G.Skill has been around for 25 years now, and since their inception have brought performance memory products to the PC enthusiast crowd. With the release of the Intel X99 chipset, and the accompanying need for DDR4 memory, G.Skill has their initial product offerings available in the form of their Ripjaws 4 series. They sent along their DDR4 3000 MHz 4X4 GB kit for us to have a look at today, which just happens to be the same kit we used for the Intel Haswell-E i7 5960X review. We didn’t do much with this memory past setting it to its stock configuration during the i7 5960X review, because we wanted to explore it more in depth here. So, let’s dive in and see what G.Skill’s entry into the world of DDR4 has to offer.
There isn’t a whole lot of information available at the G.Skill website on this kit, but here is what I was able to piece together for specifications.
G.Skill Ripjaws 4 DDR4-3000 4X4GB Specifications
|DDR4 – 288 Pin – Quad Channel Kit|
|3000 MHz (PC24000)|
|16 GB Total (4X4 GB Modules)|
|XMP||Compatible with XMP 2.0|
|Available in Red/Black/Blue|
Once the memory is plugged into the system, here is what the BIOS and CPU-Z show. Both the BIOS and CPU-Z show Max Bandwidth being 2133 MHz, but the CPU-Z memory tab shows the memory running at its rated speed and timings. It’s probably pulling that information from the JDEC standard of 2133 MHz for DDR4 memory.
The Ripjaws 4 series of memory will be available in a wide variety of speeds and capacities. Beginning at speeds of 2133 MHz and going as high as 3200 MHz, the kits are available in 16, 32, and 64 GB configurations. The 2800 MHz kits and below will operate at 1.2 V, while the higher speed kits will require 1.35 V. You can also choose from three available colors – red, black, or blue.
G.Skill has all new redesigned heatspreaders for the Ripjaws 4 modules, but they still kept the overall height at 40 mm to ensure maximum compatibility with CPU heatsinks. You’ll also notice by looking at the image below the PCB is black. Nothing kills the aesthetics of a nice looking memory module quicker than a green PCB, but that won’t be a problem here!
The Ripjaws 4 kit we are reviewing today comes packaged in a clear plastic clam-shell container. Inside, you’ll find the four memory modules and a nifty little case badge. The cardboard insert is where all the packaging’s graphics are printed and includes a marketing write-up about a few of the features.
As you can see by the above slideshow, the kit we have today is the red heatspreader version. G.Skill’s enthusiast memory has always been attractive looking, and that doesn’t change with the Ripjaws 4 series. Both sides of the modules have a sharp looking emblem, but one side has a sticker that covers most of the emblem. The sticker has information that’s nice to have handy when you’re setting timings and voltages manually. The heatspreaders and modules exude quality and feel sturdy and well-made when handling them. Here are several pictures to enjoy before we dive in to the testing phase of the review… Enjoy!
Testing for Stability at Rated Speed/Timings
The first thing we do is check for stability at the kit’s rated speed/timings. For this task, we call upon our UltraX R.S.T. Pro USB RAM tester, which was kindly provided by UltraX. The R.S.T. Pro is by far the most demanding way to test memory there is. In fact, it’s so tough on memory that many manufacturers use it to keep an eye on quality control during production runs. Depending on the capacity of the kit being tested, it can take a long time to complete the five loop testing we do. In the case of this kit, it took over seven hours to complete. After relentlessly beating on the G.Skill Ripjaws 4 for over seven hours, the R.S.T. Pro came up clean with no errors recorded. Below is a picture of the R.S.T. Pro we use and a screenshot showing it well into the sixth loop.
For in-OS testing, we use HyperPi and set it to use all 16 threads. HyperPi is a darn good stability test in its own right and will normally fail if either the CPU or memory isn’t 100% stable. If you don’t own a R.S.T. Pro, it’s probably the next best thing for testing memory stability. After passing the R.S.T Pro testing, we didn’t expect any stability problems with HyperPi and none were encountered. I think we can safely say the Ripjaws 4 kit is rock stable at advertised speed/timings/voltage.
