FEATURED G.Skill Trident Z 2x8GB DDR4-3200 CL14 - F4-3200C14D-16GTZKW

[Woomack]

Today I have Trident Z 2x8GB CL14 - F4-3200C14D-16GTZKW kit which I have for over half year but somehow never made full tests ( results are around the forums ).

The same as most Trident Z at tighter timings also here we can find Samsung "B" IC which is the best what you can find for overclocking.

Modules are black/white as you can see on the G.Skill website. I guess all know how Trident Z looks like

Here you can see more detailed info about memory modules with SPD/XMP profiles:



XMP profile works good on MSI Z170I Gaming Pro AC/Z170M Mortar, ASRock Z170 OC Formula, ASUS Maximus VIII Hero/Ranger, ASUS Maximus IX Apex, ASRock X99M Killer 3.1 and ASRock X99 Taichi. On these motherboards I had a chance to test this memory ( maybe I've missed something ).

Below are results at XMP settings and 1h stability test ( I'm just sure this memory is fully stable and I don't have to waste whole day to run stability tests ).

 

[Woomack]

Overclocking Results

3600 16-16-16 1.35V

Spoiler

3866 17-17-17 1.35V

Spoiler

4000 19-19-19 1.35V

Spoiler

3200 13-13-13 1.40V

Spoiler

3600 15-15-15 1.40V

Spoiler

3866 16-16-16 1.40V ( couldn't run any tighter till 1.55V+ )

Spoiler

4000 18-18-18 1.40V

Spoiler

4133 19-19-19 1.40V ( full auto + changed main timings )

Spoiler

4133 17-17-17 1.40V ( different manual settings )

Spoiler

3600 14-14-14 1.55V

Spoiler

 

[Woomack]

Additional results

Max clock on air, dual channel - 2253MHz/DDR4-4506

Spoiler

AIDA64/Maxxmem @4400 17-17-17

Spoiler

XTU @4000 12-12-12

Spoiler

Geekbench @3866 12-12-12

Spoiler

Cinebench R11.5 @4000 12-12-12

Spoiler

Cinebench R15 @3866 12-12-12

Spoiler

Maxxmem @4000 12-12-12

Spoiler

wPrime @4000 12-12-12

Spoiler

 

[Johan45]

Nice results for such a low end set of mems.

 

[Mandrake4565]

Nice work Woomack, thanks for sharing.

 

[EarthDog]

Well done bart, thanks!

 

[Mandrake4565]

Bart a quick question just to be clear. This is the letter that indicates what Die it is?

 

[Johan45]

Yep that's the one

http://www.samsung.com/semiconductor/global/file/product/DDR4_Product_guide_May15.pdf

 

[MARCELO287]

Great work. How much VCCIO/VCCSA voltage are you using for 3200 13-13-13 1.40V?. 1.40V Is safe for 24/7?.
Thanks a lot!

 

[Brando]

lets see 10-10-10-20 @2400mhz (please)

 

[Woomack]

@ MARCELO287
VCCIO/SA depends on used motherboard and CPU "quality". 3200 can run at stock voltages on better motherboards, 3466+ requires higher IO/SA. For instance MSI Z270 Carbon required 1.35V IO/SA at 3733-3866 while ASUS Apex runs at 4000 at ~1.25V IO/SA.

@Brando
I don't know why you want 10-10-10 so much ( except problems with higher clock ). It doesn't change much in general performance.
Not every motherboard will run at 10-10-10. Some will need higher voltages, some won't run at all. Also not every memory will run at 10-10-10 but most kits can make CL10-x-x.
I wouldn't look for new memory only to tighten 1 or 2 timings some more. Your memory will probably make something close to 10-10-10, maybe 10-12-12 or something like that. Hard to say, I would test it first.

In some other thread you said that memory can't run at more than 2400. I wasn't following that thread so my question is if you can make it work at 2666 at 100bclk manually, then enable XMP but keep that 100bclk enabled. I just can't believe it won't work at more than ~2400. If it doesn't work at higher clock then I would make RMA as something isn't right with the CPU.

 

[Brando]

@ MARCELO287
VCCIO/SA depends on used motherboard and CPU "quality". 3200 can run at stock voltages on better motherboards, 3466+ requires higher IO/SA. For instance MSI Z270 Carbon required 1.35V IO/SA at 3733-3866 while ASUS Apex runs at 4000 at ~1.25V IO/SA.

@Brando
I don't know why you want 10-10-10 so much ( except problems with higher clock ). It doesn't change much in general performance.
Not every motherboard will run at 10-10-10. Some will need higher voltages, some won't run at all. Also not every memory will run at 10-10-10 but most kits can make CL10-x-x.
I wouldn't look for new memory only to tighten 1 or 2 timings some more. Your memory will probably make something close to 10-10-10, maybe 10-12-12 or something like that. Hard to say, I would test it first.

In some other thread you said that memory can't run at more than 2400. I wasn't following that thread so my question is if you can make it work at 2666 at 100bclk manually, then enable XMP but keep that 100bclk enabled. I just can't believe it won't work at more than ~2400. If it doesn't work at higher clock then I would make RMA as something isn't right with the CPU.

it's not a big deal i was just curious. none of this is life or death, just for fun. i'm mildly amusing myself by seeing how far down the ram can go and what the optimal combo of speed/latency is. i think 2600 was about where games got the best fps according to some review i read but that was in dual channel. i can do 2600 (or 2666 i forget) by messing with xmp but i wanted to see what i could do with as few changes to settings as possible to eliminate variables in instability. again not an emergency and probably won't make a difference i can feel.

