G.SKILL Trident Z Neo DDR4-3600 MHz CL16 2x8 GB

[Black Haru]

With a market that is flooded with "Ryzen Optimized" memory kits, it can be hard for a particular kit to stand out. The G.SKILL Trident Z Neo does just that. With a striking new take on the award winning Trident Z design and specifications that are as good as it gets for Ryzen, the Trident Z Neo just might be the ultimate Ryzen memory kit.

Show full review

 

[XL-R8R]

The IC's used in these sticks are seemingly not so new:

It seems the stock rating from Samsung is 2133 @ 1.2v 15-15-15 and can also be found in this QVL document from Asus:

https://dlcdnets.asus.com/pub/ASUS/mb/LGA1151/ROG_STRIX-B360-I_GAMING/ROG-STRIX-B360-I-GAMING-Memory-Report.pdf

Searching "K4A8G085WBBCPB" will get you the desired results in this list but none of the items listed have the internal SEC code of '934' - the newest you can find are entries starting with 'SEC 7xx' - that being said, I expect the internal '934' coding is a revision of older IC's with basically a reflash for 1.35v and 3200-3600MHz.

 

[Xuper]

So Intel can't run at 1T ? any idea why?
Memory write is quite awful on AMD.almost 1/2 intel.so Next Ryzen 4000 could improve this.

 

[billeman]

Memory write is quite awful on AMD.almost 1/2 intel.so Next Ryzen 4000 could improve this.

That's the problem I have with this review. The CPU used in question is only a single-CCD cpu. The dual-CCD (3900x/3950x) processor have about double the write speed and better read speed. I don't see the point of doing a dram review and not using the best Ryzen processor for it. Should I dare to say that the comparison with intel is very skewed ?

 

[Calmmo]

So Intel can't run at 1T ? any idea why?
Memory write is quite awful on AMD.almost 1/2 intel.so Next Ryzen 4000 could improve this.

32gb neo bdie on a 2 cdd cpu

 

[Oberon]

I'm more interested in the Hynix 3600 CL16 variants of these kits. Paying almost 50% more for B-Die is crazy to me given the small benefits.

Memory write is quite awful on AMD.almost 1/2 intel.so Next Ryzen 4000 could improve this.

This was an intentional design choice by AMD and doesn't appear to have any meaningful impact on the vast majority of non-synthetic workloads. As other posters have mentioned though, any of the dual-CCD CPUs (3900X and 3950X) have "full" memory write bandwidth if it's something important to your particular use case.

 

[SQr]

I'm more interested in the Hynix 3600 CL16 variants of these kits. Paying almost 50% more for B-Die is crazy to me given the small benefits.

I paid 159USD for a 32GB 3600C16 Trident Z Neo kit with Hynix DJR. Performance is about the same as this at stock. Probably won't overlock as well though, but I think I got a good deal...

 

[TheMadDutchDude]

So Intel can't run at 1T ? any idea why?
Memory write is quite awful on AMD.almost 1/2 intel.so Next Ryzen 4000 could improve this.

It's a motherboard limitation, mostly due to trace distance from RAM to CPU. Four DIMM'd boards struggle with 4K and beyond at 1T, but 2 DIMM'd boards, designed for OC, can do it no problem.

 

[Non_NPC]

32gb neo bdie on a 2 cdd cpu

This is my Gskill RipJaws 32Gb BDie kit...

 

[JcRabbit]

I paid 159USD for a 32GB 3600C16 Trident Z Neo kit with Hynix DJR. Performance is about the same as this at stock. Probably won't overlock as well though, but I think I got a good deal...

I got the 64GB 3600 C16-19-19-39 Trident Z Neo kit of Hynix DJR for my Intel 9900K system. Currently running it 100% stable at 3800 Mhz C16-20-20-40, 1.4v.

Tried 3800Mhz with 1.40v keeping the C16-19-19-39 timings and although it booted into Windows fine, it would fail on tests. Increased to 1.45v, same result, likely because (from what I read) tRP does not scale with voltage, so it must also be increased (to 20) when you increase the frequency (to 3800).