Because we don’t yet have any other DDR4 kits to compare, all we can really do is toss some DDR3 kits (dual and quad-channel) in the mix to get an idea of how those platforms stack up. So, to that end, we have three different test beds. All the DDR3 dual-channel kits were tested on the same exact system. The DDR3 quad-channel DDR4 quad-channel kits were tested on a different system obviously.
|DDR3 Dual-Channel||DDR3 Quad Channel||DDR4 Quad Channel|
|Motherboard||ASUS Maximus VI Formula Intel Z87 Chipset||EVGA X79 Dark|
Intel X79 Chipset
Intel X99 Chipset
|CPU||i7 4770K Haswell||i7 4960X Ivy Bridge-E||i7 5960X Haswell-E|
|Memory||Various – See Below Comparison Kits||G.Skill TridentX DDR3-2400 MHz 4X4 GB 10-12-12-31||G.Skill Ripjaws4 DDR4-3000 MHz 4X4 GB 15-15-15-35|
|Video Card||EVGA GTX 780 Ti Classified||ASUS HD7970 DCUII TOP||EVGA GTX 780 Ti Classified|
|OS||Windows 7 X64||Win 7 X64||Win7 X64|
|Cooling||Custom Water||Custom Water||Custom Water|
Below is the list of the memory kits used for comparison today.
|Kingston||HyperX Beast||DDR3 2400||Dual||4×4 GB||11-13-13-30-2T||1.65 V|
|G.SKill||TridentX||DDR3 2666||Dual||2×4 GB||11-13-13-35-2T||1.65 V|
|G.Skill||TridentX||DDR3 2400||Dual||2×8 GB||10-12-12-31-2T||1.65 V|
|G.Skill||Ripjaws 4||DDR4 3000||Quad||4×4 GB||15-15-15-35-2T||1.35 V|
Because of the difference between the three platforms, most of the benchmark results you see below will favor the X79 and X99 and their increased CPU cores over the Z87 platform. Single threaded benchmarks will tend to favor the faster CPU core speeds, which is where the X99 and DDR4 platform will fall a bit behind.
The graphs below are based off of percentages with the G.Skill Ripjaws 4 kit being the basis, and therefore, always being 100%. The raw data used to make each graph can be found in the table below each chart. The first set of tests are from AIDA64 Engineer Edition and include the memory read/write/copy/latency benchmarks. The higher percentage is better, except for the latency test, where lower is better. As expected, the Ripjaws 4 kit wins out in the read and write tests, but falls a few percentage points behind the TridentX quad-channel kit in the copy test. Latency testing is pretty much a draw between the two quad-channel kits, but they are both no match for the the dual channel kits as we’d expect.
|AIDA64 Memory Benchmarks – Raw Data|
|Kingston HyperX Beast DDR3-2400 (DC)||35903||37565||34834||44.7|
|G.Skill TridentX DDR3-2400 (DC)||35656||37662||34816||43.6|
|G.Skill TridentX DDR3-2666 (DC)||32966||41505||35597||40.8|
|G.Skill TridentX DDR3-2400 (QC)||56059||41272||55667||65.0|
|G.Skill Ripjaws 4 DDR4-3000 (QC)||58303||56900||53595||64.0|
SuperPi is a single threaded benchmark that normally prefers clock speed over all else, as we can see by the results below. However, wPrime uses all available threads, and the advantage clearly goes to the Ripjaws 4 and the eight core CPU in that system. These results are timed, so the lower percentage values are better in the chart.
|SuperPi & wPrime Benchmarks – Raw Data|
|Kit||SuperPi 1M||SuperPi 32M||wPrime 32M||wPrime 1024M|
|Kingston HyperX Beast DDR3-2400 (DC)||8.955||459.280||5.819||180.694|
|G.Skill TridentX DDR3-2400 (DC)||8.954||466.924||5.851||181.150|
|G.Skill TridentX DDR3-2666 (DC)||8.923||462.790||5.881||182.257|
|G.Skill TridentX DDR3-2400 (QC)||9.260||508.142||4.557||134.715|
|G.Skill Ripjaws 4 DDR4-3000 (QC)||10.359||536.894||3.525||103.647|
7-zip for compression, x264 for video encoding, and Cinebench for rendering are all tested next. As you can see, these benchmarks also take advantage of the higher CPU core count and again show big advantages for the Ripjaws 4 kit because of that. If nothing else, we’ve learned the DDR4 kit doesn’t hold back the power of the i7 5960X processor!
