says Quad Channel. So what is the difference between these 2 sets of Ram? They have the same part number except for the suffix that I am pretty sure just indicates color. They both are 8GB x 4 DIMM kits. The reason I am looking at the 3200MHz DIMMs is that I keep reading on this forum about how they are the best ones to get and I am wondering if I made a mistake when I bought my kit.

My MOBO is an Asus X99-Deluxe II which supports Quad Channel Ram- up to 128GB DDR4 @3333. When I was selecting components, I bought a set of G.Skill F4-3333C16Q-32GTZB which I got on sale at a very good price - a bit less than F4-3200C14Q-32GTZ. If I understand how this works, per the specs, the F4-3333C16Q-32GTZB has been tested (binned) to 3333MHz which is a guaranteed higher operating frequency than the F4-3200C14Q-GTZ which has only been tested to 3200MHz.

The other very crucial difference is the 3200MHz set has published timings of 14-14-14-34-2N verses 16-18-18-38-2N for the 3333MHz set. So the 3200MHz sticks have a CAS of 14 clocks, tRCD-14, tRP-14, tRAS-14, CR-2. The 3333MHz sticks have a CAS of 16 clocks, tRCD-18, tRP-18, tRAS-38, CR-2. So they can operate at a higher clock frequency but they take 2 clocks longer for CAS, and 4 clocks longer for RAS-to-CAS, Row Precharge, and Row Active times. They both have a CR of 2N (2T).

So at 3200MHz, 1 clock cycle takes .3125ns so 14 clocks takes 4.375ns. The whole cycle (CAS, tRCD, tRP, tRAS) takes 76 clocks = 23.75ns, then 2 clocks for CR = .625ns for a total of 24.375ns.

At 3333MHz, 1 clock cycle takes .300ns so the whole cycle for 16-18-18-38 which is 90 clocks = 27.00ns, then 2 clocks for CR = .600ns for a total of 27.600ns which is 3.225ns longer than the 3200MHz DIMMs.

Now I have been (unsuccessfully) struggling to get my system which is a Haswell-E i7-5820K to even boot with the XMP settings on the DIMMs of 3333MHz 16-18-18-38-2N, let alone trying to go to 1N (1T) or tighten any of the timings. I have since read that I might as well forget trying to get a Haswell-E based system to run at with Ram at 3333MHz due to IMC limitations though I might get it to run with Ram at 3200MHz.

So, my questions are:

1 - What's the deal with the G.Skill Dual Channel verses Quad Channel part numbers, that seem to be identical, and have identical specs except that one says Dual Channel kit and one says Quad Channel kit (both are 8 Gig x 4 DIMMs)?

2 - Are my calculations of the total time in ns for each kit above correct or do I not get it?

3 - Did I blow it buying F4-3333C16Q-32GTZB with latencies of 16-18-18-38-2N instead of an F4-3200C14Q-32GTZ set with latencies of 14-14-14-34-2N? If my calculations are correct, the 3333MHz kit will take 3.225ns longer per total command cycle than the 3200MHz kit, assuming I could ever get it to run at 3333MHz.

4 - If I set the system for 3200MHz Ram clock with the F4-3333C16Q-32GTZB DIMMs, should I be able to tighten their timings to 14-14-14-34-2N which would in effect be the same as the 3200MHz kit or is the F4-3333C16Q-32GTZB kit just inherently slower due to the additional clocks required for each timing stage despite the higher maximum clock rating?

5 - And last - the F4-3200C14Q-32GTZ is currently ON SALE at a good price. Would it be worth it to go ahead and buy this kit and swap them out for what I have? Will I see better stability or enhanced performance (benchmarks aside)?

If anyone actually reads all this and can offer some advice, I would really appreciate it. Thanks in advance!!
 
See less See more
1
#2 ·
G.Skill just bins the RAM to achieve those settings, and higher frequencies like that are difficult to achieve. Like you said , instead of starting with XMP speeds, I bet you can manually bring it down to 3200 or 3000 and tighten timing until access latencies are similar to the 3200 set... But trying yourself is the only way to know for sure.
 
