G.Skill Trident Z 3200 MHz C16 DDR4 (2x 8 GB)

[cadaveca]

With DDR4 prices taking a dive recently, there's no time like the present to get yourself a new Z170 system, or perhaps you want to move from 8 GB to 16 GB. G.Skill's Trident Z kits and their sweet new looks and affordable pricing just might make you click that buy button.

Show full review

 

[the54thvoid]

Weird - Forum entry ^^ for 29th Sept but review dated 10th October.

Those things so fast they send you back in time Dave?

I keep virtual shopping for a new set up and I haven't seen any of these DDR Tridents anywhere in the UK from the major e-tailers. If I see them and another Skylake price offer, I might just make the jump.

 

[cadaveca]

Weird - Forum entry ^^ for 29th Sept but review dated 10th October.

Those things so fast they send you back in time Dave?

I keep virtual shopping for a new set up and I haven't seen any of these DDR Tridents anywhere in the UK from the major e-tailers. If I see them and another Skylake price offer, I might just make the jump.

LoL. That's when the post with the review was made. You'll find the same in many reviews, if they aren't copied into a new thread before posting to the front page.

These sticks are up on Newegg, as are other models that have been available for some time now. I am sure they will be in EU soon.

 

[yagga]

Hey cadaveca, can you please run some SSD style multitasking torture test on these things (eg anand's destroyer)?

Why does no one test to see if these insane RAM modules perform faster when they're getting hammered with an avalanche of requests from all angles? I do a lot of work with multiple VMs running and would like to know if there's a benefit to upgrading, but the available data has little to say about that.

 

[cadaveca]

Hey cadaveca, can you please run some SSD style multitasking torture test on these things (eg anand's destroyer)?

Why does no one test to see if these insane RAM modules perform faster when they're getting hammered with an avalanche of requests from all angles? I do a lot of work with multiple VMs running and would like to know if there's a benefit to upgrading, but the available data has little to say about that.

You need to be a bit more specific in what you are looking for. Anandtech's SSD testing is not testing memory. There is no currently existing application with usable output for reviews for this other than bandwidth/latency, and that is why nobody does such testing. There needs to be some sort of valued output that can be graphed for comparison.

 

[yagga]

Thanks for the reply.

Anand's benchmarks run application traces, correct? To test each SSD they just swap each one out (keeping the underlying system the same), then they record the data rate and how long it takes the test to complete. The faster both those metrics, the faster the SSD.

What i would like to see is the same type of testing, except instead of swapping out the SSD (or whatever storage you're using) you keep that the same (secure erase it first to try to keep a reliable baseline), swap out the memories instead, and run the test on those. The faster the traces finish the faster the memory.

Another idea would be to setup a ramdrive and run the traces from that.

Do both and we'll really be getting somewhere.

Does this make sense?

 

[cadaveca]

It does, but I need a benchmark I can start, walk away from, and come back and it has a score, or time. I do not have the time to be sitting monitoring a system for every second while it is in testing... doing reviews isn't a job that pays me by the hour. It's just a hobby that I do when my family doesn't require my time. So the testing that I do needs to be fairly autonomous, and needs to eb something that normal people can replicate (in other words, I'm not about to use custom software, since end users won't run custom software).

I use the exact same system, just swap out the ram sticks. I pop in a set, set XMP, test stability a bit, and then I run my benchmarks. Then I will push OC for a bit, then I re-run my benchmarks. If it was a matter of downloading and running a benchmark, I'm all for it. I run about 20 benchmarks over and above what I report in my review already, tracking the usability of the performance metrics to relate performance differences, so adding in another for a bit, testing, and then validating the results is no big deal for sure.

Does that make sense?




Ultimately, I'd love to be able to provide you the content you are requesting. What I really need is a link to a software to run. IDeas about how it can be done aren't adequate for me, unfortunately (and obviously nobody else either, or you'd not be asking me for this, right? )

 

[yagga]

Makes perfect sense. I've got no info on how Anand's tests are run, can you get in contact with them through back tech journalism channels and ask for the secret sauce? Or at the very least tell them how awesome their testing suite make a mention of how rad it would be if they would use it for RAM as well? Gotta get this party started, it's loooooong overdue... And maybe we can finally shut the jorge's of the world up if this comes back with the positive results i'm expecting.

If we can get this going and clear performance gains present themselves in the test, i'll be the first inline to upgrade with, i'm sure, many more behind me. Get on that horn man and start calling around, we'll change the world and you'll be the man who got it done.

Capiche?

 

[cadaveca]

Sure, but I think you missed the bit about that Anandtech doesn't offer what you are looking for either, so... you'd think they'd have done some testing and found it pointless.

 

[W1zzard]

Anand is using disk traces. They record timestamp, address and size of each disk access and then replay those without the actual application running, as quickly as possible, ignoring the timestamps. I think you misunderstood this as "a method that replays certain tasks in the actual application".
Anand's approach creates a pure disk load without any processing time for the data, not useful for testing anything other than disk io patterns.

and obviously real workloads don't scale that way, for each access in real life there is some processing time involved, which means if your storage gets 100x faster, the time you wait in your application doesn't change nearly that much because it performs other tasks too.

 

[yagga]

Thanks for clarifying W1zzard. What's your opinion of a memory test like this, do you think it can be done, would it be worth it, and if so can you think of a way to do it?