|7zip, x264 & Cinebench Benchmarks – Raw Data|
|Kit||CB R10||CB R11.5||7zip||x264 Pass 1||x264 Pass 2|
|Kingston HyperX Beast DDR3-2400 (DC)||31410||8.84||24853||192.47||50.35|
|G.Skill TridentX DDR3-2400 (DC)||31375||8.85||24346||193.79||50.07|
|G.Skill TridentX DDR3-2666 (DC)||31582||8.90||24951||196.67||50.26|
|G.Skill TridentX DDR3 2400 (QC)||35523||11.23||33358||194.01||60.75|
|G.Skill Ripjaws 4 DDR4-3000 (QC)||43884||15.26||42473||204.95||83.42|
In all reality, these results are an unfair fight in most cases because of the X99/Haswell-E platform advantages. We’re quickly learning about the advantages the X99 platform and DDR4 memory offers.
First up was an attempt at lowering the timings, which provided a respectable result. Raising the voltage to 1.5 V allowed the Cas Latency (CL) to be lowered to 13, which is a nice result. The Ras to Cas Delay (tRCD) and Ras Precharge (tRP) were having no part of being lowered. The Ripjaws 4 have no trouble running at 1T command rate either. The memory was stable 13-15-15-36-1T and passed the HyperPi 32M run about 10 seconds faster than at stock settings.
Overclocking the the MHz side, things get a bit dicey. In order to get any overclocking done at all, the 3rd tier timings had to be manipulated. This comes at the expense of overall performance, but some overclocking can be done if you relax those settings. Even with relaxed 3rd tier timings, don’t expect a massive amount of overclocking… it seems to be the nature of the beast with DDR4 memory. Just as I expect default timings to improve as the DDR4 technology matures, I’m sure better overclocking will too. We had to settle on a +115 MHz overclock, which isn’t too bad for an early adopter kit. The kit didn’t seem to respond to any voltage over 1.5 V and pretty much reached its limit at 3115 MHz. It passed another run of Super Pi 32M here, but you can see it took a pretty good performance hit. It took over 33 seconds longer to complete the Hyper Pi 32M run when compared to just overclocking the timings.
DDR4 Memory overclocking seems to benefit more by working with timings, or at least with the Ripjaws 4 and the ASUS X99 Deluxe we’re using. Having said that, different motherboard and memory manufacturers may work things differently with their offerings. It’ll be interesting when we get around to testing other motherboards and memory kits to see how it all shakes out.
G.Skill did a very nice job with their initial DDR4 offering, and it performs terrific at stock settings. It’s rock solid stable at advertised speed/timings/voltage as witnessed by our stability testing. As far as overclocking goes, lowering the timings provided a substantial performance gain. Raising the MHz actually hurt performance because of the necessity to raise the 3rd tier timings. Again, that could just be motherboard specific and a whole different story on another motherboard. Time will tell. The redesigned heatspreaders do their job nicely and provide a good amount of aesthetic value. Whether you decide on red, black, or blue, this kit will look nice in your system build.
As I’m sure you’ve probably noticed by now, DDR4 memory doesn’t come cheap. The G.Skill Ripjaws 4 DDR4-3000 kit will set you back $329. Sure, that’s expensive, but right now it’s the only 3000 MHz kit available at Newegg. Couple that with Corsair pricing their 2800 MHz kit with lower timings for $379, and I think it’s a fair price.
If you’re going to be an early adopter and jump into the X99/Haswell-E game, the G.Skill Ripjaws 4 DDR4-3000 kit would make excellent memory choice. It’s fast, it’s stable, and it looks darn good too!