G.Skill Trident Z 16GB kit pairs DDR4-3600 speed with low latency

Not content with making ever-larger and ever-faster RAM kits, G.Skill is pushing speed and latency boundaries this time around. The company's latest 16GB Trident Z kit clocks in at 3600 MT/s with 15-15-15-35 timings, and it spreads its capacity across two 8GB DIMMs. This responsive kit turns in those numbers while running at 1.35V. We've built systems with Trident Z RAM in the past, and the two-tone aluminum heat spreaders on these DIMMs look pretty snazzy, too.

As DDR4-3600 goes, a two-DIMM 16GB kit is pretty unusual, and such a kit with 15-15-15-35 timings appears to be a first. A quick survey of Newegg suggests that similarly-specced kits with four 4GB DIMMs are most common right now. G.Skill and Corsair both make 16GB DDR4-3600 kits with 16-16-16-36 timings, though G.Skill offers a two-DIMM kit with those specs while Corsair's sticks with four.

Builders who need this Trident Z kit's uncommon combo of low latencies and high speeds will be able to pick one up at the end of April. G.Skill didn't announce pricing info, but we'd expect a hefty premium over the company's already-pricey 16GB DDR4-3600 kit with 16-16-16-36 timings. That kit sells for $154.99 on Newegg right now. Slightly more pedestrian 16GB DDR4-3200 kits with 15-15-15-35 timings sell for about $100.

Comments closed
    • ronch
    • 4 years ago

    Good to see G.Skill is still rockin’ hard these days. I got my pair of G.Skill DDR3-1866 DIMMs more than 3 years ago, and unlike some brands that by now would’ve been dead or not showing much signs of life, G.Skill is still on top of things.

    • kuttan
    • 4 years ago

    Which vendor G.Skill usually uses for their memory chip ICs ?

      • Krogoth
      • 4 years ago

      They use Micron or Samsung chips.

    • Entroper
    • 4 years ago

    I haven’t been able to get my Skylake build to POST at 3200 MHz speeds. I have that DDR4-3200 15-15-15-35 kit, works fine at 3000 MHz, but any faster and it just waits for a while and then reboots to safe mode at 2133 MHz.

      • CScottG
      • 4 years ago

      That’s likely an m-board (and possibly CPU) limitation.

      Ex. Asrock has some 4000 RAM spec.ed, but only tested with very few modules and only with the i7-“k” (Skylake).

      -you have to be really careful about what the m-board will support – and IMO that’s bordering on fraud to say something like DDR 3200-4000 compatible without an OBVIOUS disclosure as to what’s required to actually get there.

      • Krogoth
      • 4 years ago

      [b<]Motherboard[/b<] (memory tracing) or the memory controller on the CPU itself is what is holding your memory back. Running factory overclocked and overvolt memory is almost always a hit or miss affair. The memory, motherboard and memory controller have to play nice with each other.

    • brucethemoose
    • 4 years ago

    Following the rough Frequency/CAS Latency formula, this is the fastest DDR4 kit I’ve seen.

    • Visigoth
    • 4 years ago

    I don’t like the 1.35 volts. They should’ve stayed with 1.2 volts. Someday, maybe.

      • brucethemoose
      • 4 years ago

      If you want low access latencies on the current DRAM fabrication node, you have to up the voltage. No way around it.

        • yuhong
        • 4 years ago

        Obviously, but I do wonder how much better Samsung 4Gbit DDR4 E-die compares with the D-die for example.

    • southrncomfortjm
    • 4 years ago

    What practical effect would the slightly faster timing really have? Is there any conceivable way an end-user would actually feel their machine doing anything faster? Are there some kinds of work that would actually notice a difference (encoding maybe?)

    Or is this really just about it being a 2x8GB configuration rather than the slightly faster timings?

      • Freon
      • 4 years ago

      2x8GB is the standard for a 16GB kit.

      As for low timings and high speeds, I have a hard time finding much evidence it helps. It seems only Witcher 3 running 1080p with a 980 Ti really seems to care much beyond 2666mhz or so.

      2666mhz kits are approaching the <$70 range now. I’ve seen 3000mhz for under $80 on sale. None are particularly low latency, but again, unsure it matters much. +/- one or two CAS numbers isn’t actually a lot of linear time.

        • grazapin
        • 4 years ago

        I got a G.Skill kit 2x8GB DDR4 3000 15-16-16-35 for $68 after a discount at Newegg. Normal price is $75. That $155 kit seems like an insane price premium for the tiny, probably unnoticeable performance increase you might get.

        Spending that extra $80 on a better video card or CPU makes way more sense.

      • nanoflower
      • 4 years ago

      From what I’ve read the faster memory really tends to help things like APUs since they tend to be constantly moving memory around. For your average CPU and discrete graphics card I’m not sure you could notice the difference except in certain memory focused benchmarks.

      • ImSpartacus
      • 4 years ago

      If you’re on a classic cpu+ dgpu gaming machine, then you won’t notice any performance bump unless you have weird unrealistic bottlenecks (very weak cpu paired with very strong dgpu, etc).

        • jdevers
        • 4 years ago

        At the same time, if you are willing to spend that kind of premium on memory to make an APU work better you might want to reconsider the APU in the first place.

      • brucethemoose
      • 4 years ago

      Yes, it makes a difference:

      [url<]http://m.hardocp.com/article/2015/08/05/intel_skylake_core_i76700k_ipc_overclocking_review/6#.VwvY3oo8LCQ[/url<] Latency seems to be the key factor vs raw speed though. And you have to be in a scenario where you're actually CPU bottlenecked (which, as TR has shown, happens more often than you'd think during the slower frames.)

        • southrncomfortjm
        • 4 years ago

        I can see the memory speed+latency making a difference in those extreme tests where they are trying to bottleneck the CPU, but there are too many variables at play. The memory timings *and* speeds are different in those tests, so you can’t draw a conclusion that it is the memory timings that are making a difference.

          • brucethemoose
          • 4 years ago

          They list the timings on the 3rd page, and overall latency is going down as speed goes up.

          You’re right, no-one has tested the impact of bandwidth vs latency separately, as far as I know… maybe that’s something TR should do.

            • Krogoth
            • 4 years ago

            It has done on a number of types and various platforms. The only time were it had a meaningful impact was on budget platforms using a iGPU or ancient platforms (K7-Netburst and earlier). You are just better off allocating your memory budget towards having more memory capacity then getting “faster” memory.

            • brucethemoose
            • 4 years ago

            I’ve never seen anyone do a test on Skylake/Haswell-E where latency or frequency are kept constant, though.

            It’d be interesting if TR bought this kit and tested it from 2133-3600Mhz while keeping the timings constant (which isn’t constant latency, but still interesting), or tested from CAS 15-25 at a constant 3600Mhz.

      • UberGerbil
      • 4 years ago

      You might want to talk to some of the folks posting [url=https://techreport.com/forums/viewtopic.php?f=36&t=116660<]in this thread[/url<].

        • southrncomfortjm
        • 4 years ago

        Eh, not that big of a deal. Just curious. I’m perfectly fine with my DDR3-1600 with 8-8-8-10(?) timings. Just nice to have some perspective on these news articles that don’t provide too many details.

        • Krogoth
        • 4 years ago

        That’s mostly from bandwidth not so much from lower latency. The massive caches on modern CPUs hide most of the tangible benefits that would come from having a lower memory latency.

      • Krogoth
      • 4 years ago

      It is limited to certain real-world applications (ie not games). It is mostly for epenis-types.

Pin It on Pinterest

Share This