G.Skill’s DDR4-4400 kit seizes the four-module memory speed crown

Super-fast memory kits composed of two modules are all well and good, but some folks need large capacity to go with wickedly-high clock speeds. Last week Corsair claimed to have made the world's fastest four-DIMM 32 GB memory kit with its 4333 MT/s set. However, today G.Skill has trumped that figure with its DDR4-4000 four-module kit. The DIMMs' Samsung B-die chips are rated for CL19-19-19-39 timings and use Samsung B-die chips clocked at a sky-high 4400 MT/s.

The memory maker demonstrated the four-stick kit running in dual channel mode on an Asus ROG Maximus X Hero (Wi-Fi AC) motherboard and a speedy Intel Core i7-8700K processor. This kit pushes the limits of XMP certification with a 1.5 V input voltage, though it supports XMP 2.0 to hopefully ease memory overclocking.

G.Skill didn't provide any pricing or availability information, but the fastest memory always costs a pretty penny. We imagine kits will hit stores in the coming weeks. The company offers lifetime warranty on all its memory modules.

Comments closed
    • torquer
    • 2 years ago

    Is there really any point? Outside memory bandwidth benchmarks I can’t think of anything that gives much of a crap about memory speed anymore

      • tacitust
      • 2 years ago

      There were some benchmarking videos made a few months back that showed faster memory being a significant help to video games when being played on an Intel i3-6100 processor with a mid-range video card.

      Of course, at that level, just upgrading to an i5 or i7 would give you a better bang for your buck anyway, and improvements when boosting the memory at the high-end would be much smaller.

      • Firestarter
      • 2 years ago

      [url<]https://techreport.com/review/31179/intel-core-i7-7700k-kaby-lake-cpu-reviewed/11[/url<]

        • Chrispy_
        • 2 years ago

        Based on the kits used, that result nicely confirms that true-latency is the key metric once you have [b<]enough[/b<] bandwidth. 3866 CL18 = 4.65ns true latency 3000 CL15 = 5.00ns true latency so the exotic 3866 kit was actually 7% faster and turned in scores that were 7% higher. I would expect the much cheaper 3466 CL16 kits to actually be even quicker in ARMA 3, because that kit has a true latency of just 4.61ns. Here's [url=http://www.overclock.net/content/type/61/id/2601678/flags/LL<]yet more proof that bandwidth and frequency are irrelevant[/url<], and that these high frequencies are performing better because they also happen to have lower true latency (source, [url<]http://www.hardware.fr/articles/940-5/cpu-ddr4-vs-ddr3-pratique.html).[/url<] Notice how 2400 CL11 is faster than 3000 CL16. When looking at different RAM, just compare CAS latency / clock speed and whichever is lowest is fastest.

          • Firestarter
          • 2 years ago

          G.Skill has a 2x8GB kit, 3600/CL15 that is very low latency and not too expensive from what I’ve found. Model name is G.Skill Trident Z F4-3600C15D-16GTZ: [url<]https://www.newegg.com/Product/Product.aspx?Item=N82E16820232306[/url<]

          • Freon
          • 2 years ago

          That’s just a hypotheses that needs to be tested. I.e. just bump up the latency of the 3866 to CL19 or CL20 and see what happens.

          Even then, It’s still just one game, and the times we do see speed vs. latency benchmarks it seems like a crapshoot either way and may still be application specific.

          I think you’re extrapolating too far with very limited data.

            • Chrispy_
            • 2 years ago

            That one game is the single gold standard for “most bandwidth hungry game there is”

            If the benefits of higher bandwidth RAM are minimal in that, you can all but guarantee the differences vanish elsewhere.

        • Freon
        • 2 years ago

        ARMA 3 seems to be the exception to the rule, a very poorly written engine. Same goes for H1Z1 which seems to show similar scaling with single threaded performance.

          • travbrad
          • 2 years ago

          Planetside 2 (same engine as H1Z1) shows some improvements with faster memory too, although it’s still more dependent on IPC/clockspeed than memory.

          These games are exceptions though as you said, and even with those games there may be a point of diminishing returns with memory beyond a certain performance level.

      • Chrispy_
      • 2 years ago

      If you find an application that can use more bandwidth (there are a few games like ARMA 3 and Ryse: Son of Rome, to name a couple) then more bandwidth can provide significant improvements, but only if it doesn’t come at the cost of true latency.

      There’s an overwhelming amount of evidence that true latency (clockspeed/CAS latency) is the single most important factor – to the point that even in games like ARMA 3 and Ryse, if the latency is too high, the frequency of the RAM doesn’t matter.

