Table of Contents
The premium DDR4 memory market just became a whole lot more interesting! A new kit of memory from Corsair has landed in our hands, and we are going to take a good look at this memory to see what makes it noteworthy. Corsair has added a new member to their flagship Dominator line. Today, we are looking at the Corsair Dominator Platinum RGB. This new Dominator module, which was first introduced back in 2010, has been redesigned from the ground up to achieve the highest Intel XMP standards for modern DDR4 memory.
Specifications and Features
Corsair launched the Dominator Platinum RGB series with an impressive twenty different memory configurations to choose from. The individual stick density comes in the form of 8 GB or 16 GB, with a wide range of kit options available. Whether you need the massive and speedy 128GB 3600c18 kit @ $1549.99, or the humble 16GB 3000c15 kit @ $164.99, they have your modern DDR4 needs covered. While it is not currently available on any retail websites, Corsair spotlights a truly remarkable 16GB 4800c18 kit @ $864.99. With a price tag of over $800 for a 16 GB kit of DDR4, Corsair has broken all barriers of the premium memory market.
No matter what the budget or the requirement is, all of the memory in this lineup comes with the same awe-inspiring build quality and truly unique RGB lighting. Corsair launched this new line with an industry leading, 10-layer PCB. Mounted to the new custom PCB are hand-selected ICs. This extra step of hand selecting memory ICs is to ensure that each stick is perfectly matched for better overall compatibility. To keep the memory running cool, Corsair implemented their patented DHX cooling technology, which utilizes both sides of the memory heat spreaders to cool the ICs.
To outshine the competition, Corsair has broken the norm of conventional RGB design and started from the ground up with a brand new type of RGB LED. Each module of the new Dominator Platinum RGB has:
12 ultra-bright individually addressable CAPELLIX RGB LEDs line the top of the module, shining brighter, lasting longer, and consuming less power than conventional LEDs. –Corsair
The new Capellix LED are substantially smaller and brighter than typical RGB LEDs commonly used on DDR4 modules. The reduced size affords Corsair the opportunity to pack more RGB LEDs on to each module while still retaining the ultra bight look that the consumers demand.
In the table below we examine the particular details of the memory being evaluated today.
|CORSAIR DOMINATOR PLATINUM RGB|
|Capacity||32 GB (4×8 GB)|
|Speed Spec||PC4 25600|
|Rated Frequency||DDR4 1600MHz (3200MHz Effective)|
|Kit Type||Quad Channel|
When purchasing DDR4 memory the main factors to consider, other than memory size or physical features, are the operating frequency and timings. XMP is a memory profile stored inside the actual memory, which allows the user to easily apply the rated frequency and timings. This kit has an XMP certified rating of 1600MHz (3200MHz Effective) with primary timings of CL14-14-14-34. The voltage for this XMP rating is a scant 1.35v.
Below is a screenshot of Thaiphoon Burner, which is a wonderful free tool that allows one to read the Serial Presence Detect (SPD) firmware of the DRAM. The SPD information is critical in determining how the stick will perform and how the computer will recognize it.
As the Thaiphoon Burner screenshot shows, this specific kit of memory is composed of Samsung B-DIE ICs. This memory uses a total of 8 ICs, all of which are located on only one side. As you might know, Samsung B-DIE has become synonymous with high frequency and tight timings. Overclockers, gamers, and system builders alike all flock to memory composed of Samsung B-DIE ICs. They have become the gold standard to which all other memory ICs are compared. Initially, the Samsung B-DIE ICs were sought after for their ability to operate at low primary timings, such as CL14-14-14. However, with the advancement of memory PCB’s and as the IC technology matures the industry has seen gradually increasing frequencies to go along with low primary timings.
Corsair has changed the marketplace standard for high-end memory. Simply creating beautiful memory is not good enough. Consumers paying premium prices expect the entire experience to be premium as well. Corsair does not disappoint with the packaging. The memory comes in a full-color box with embossed lettering and a glossy image of the memory that hides beneath.
Opening the box we find the memory safely nestled in individual plastic clam-shell packaging. The individual sticks are tucked away nicely in high-density foam. All of this creates the high-end feel that we have come to expect from the Corsair Dominator line of memory.
The making of trend-setting, premium memory comes down to perfecting each individual piece and then assembling them in a meaningful way. Starting back with the launch of the Dominator series in 2010, Corsair set the quality standard for premium memory. They have now simply raised the bar even higher with the new Dominator Platinum RGB. Once out of its packaging, the overall feel in the hand is quite impressive. The final fit and finish of this memory are simply unparalleled, as everything fits together perfectly. Even to a casual observer, it’s clear that Corsair went to great lengths to make this memory special.
Our patented Dual-Path DHX cooling technology cools the memory through both the PCB and the external housing – allowing the module to stay cool even under extreme stress. –Corsair
To start off, the side plates are forged aluminum with matte black anodization. The color is subtle and will easily match most of the existing computer components on the market today. The top bar is a zinc alloy with a matte black finish which matches the color and tone of the side plates, but it differs ever so slightly by being a bit less matte. The module has aluminum heat sink fins which sit on top and offer additional cooling support. While they no doubt offer exceptional cooling, one potential downside is that the modules are very tall and might conflict with the larger air coolers on the market.
The memory is only offered in one color option, but we don’t feel this hinders the product line in any way as this memory is all about the RGB lighting. The all-black coloring blends in with the other system components and lets the RGB stand out.
When it comes to lighting, the Corsair Dominator Platinum RGB offers a very unique light show. Apart from most other RGB memory on the market today, the majority of the stick’s lighting has been covered by the top bar. The zinc-alloy top bar, in conjunction with the built-in plastic light diffuser, focuses the light into 10 individually, addressable light squares. In addition, the center of the stick has two addressable LEDs which light up the namesake, Dominator.