#3 ·
Quote:
Originally Posted by SpeedyIV View Post

So at 3200MHz, 1 clock cycle takes .3125ns so 14 clocks takes 4.375ns. The whole cycle (CAS, tRCD, tRP, tRAS) takes 76 clocks = 23.75ns, then 2 clocks for CR = .625ns for a total of 24.375ns.

At 3333MHz, 1 clock cycle takes .300ns so the whole cycle for 16-18-18-38 which is 90 clocks = 27.00ns, then 2 clocks for CR = .600ns for a total of 27.600ns which is 3.225ns longer than the 3200MHz DIMMs.
This isn't how memory timings work.

All other things being equal memory set to 3200MT/s 14-14-14-32 is generally going to perform slightly better than 3333 16-18-18-38, but you can't just add the timings together because they interact with each other in a complex way, not simply serially.
Quote:
Originally Posted by SpeedyIV View Post

Now I have been (unsuccessfully) struggling to get my system which is a Haswell-E i7-5820K to even boot with the XMP settings on the DIMMs of 3333MHz
Not unusual.
Quote:
Originally Posted by SpeedyIV View Post

What's the deal with the G.Skill Dual Channel verses Quad Channel part numbers, that seem to be identical, and have identical specs except that one says Dual Channel kit and one says Quad Channel kit (both are 8 Gig x 4 DIMMs)?
Hard to say, other than one is officially certified to work on quad channel platforms and the other with two DIMMs per channel. Realistically, the binning probably isn't any different, but their may be some minor SPD/XMP timing differences to account for the intended platforms.
Quote:
Originally Posted by SpeedyIV View Post

Did I blow it buying F4-3333C16Q-32GTZB
I'm not an expert on what G.Skill kits do best, but I would be surprised if this kit was any worse than the 3200 kit at 3200.
Quote:
Originally Posted by SpeedyIV View Post

If I set the system for 3200MHz Ram clock with the F4-3333C16Q-32GTZB DIMMs, should I be able to tighten their timings to 14-14-14-34-2N
Quite possibly, but again, I don't have any hands on experience with these kits, nor have I researched their capabilities in detail.

You may want to check out the Haswell-E/Broadwell-E DDR4 memory stability thread and ask there.
Quote:
Originally Posted by SpeedyIV View Post

the F4-3200C14Q-32GTZ is currently ON SALE at a good price. Would it be worth it to go ahead and buy this kit and swap them out for what I have? Will I see better stability or enhanced performance (benchmarks aside)?
If you can return your existing memory and have more left over than the cost of the 3200 kit, it might be worth it. I highly doubt you'll see any perceptible difference in performance between the two kits though.
 
#5 ·
Quote:
Originally Posted by PontiacGTX View Post

if you plan to use DDR4-2666MHz+ then upgrad eto Broadwell-E or Get skylake, Haswell-E doesnt performs much better with higher ram frequency(or at releases benchs shown it)
Good suggestions but Broadwell-e is out of my price range for now and I don't want to switch to Skylake as I have an x-99 MOBO. I have gotten my Ram kit to 3000 MHZ which is a pretty good increase from stock. Have also gotten my 5820k to 4.6 GHz without too much trouble, though not with the Ram running at 3000 simultaneously. I have learned a lot since building this rig and now know that I will most likely never get a good solid 3200 or 3333 on the Ram while the CPU is running over 4 GHz. I will just have to live with the limits of Haswell-e and remember that in most if not all situations, CPU clock speed is the most significant performance increase for real world use.

Thanks!!
 
#6 ·
Quote:
Originally Posted by Blameless View Post

This isn't how memory timings work.