      What these games show is that they can use more bandwidth than 2133 and 2400 can provide. Diminishing returns have already kicked in with 2666, offering 95% of the performance of 4000+ kits, at less than half the price, usually. If you’re buying 2666 CL15 which is super affordable (about 10% more expensive than the bargain basement 2133 RAM) then you’re being smart by having enough bandwidth, but still focusing on performance/$ and lower true latencies.

      If you can find cheap 2800, 3000, or 3200 RAM with tight timings – then that too is a smart buy because more bandwidth can’t hurt, so long as the true latency isn’t higher.

      The problem comes when you spend lots of money on faster RAM:
      [list<][*<]$160 buys you 16GB of 3000 CL15 which has [b<]enough[/b<] bandwidth. [/*<][*<]$365 buys you 16GB of 4400 CL19 which has [b<]more[/b<] bandwidth.[/*<][/list<] The thing is, both of those kits have the same true latency and unlike DDR4 2133, the 3000 kit doesn't leave much (if any) performance on the table. The $200+ price difference between those two kits gets you an 8700K instead of an i3, or a GTX 1080 instead of a 1060. Don't be stupid, and don't buy [i<]expensive[/i<] RAM.

    • Shobai
    • 2 years ago

    First para: sentence beginning with ‘However, today G.Skill…’ has the wrong speed listed. The following sentence should probably be reworked to remove the second usage of ‘Samsung B-die chips’.

    • Chrispy_
    • 2 years ago

    LOL, CL 19? My mother slow cooks a hog faster than that….

    True latency = CAS latency / clock speed, and all these high-frequency modules have such terrible timings that their true latency is barely faster than affordable DDR4 3200.

    Hell, the fastest RAM you can probably buy for real-world use is [url=https://www.newegg.com/Product/ProductList.aspx?Submit=ENE&N=100007611%20601190328%20600528730%20601203950&IsNodeId=1&bop=And&Order=PRICE&PageSize=36<]DDR4 3466 at CL16[/url<]. For a start, your processors IMC has a hope in hell of actually driving the RAM at those speeds, and secondly, it's barely 50% more costly than the cheapest RAM you can buy. At CL16 it's going to be equivalent to 4133 CL19 but unlike those high speed kits it will run at 1.2V and not require you to dial back your K-series overclock to reach it. Edit: See [url=https://www.techpowerup.com/reviews/Intel/Core_i7_8700K_Coffee_Lake_Memory_Performance_Benchmark_Analysis/9.html<]this Coffee-Lake memory-scaling article[/url<] by TPU? Yeah. That DDR4-4000 is never in the lead, and in every graph, even when the 2133 and 2400 stuff falls away, the 3000 CL14 is right there next to the 4000 CL19. I know it's not a perfect analogy, but true latency is like a car's acceleration and frequency is like the top speed. Most applications are tracks with lots of curves and acceleration is by far the most important aspect. Occasionally it won't be a curvy track, it'll be a five mile straight - and this vanishingly rare scenario, the DDR4 4600 is going to win ONE RACE, assuming the track is long enough to let it catch up with the lower latency RAM that was faster off the line, through the first quarter-mile, and still in the lead at the two-mile mark.

      • Firestarter
      • 2 years ago

      60 ns EDO RAM has the real latency right in the description, it would be nice if RAM modules got advertised like that today

    • RickyTick
    • 2 years ago

    Maybe I’m misreading or misunderstanding ram speeds, but G.Skill has DDR4 4600 ram available at NewEgg now. Isn’t this “faster” than DDR4-4000?
    [url<]https://www.newegg.com/Product/Product.aspx?Item=N82E16820232629[/url<]

      • takeasneededforpain
      • 2 years ago

      I think they are referring to 4-stick kits…

        • RickyTick
        • 2 years ago

        Oh, I’m sure you’re right. Thanks

        • freebird
        • 2 years ago

        Yes, this article is for 4x8GB = 32GB RickyTick’s link is to 2x8GB= 16GB. I really don’t understand this “speed” race with the higher latency needed to achieve it. My 4x16GB for 64GB of 3000Mhz with CL14-14-14-32 is plenty fast enough and could probably be overclocked much higher with increased latency settings, had it running at 3200Mhz with the same settings for awhile, but sometimes my Ryzen 1700/X370 mobo wouldn’t reboot and it would reset to default 2133. Just glad I bought my 64GB back in March before memory prices accelerated past the insane GPU prices.

          • derFunkenstein
          • 2 years ago

          Bigger numbers are always better, CL included amirite? #eyebrowWiggle

          • Firestarter
          • 2 years ago

          that DDR4 4400 kit has a lower real latency than your DDR4 3000 kit and higher bandwidth too. Assuming you get both to run at their advertised speeds, the DDR4 4400 kit will be faster in every situation. Whether that is worth the extra money is not for me to judge until I’m buying it myself

Pin It on Pinterest

Share This