EDIT: nah i double checked. can't go past 2400 ram speed without a 125 strap. with a 125 strap i have to either drop from 4300 to 4200mhz cpu speed or jump to 4375mhz and pump more voltage which my cooler isn't quite up for. i don't think slightly more bandwidth would replace that 100mhz lost but i could be wrong.

 

[Cheetos]

Spoiler

View attachment 188131

4133 17-17-17 1.40V ( different manual settings )

Created an account on this forum just to ask this question, amazing testing documentation and overclocking results by the way!

I am currently building from scratch a gaming computer and seeing how some games benefit a bit from faster memory I've ended up buying the RAM kit you just reviewed.
If one should be to run this memory kit and expect it to last for years of stable gaming, would it be unwise to run it at the above settings you achieved ? (4133 17-17-17 1.40V)

Running this memory this fast won't cause it to overheat ?

Running a 1.35v kit at 1.4v doesn't seem too bad, is it ?

 

[Woomack]

This memory will heat up about the same regardless of frequency. The main factor which causes memory to generate more heat ( and shorten its life ) is voltage. 1.35-1.40V is tested as safe by memory manufacturers. Actually most 4000+ memory kits have 1.40V in XMP profile.
I can't guarantee you that your setup will run stable at 4000+ CL17-17-17 at 1.35-1.40V. It depends on many factors like motherboard, memory controller etc. However there is high chance that it will work when you have Kaby Lake CPU and motherboard on which manufacturer is declaring support for 4133+ memory kits.
The best motherboard right now is ASUS Apex but there are some others with 4133+ support. Good overclocking motherboards aren't cheap so sometimes it's good to think if premium price is worth it. Memory is scalling the best up to ~4000 but each memory ratio like from 3600 to 3733 and next to 3866 is giving maybe 0.5% performance increase in applications which are using high memory performance. In most cases 3600 at lower timings will run about as good as 4000 at more relaxed timings. At the same time motherboard which fully supports 3600-3733 memory costs 50% of what you have to pay for guaranteed 4133+.

 

[wingman99]

Nice results for such a low end set of mems.

3200 CL 14-14-14 speed memory is the sweet spot for cost with the second best performance.

The time between sending a column address to the memory and the beginning of the data in response.
DDR4-2800 CL 14 =10.00 ns 11.07 ns 12.50 ns
DDR4-3200 CL 14 = 8.75 ns 9.69 ns 10.94 ns
DDR4-3600 CL 15 = 8.33 ns 9.17 ns 10.28 ns
DDR4-4000 CL 18 = 9.00 ns 9.75 ns 10.75 ns
DDR4-4266 CL 19 = 8.91 ns 9.61 ns 10.55 ns

https://docs.google.com/spreadsheet...mV_44QazuvsSjjTu-H4KyVfpg/edit#gid=2008389493

 

[EarthDog]

Think he meant overclocking headroom.

 

[Cheetos]

This memory will heat up about the same regardless of frequency. The main factor which causes memory to generate more heat ( and shorten its life ) is voltage. 1.35-1.40V is tested as safe by memory manufacturers. Actually most 4000+ memory kits have 1.40V in XMP profile.

Great! That's exactly what I want to know

I can't guarantee you that your setup will run stable at 4000+ CL17-17-17 at 1.35-1.40V. It depends on many factors like motherboard, memory controller etc. However there is high chance that it will work when you have Kaby Lake CPU and motherboard on which manufacturer is declaring support for 4133+ memory kits.
The best motherboard right now is ASUS Apex but there are some others with 4133+ support. Good overclocking motherboards aren't cheap so sometimes it's good to think if premium price is worth it.

I am waiting to receive my parts through the mail, ordered gigabyte Z270x gaming 7 motherboard with i7-7700k. I do understand the factor of "luck" that comes with any overclocking components as some parts simply are bad overclockers out of the box. Anyways, the z270 g7 board shows up as compatible on GSKILL's website and it is rated for 4133+ so I'm hoping for the best. Will report results here when it's up and running.
I did look into the APEX board but it was 140$ more expensive than the one I chose and couldn't justify this price.

Memory is scalling the best up to ~4000 but each memory ratio like from 3600 to 3733 and next to 3866 is giving maybe 0.5% performance increase in applications which are using high memory performance. In most cases 3600 at lower timings will run about as good as 4000 at more relaxed timings. At the same time motherboard which fully supports 3600-3733 memory costs 50% of what you have to pay for guaranteed 4133+.

What do you consider as the most efficient way to compare performance increase from memory overclocking ? I'd like to be able to say with confidence what kind of benefit I'm looking at from different OC settings.

 

[Woomack]

To know performance gain you have to test it on your hardware and applications which you are using. You can compare bandwidth in benchmarks but most of them are far from real difference in popular applications.
I would simply run memory at ~3200 cl14, ~3600 cl15/16, ~3866 CL16/17, ~4000 17/18 and run some tests like when you play games then some of them have internal benchmarks. Can run fraps or whatever is popular now to measure fps in games.
In most cases difference is so low that it doesn't matter how high you set your memory. Some titles can use fast memory and if I'm right then it's helping more at higher screen resolution.

 

[wingman99]

I have a memory latency of 10.6 ns and Woomack yours is 150.0 ns, did you have something running in the background?

 

[Woomack]

no, this soft is just not designed for newer OS, MM is always showing wrong latency