FEATURED G.Skill Trident Z Neo 2x8GB DDR4-3600 CL14 - F4-3600C14D-16GTZNB /w Ryzen 4650G

[Woomack]

Today I will present some results on one of the best enthusiast memory kits designed for AMD Ryzen 3000 - G.Skill Trident Z Neo 16GB DDR4-3600 CL14. I said it's designed for Ryzen 3000 as the memory is on the market for some months already and it perfectly fits optimal memory frequency of every Ryzen 3000 CPU (excluding APU) as at this clock, every Ryzen 3000 should work with infinity fabric at 1:1 ratio. However, soon there will be a premiere of Ryzen 4000/5000 (however it will be called) processors and many users are preparing for an upgrade. Since I was able to get Ryzen 4650G then I will use it for most tests in this review.

Link to G.Skill product website <-click.

Product photos:



...
Some more product photos:

Spoiler

The tested memory kit has one XMP profile which is working perfectly fine on AMD B550 and X570 motherboards. I had no problem with the profile on Ryzen 9 3900X and Ryzen 5 4650G on ASUS Strix B550-I Gaming, Crosshair VIII Impact, ASRock B550 PG Velocita and ASRock B550 Extreme4 motherboards. Presented results are mostly made on the Strix B550-I Gaming as it overclocks the best.

XMP#1: DDR4-3600 CL14-15-15-35 1.45V



As you can see, there are no problems with stability so we can move to overclocking tests.

It was also expected to see Samsung B-die under the heatsinks as it's still the best memory for overclocking and competitive benchmarking.

 

[Woomack]

Overclocking Results

Here are two results made earlier on the 3900X and ASUS Strix B550-I Gaming and later I will show some results on the Ryzen 4650G and the same motherboard.

DDR4-3733 CL14-13-13-28 1.50V - stable


DDR4-4800 CL16-17-17-36 1.60V - some problems under longer full load, CL18-18-18 is fully stable

 

[Woomack]

Results with IF/mem ratio 1:1


DDR4-3600 CL14-15-15-35 1.45V (XMP)




DDR4-4000 CL14-15-15-35 1.55V




DDR4-4200 CL16-16-16-36 1.50V




DDR4-4400 CL16-17-17-38 1.55V




DDR4-4533 CL16-17-17-38 1.60V




Results with IF/mem ratio 1:2
Results more for bandwidth and latency comparison. For some reason I couldn't stabilize anything above DDR4-4666 while exactly the same memory kit was working fully stable at DDR4-4800 CL18-18-18 with the 3900X CPU.


DDR4-5000 CL18-18-18-38 1.65V




DDR4-5100 CL20-20-20-46 1.70V




DDR4-5200 CL20-20-20-46 1.75V




Some additional comments:
- Looks like at least my CPU requires SOC 1.35V+ for DDR4-4200+ with IF 2100MHz and tighter timings to pass more demanding tests and 1.4V at DDR4-4400/DDR4-4533.
- Memory latency is higher on Ryzen 4650G than on Ryzen 3900X, when I compare for example XMP DDR4-3600 then is about 7ns difference. It can be also related to automatic sub-timing settings on various CPUs.
- At DDR4-4200+ and 1:1 ratio, Ryzen 4650G doesn't like as tight timings as 3900X with 1:2 ratio.
- Samsung IC overclocks significantly higher on Ryzen 4650G. 3900X wouldn't let me to pass DDR4-4866 while on 4650G I could reach 5200. On the other hand, it's harder to stabilize tight timings but it can be a matter of my CPU.

Memory clock at about 4200 or 4400 seems optimal and I guess that most users will try to reach it once retail Ryzen 4000/5000 hit the stores. G.Skill has many memory kits rated at anything between DDR4-3600 and 5000 so anyone can pick something optimal. I think that finally more users will start to search for 4000+ memory kits which prices are going down.
Even though the Neo DDR4-3600 CL14 seems like not the highest frequency kit then it overclocks great and under the heatsinks we can expect only the best Samsung IC what guarantees a wide range of settings and the best performance on any modern platform.


I'm done for now but will work some more on additional settings. Feel free to comment or ask questions.

 

[Voodoo Rufus]

Where did you get the CPU from?