As for tRAS, I tried setting it at 38 and to my surprised performance apparently DECREASED a bit in relation to the previous value of 40, so I set it back to 40. I also read somewhere that if tRAS is too low on Asus boards the board/memory controller will silently and internally override it with its own minimum value (which would explain the slight decrease in performance in relation to the previous 40 value but without any memory errors).

I also tried 4000Mhz at 18-22-22-42 with 1.45v (hey, you never know when you get lucky lol) but the board wouldn't even post (well, actually it did, it just kept trying to retrain the memory until it gave up with an error and post in safe mode). Since this is a production system, I then decided to call it a day and run the Neo kit stable at 3800 Mhz.

 

[billeman]

This is my Gskill RipJaws 32Gb BDie kit...View attachment 141601

Those are mighty impressive results Dual-rank FTW

 

[gamefoo21]

Sigh... Bunch of noobs.

AMD cripples the 3800x and lower by cutting the ram write bandwidth to 50%.

It's built into the CPUs and how the memory controllers are binned.

 

[EarthDog]

Sigh... Bunch of noobs.

AMD cripples the 3800x and lower by cutting the ram write bandwidth to 50%.

It's built into the CPUs and how the memory controllers are binned.

Or, you know, it's just how the arch was designed (it has less CCXs, 1, and therefore 1/2 the write bandwidth). It has nothing to do with binning.

 

[TheMadDutchDude]

Sigh... Bunch of noobs.

AMD cripples the 3800x and lower by cutting the ram write bandwidth to 50%.

It's built into the CPUs and how the memory controllers are binned.

*sighs harder...*

Or, you know, as above, it is in the design where it only has one CCX instead of two...

Maybe you should check your facts before posting such a message.

Write bandwidth very little to do with the 'real-world' performance for 99% of the user base.

 

[gamefoo21]

I

*sighs harder...*

Or, you know, as above, it is in the design where it only has one CCX instead of two...

Maybe you should check your facts before posting such a message.

Write bandwidth very little to do with the 'real-world' performance for 99% of the user base.

*Sighs hardest*

So why does AMD gimp the memory controller on the single chips? Your telling me that AMD has magic silicon that can read at a full 128bit bus width but only writes at 64, and adding a second chiplets suddenly awakens the extra write pins?

AMD specifically chose this path because their memory controller is still crap. They made the trade off because gamers will buy the 3800 and lower, bundling it with fast ram, professionals will buy the 3900 and up and use slow ram...

It's a compensation method, to make up for a memory controller that still can't officially break 3000mts when all 4 dimms are populated.

Sigh...

Trust me, I know all too well about Ryzen 3000s memory controller design issues. It also sits on the IO die, and it's why it's a latency monster.

So when you bench a 3800, 3700, 3600 etc... It'll get hammered on write bandwidth and there's nothing the end user can do.

I was specifically responding to the whiners who didn't seem to understand that AMD cripples the memory controller part of the IO chiplets.

Personally I'm excited to see how the Zen 2 APUs do, unless AMD artificially cripples the desktop ones, I won't be surprised if the 8/16 versions end up faster than the 3800/3700, since that IO die will be part 9f the main die, and there will be serious latency benefits to it...

But oh wait monolithic designs are bad right?

 

[EarthDog]

*yaaaaaaaaawn...

It has nothing to do with binning.

*rolls over, never to be bothered again...

 

[TheMadDutchDude]

Just let the man exercise his right to his free speech, even though he doesn't understand how the CCX system works. I'm out.

Hmm, one final thing, then. Their memory controller is crap, right? Yet as of right now it has the highest ever obtained frequency for DDR4. Yah, it's a turd.

OGS`s Memory Frequency score: 3027.2 MHz with a DDR4 SDRAM

The DDR4 SDRAM @ 3027.2MHzscores getScoreFormatted in the Memory Frequency benchmark. OGSranks #1 worldwide and #1 in the hardware class. Find out more at HWBOT.

hwbot.org

You still get the benefit from faster RAM on the 3900X and up. Please, just stop while your head is still above the ground. You're just digging yourself a hole, fella.