 

[Luka KLLP]

Pardon my language, but those sticks are sexy as f*ck

Call me cliche, but I still love the black-red theme. Great review!

 

[W1zzard]

Thanks for clarifying W1zzard. What's your opinion of a memory test like this, do you think it can be done, would it be worth it, and if so can you think of a way to do it?

My ssd bench uses actual real applications, doing real work, i'll do a few runs with different memory speeds, but i'm afraid the difference is gonna be very small, maybe even smaller than randon variations between tests

 

[yagga]

Thank you sir!

 

[Woomack]

I have about the same OC results on my kit but I had to update BIOS in M8 Hero to make it. On older BIOS I could barely boot at 3333 regardless of voltage or timings. On new BIOS ( 1001 ) memory is stable @3333 16-16-16 1.37V ( at 1.35V was crashing after about 30min load ). I will make some more tests and review should be ready soon.

 

[buildzoid]

So what would be the maximum safe voltage to run DDR4 at? 1.4V?

 

[cadaveca]

So what would be the maximum safe voltage to run DDR4 at? 1.4V?

...yikes. I do not feel confident answering this question. There are different CPUs and different ICs and each may have different levels of tolerance. The only appropriate answer is the specified voltage.

G.Skill has some Trident Z kits with "varied voltage", listing 1.4V as the max there. Since there is a chance that there are different ICs on different sets, we can't just say that 1.4V is OK, but I do feel it should be relatively safe (if we go by the whole stock + 10% rule).

At the same time, I do know of users pushing 1.5V 24/7, but as it has not been very long with DDR4 on the market, it is hard to say how this may affect the lifespan on these ICs. What I can say is that Hynix likes tight timings and Samsung likes high speed, as we are used to form DDR3.

 

[buildzoid]

...yikes. I do not feel confident answering this question. There are different CPUs and different ICs and each may have different levels of tolerance. The only appropriate answer is the specified voltage.

G.Skill has some Trident Z kits with "varied voltage", listing 1.4V as the max there. Since there is a chance that there are different ICs on different sets, we can't just say that 1.4V is OK, but I do feel it should be relatively safe (if we go by the whole stock + 10% rule).

At the same time, I do know of users pushing 1.5V 24/7, but as it has not been very long with DDR4 on the market, it is hard to say how this may affect the lifespan on these ICs. What I can say is that Hynix likes tight timings and Samsung likes high speed, as we are used to form DDR3.

Well I was more concerned for the CPU IMC. I have a 750K with a half dead IMC(one channel doesn't work) and the IMC on my 3960X seems to be giving up. I few dead sticks of DDR4 are nothing compared to a 5960X with no memory controller.

 

[cadaveca]

Well I was more concerned for the CPU IMC. I have a 750K with a half dead IMC(one channel doesn't work) and the IMC on my 3960X seems to be giving up. I few dead sticks of DDR4 are nothing compared to a 5960X with no memory controller.

Well, I do remember hearing that 1.5V was the max for XMP 2.0. Also, Skylake supports both DDR3L and DDR4, so I do feel kind of safe with Skylake @ 1.5V because of that DDR3 bit, but with Haswell-E, I feel as you do. Since added voltage doesn't really seem to help frequency, and only really affects timings, I do question the need to overvolt memory on X99 unless benching.

 

[buildzoid]

Well, I do remember hearing that 1.5V was the max for XMP 2.0. Also, Skylake supports both DDR3L and DDR4, so I do feel kind of safe with Skylake @ 1.5V because of that DDR3 bit, but with Haswell-E, I feel as you do. Since added voltage doesn't really seem to help frequency, and only really affects timings, I do question the need to overvolt memory on X99 unless benching.

Yeah I'm going to be benching pretty much everything using the 5960X so being able to run 3000 with tight timings is the goal.

 

[cadaveca]

Well, you know the risks. So I feel OK telling you that you'll see the best results well over 1.5V, in the 1.65V+ territory.

 

[Woomack]

MSI Z170I Gaming Pro AC likes this memory more than ASUS Maximus VIII Hero. ~1.45V for 3600 but 1.4V+ is required to pass ~3400. Still great result for 8GB modules.

 

[eVampire]

Hello everyone, I'm a bit confused about the Test System Setup, it is indicated that the motherboard used is an Asus Z170-Deluxe but the photos posted under Installation & Setup was a Gigabyte GA-Z170X-Gaming G1. I think this was just a minor error. By the way this is a great review. I plan on getting one of these in my upcoming upgrade.

 

[cadaveca]

Hello everyone, I'm a bit confused about the Test System Setup, it is indicated that the motherboard used is an Asus Z170-Deluxe but the photos posted under Installation & Setup was a Gigabyte GA-Z170X-Gaming G1. I think this was just a minor error. By the way this is a great review. I plan on getting one of these in my upcoming upgrade.

Yeah, I used the ASUS board for testing but Gigabyte for pictures. You can see in the CPU-Z that ASUS was used. Benefit of also doing board reviews is that I have many boards; it just so happened I was also working on that Gigabyte board's review, and liked how the sticks look in it.

I've swapped the Z170 DELUXE out for a MAXIMUS VIII EXTREME at this point. expect a review on that board and more memory reviews soon!

 

[Steevo]

Have any way of performing a DDR3 VS DDR4 comparison?