 

[Woomack]

Local online store and they took it from Polish distributor. They had 5 of them and I was surprised as all others in the EU sell them in bundles with motherboards and RAM or as a whole PC.
I won't hide I was expecting a bit different questions in this thread

 

[Voodoo Rufus]

Ok, back on topic then since I'm not in Poland. Those are some tight timings for 3600 ram, correct?

 

[Woomack]

Pretty much the tighest timings in a DDR4-3600 memory kit available in stores. It can also run tighter than in XMP but at safe/reasonable voltages not much more than CL14-13-13 or 14-14-14. For Ryzen 3000 it's about the best you can get out of the box. For Ryzen 4000/5000 better results will be at a higher memory clock but as you see, this memory can handle a lot. G.Skill also has DDR4-3800 CL14-16-16 1.50V kit but not every Ryzen 3000 will work at DDR4-3800. At least maybe 50% of what I tested couldn't run stable and I had to set DDR4-3733. That's why recommended maximum clock is DDR4-3600 but with Ryzen 4000/5000 it will be higher. I had no chance to test more than one chip so I can't tell what is the average max clock. We will find out in some weeks.

 

[Woomack]

Here is an additional stability test at DDR4-4533 CL16-17-17-38 1N, 1.55V with IF 2266MHz / 1:1, SOC 1.35V.
This setup doesn't like too low tRC and tRAS. On the other hand there is a significant difference in memory bandwidth and latency at tighter sub-timings. I can even say that we can achieve more because of that than because of tighter main timings. Going from auto ~700 tRFC to ~400 and tRC from 100+ to 72 gives about 2GB/s more and 3ns lower latency. Additional sub-timings let to reach 70GB/s memory bandwidth at DDR4-4533 what used to be possible almost only on quad-channel platforms (at least counting reasonable voltages and ambient temps).

 

[TerranBrackiatt]

quick Q:

did the XMP profile ALSO overclock the CPU in any way?

reason i ask:

on Nevermore, the default setting is 667 (DDR3 1333) for my g.skill RAM, with the 4930k running stock (3.4GHz/ 3.9GHz boost). but as soon as i click the XMP profile for 1200 (DDR3 2400) in the BIOS, the mobo ALSO overclocks the processor to all-core 3.9GHz (since this is the boost speed, i obviously get no more boost). I assumed this had to do with the memory controller on the CPU being rated for DDR3 1600, while using DDR3 2400 RAM. not complaining, but curious so i can plan for my next build.

you can see in my sig what i plan on upgrading soon, and if the XMP profile overclocks the processor, (i assume disabling the Boost * ), i just might select DDR4 3200 RAM instead of the DDR4 3600 or higher. i don't really want to mess with overclocking the Ryzen9 5900x (as i'm out of touch with the new CPUs and frankly would rather spend my free time gaming versus testing OC stability), but i still want the Boost stuff to work.

* i suppose this becomes another question,
~ are the XMP profile and CPU clocks totally separated? (i ask out of ignorance of the past 7 years of CPUs)
~ IF the XMP profile overclocks the ryzen 9 3900x, does it OC the same way my Rampage IV black did? (all core at the boost cap) or does it select some all core speed in between the default and boost speeds? or some other configuration?

PS: i've been meaning to ask this Q in its own thread for awhile, but wasn't sure where to ask it: Motherboards, RAM, CPU, or general hardware, as the Q encompasses all 3 parts, and how they interact together. Then i get distracted by work (18-wheeling.) My apologies if it's out of place here.

 

[Woomack]

On some motherboards when you manually change any option then it may switch to manual/auto OC setting. Usually it's optimized motherboard profile. However, in most cases XMP will only affect memory settings and may also affect additional voltages related to memory controller.

For Ryzen 3000 recommended is DDR4-3600. For Ryzen 5000 it will be probably anything between 3600 and 4000. DDR4-3600 costs about as much as DDR4-3200.

 

[S4boost]

I just purchased the 2x16 3600mhz cl14 kit, have you tested that kit yet? My goal is to get fclk to 2000mhz if my 5600x will handle it. Would you run 1.55 daily on a gaming build as shown in your 4000 cl14 test?