It's not nerfed, it's not binned, it is not binned better or worse... as a matter of fact, it's none of what you are claiming.

Peace. Have fun in your made-up fairytale.

 

[gamefoo21]

Just let the man exercise his right to his free speech, even though he doesn't understand how the CCX system works. I'm out.

Hmm, one final thing, then. Their memory controller is crap, right? Yet as of right now it has the highest ever obtained frequency for DDR4. Yah, it's a turd.

OGS`s Memory Frequency score: 3027.2 MHz with a DDR4 SDRAM

The DDR4 SDRAM @ 3027.2MHzscores getScoreFormatted in the Memory Frequency benchmark. OGSranks #1 worldwide and #1 in the hardware class. Find out more at HWBOT.

hwbot.org

You still get the benefit from faster RAM on the 3900X and up. Please, just stop while your head is still above the ground. You're just digging yourself a hole, fella.

It's not nerfed, it's not binned, it is not binned better or worse... as a matter of fact, it's none of what you are claiming.

Peace. Have fun in your made-up fairytale.

Ooo a single Dimm on super cooling with Micron that's a timing and clock slut(loose timings with lots of clocks) memory that's probably golden binned.

The latency on the AMD setup is garbage. Threadripper is even worse.

Oh who has the fastest memory controller:

Splave`s MaxxMem Read Bandwidth score: 47137 MByte/sec. with a DDR4 SDRAM

The DDR4 SDRAMscores getScoreFormatted in the MaxxMem Read Bandwidth benchmark. Splaveranks #1 worldwide and #1 in the hardware class. Find out more at HWBOT.

hwbot.org

.

Hot damn on a modded Z170 board too. He's ranked #1, AMD guy is #2...

Oh what about a different bench:

FUGGER`s AIDA64 - MemoryRead(alpha) score: 185079 MB/s with a DDR4 SDRAM

The DDR4 SDRAMscores getScoreFormatted in the AIDA64 - MemoryRead(alpha) benchmark. FUGGERranks #null worldwide and #null in the hardware class. Find out more at HWBOT.

hwbot.org

Intel again...

Oh look a 7700K...

Luumi`s MaxxMem score: 5339 marks with a DDR4 SDRAM

The DDR4 SDRAM @ 4214MHzscores getScoreFormatted in the MaxxMem benchmark. Luumiranks #3 worldwide and #3 in the hardware class. Find out more at HWBOT.

hwbot.org

So it's become the FX of memory controllers, lots of Hz but it lacks the performance to match.

For real world, Intel's memory controller doesn't shit the bed on 4 dimms. What's the spec for Ryzen again? 3200 for 2 dimms, 2,666 for 4 Dimms? It still can't handle 4 dimms, and it's been a weakness since AMD initially designed their IMC back for the Athlon 64.

Yes because AMD has never said they didn't choose to limit the write performance.

Either way it's a design limitation that appeared on the chiplet design and wasn't there on the monolithic Zen+ afaik.

Yeah, I have never been a fan of a the single chiplet 3000 series because they scream...

"All of the latency but half of the bandwidth."

But I don't understand how the CCX design works... Oh wait... The memory controller is part of the I/O die, it's only tied to the chiplets through the IF... So the IF is Half Duplex? It can only send data at half rate, so it needs to bounce write data from one core over the second CCX to get full speed. That just makes for worse latency... Unless memory benches became multi-core suddenly...

So how does core 1 in a 2 ccx setup get full memory write speed, while core 1 in a single CCX get half the speed? Unless AMD nerfs the memory controller for max speed... Oh look at that clock champ up there it's a 3600x...

The memory controller is part of the I/O die:

AMD has an awesome CPU design series that's let down by their interconnect tech, a memory controller that annoys me and I really hope when they shrink it to 7nm it won't be so horrible at many dimms.

I apologize for wanting fast in all of my tasks, and I really do want a 3900 or 3950x to replace my Intel rig.

I'm a goon who rocks a repaired V2 because it renders and does work faster than a 2080ti but it games like a hopped up 5700xt.

But keep on hating... ;-)