All other things being equal memory set to 3200MT/s 14-14-14-32 is generally going to perform slightly better than 3333 16-18-18-38, but you can't just add the timings together because they interact with each other in a complex way, not simply serially.
Thank you very much for addressing each of my questions. I am digging through the internet reading up on DDR timings and how they interrelate. Its interesting (though depressing) that despite clock frequency increases, memory latencies have really not gotten any better. My searching led me to stumble across an article on crystal phase change based memory that has the potential for picosecond timings. The article is at the following link if anyone is interested. Interesting stuff but academic to my current situation. http://www.extremetech.com/computing/233691-phase-change-memory-can-operate-thousands-of-times-faster-than-current-ram
Quote:
Originally Posted by Blameless View Post

What's the deal with the G.Skill Dual Channel verses Quad Channel part numbers, that seem to be identical, and have identical specs except that one says Dual Channel kit and one says Quad Channel kit (both are 8 Gig x 4 DIMMs)?

Hard to say, other than one is officially certified to work on quad channel platforms and the other with two DIMMs per channel. Realistically, the binning probably isn't any different, but their may be some minor SPD/XMP timing differences to account for the intended platforms.
I still don't get it. The 2 G.Skill kits I mentioned have the SAME part number except the last 2 characters that I still think specify the color scheme of the DIMMs. Maybe they actually are the same. I think I will post this question on the G.Skill forum. Maybe someone there can clarify this.
Quote:
Originally Posted by Blameless View Post

If I set the system for 3200MHz Ram clock with the F4-3333C16Q-32GTZB DIMMs, should I be able to tighten their timings to 14-14-14-34-2N

Quite possibly, but again, I don't have any hands on experience with these kits, nor have I researched their capabilities in detail.
I am going to try this. I also just ran RAMMON on the DIMMs and am going to use the best listed timings as a target. From what I have read, the performance increases from overclocking memory are much less significant that what is gained by overclocking the CPU. So, if I can't get stable performance pushing the Ram, I will back it off and go for the highest possible CPU clock. Does this make sense?
Quote:
Originally Posted by Blameless View Post

If you can return your existing memory and have more left over than the cost of the 3200 kit, it might be worth it. I highly doubt you'll see any perceptible difference in performance between the two kits though.
No I can't return the 3333MHz kit. It's been installed for several months. Someday maybe I will upgrade the 5820K to a Broadwell-e CPU and see if I can get the Ram to 3333 again. When I bought it, I thought you just enabled XMP and were off to the races. I was not aware of the limited ability of Haswell-e to handle overclocked Ram. One of it's favorable aspects was support for quad channel Ram so it was not apparent to me that the Haswell-e platform would suffer from this particular limitation. Live and learning every day. Thanks again!
redface.gif
 
#7 ·
Quote:
Originally Posted by SpeedyIV View Post

Thank you very much for addressing each of my questions. I am digging through the internet reading up on DDR timings and how they interrelate. Its interesting (though depressing) that despite clock frequency increases, memory latencies have really not gotten any better.
http://www.anandtech.com/show/3851/everything-you-always-wanted-to-know-about-sdram-memory-but-were-afraid-to-ask

Slightly dated, but DDR4 follows the same fundamental principles, even if adds a few new features to mask latency and increase parallelism.

The simplified (but still fairly arcane) state diagram does a good job of illustrating what I mean when I say timings aren't purely linear:

DRAM%20Simplified%20State%20Diagram.png

Quote:
Originally Posted by SpeedyIV View Post

From what I have read, the performance increases from overclocking memory are much less significant that what is gained by overclocking the CPU.
This is virtually always the case.
Quote:
Originally Posted by SpeedyIV View Post

So, if I can't get stable performance pushing the Ram, I will back it off and go for the highest possible CPU clock. Does this make sense?
I'd recommend finding the optimal CPU OC first and worrying about the memory last.
Quote:
Originally Posted by SpeedyIV View Post

I was not aware of the limited ability of Haswell-e to handle overclocked Ram. One of it's favorable aspects was support for quad channel Ram so it was not apparent to me that the Haswell-e platform would suffer from this particular limitation. Live and learning every day. Thanks again!
redface.gif
Overclocking isn't always a sure thing and Haswell-E doesn't officially support anything over DDR4-2133...so being able to get ~3200MT/s out of it's memory controller is still fairly impressive.
 
This is an older thread, you may not receive a response, and could be reviving an old thread. Please consider creating a new thread.
Top