The use of the all new Capellix LED allows for an unprecedented 12 uniquely-programmable LEDs. The LEDs are substantially smaller which allow Corsair to include 12 of them on top of the module. Not only are the LEDs brighter and smaller than conventional ones, but they are in fact more efficient, resulting in less power draw and ultimately longer life.
To accompany and control the memory, Corsair has redesigned and merged its software into one cohesive and all-encompassing suite. Corsair Utility Engine, or iCUE for short, can now control all of your Corsair devices such as the keyboard, mouse pad, mouse, headset, headset stand, memory, AIO cooler, and power supply. It can also integrate into other Corsair controllers, such as the Commander PRO or Lighting Mode Pro. The software has many different powerful tools such as the ability to create macros and update the firmware.
In terms of memory control, Corsair has packed a truly astonishing amount of lighting options for the Dominator Platinum RGB. There are 17 predefined lighting options that all offer a different light show. Stepping outside of the predefined light patterns, the user has endless possibilities of what can be accomplished. The custom fields are so flexible that we will jokingly say that to program a game of Tetris is not completely out of the question.
Corsair also added temperature controlled LED control to the iCUE software. The module can be programmed to display a particular color based on a specific temperature, or range of temperatures. While Samsung based memory modules don’t often heat up very much, the temperature range can be tuned so that the color changes based on small temperature increases.
Due to the fact that the light is restricted by the top bar, very unique light shows can be programmed and displayed with this memory. The predefined light show called STACK was particularly interesting. Never before have we seen memory perform acrobatics like this. In the video below you can see the Stack light show running with the speed on ‘slow’ and the color set to ‘shift’. All of these options can be modified from the predefined settings.
Testing and Overclocking
The overall objective is to evaluate the memory from the perspective of overclocking for daily usage. Most of the overclocking settings we will attempt are designed to increase productivity and overall system performance. Additionally, we will briefly examine how Ryzen Threadripper responds to memory overclocking and examine what types of overclocks can be carried out with reasonable voltage levels. To accomplish this task, we will turn to benchmark programs to examine the performance of the memory and overall system under various conditions. The approach is to first test the Intel XMP profile. Once we have established that the XMP profiles are working on the test system, then the real fun begins as we evaluate the memory from an overclocking perspective.
There are several ways of going about this, however, our primary focus will be the overclocking potential without excessive voltage. According to the XMP 2.0 certifications, the absolute maximum allowable voltage is 1.50 V VDDR. Thus, all overclocking endeavors will be conducted with less than 1.50 V.
Below are the test system and resulting memory speeds that will be used to evaluate the memory and run the benchmarks.
|CPU||RYZEN Threadripper 2990WX Overclocked to 4.0 GHz|
|Cooler||Alphacool Eisblock XPX CPU Block with custom water loop|
|Motherboard||ASRock X399M TAICHI sTR4|
|Graphics Card||ASRock Phantom Gaming X Radeon RX 580|
|Solid State Drive||Team Group L5 LITE 3D SSD|
|Power Supply||Seasonic 1200W Platinum PRIME|
|Operating System||Windows 10 x64|
|Memory Speeds Compared|
|Intel XMP ~ 3200, 14-14-14-34 ~ 1.35 V|
|Test Case 1 ~ 3200, 14-13-13-24 ~ 1.40 V Tight Sub Timings|
|Test Case 2 ~ 3600, 14-14-14-34 ~ 1.45 V|
|Test Case 3 ~ 3600, 14-13-13-24 ~ 1.50 V Tight Sub Timings|
As is the case with all overclocking adventures, your results may vary, so proceed only if you assume all risk. To view and examine all of the various memory profiles we use two primary tools which include AIDA64 and the Ryzen Timing Checker. AIDA64 is a powerful system diagnostic and benchmarking tool that can be purchased for a reasonable price. Next, we will be using the Ryzen Timing Checker, which is a free piece of software that allows users to see all of the major timings associated with DDR4 which have been applied in the bios.
Below is the XMP profile. This particular kit of memory comes with an XMP profile of 3200, CL14-14-14-34.
The improvements in XMP profile speeds is greatly attributed to modern manufacturing processes and memory PCB layouts; however, it might not be attainable on all motherboards. The profile is intended to be a one-click overclock, but it likely only applies to enthusiast-grade motherboards. If you are building a computer based on a budget motherboard, don’t be surprised if you cannot achieve stability with this XMP.
Memory overclocking is all about pushing the frequency higher and lowering the timings. One hurdle for memory overclocking is the motherboard. The distance between the CPU and the memory modules has a direct relationship with the overclocking potential of the memory. Motherboards with a larger distance between the CPU and the memory will have greatly-reduced overclocking potential compared to motherboards with a shorter distance. Therefore, motherboards with only two dual, in-line memory modules (DIMMs) will have a better likelihood of increased memory overclocking potential than motherboards with four DIMMs, because the distance is inherently shorter.
The motherboard being used is the ASRock X399M Taichi. This is a very reasonably-priced motherboard which has elite-level memory overclocking capabilities due to it only having four memory slots. Within the X399 chipset, this is the only motherboard that offers four DIMM slots, compared to the standard 8 DIMM slots for all other motherboards for this socket. Similar to how two memory slots are better than four for overclocking, four memory slots are better than eight when it comes to quad channel memory. We are starting with the best case scenario for overclocking memory on this chipset.
Once the XMP profile has been successfully tested, we can dive into overclocking. The methodology is to set our maximum working voltage of 1.490 V, to see what could be accomplished, and then lower the voltage to find the stability point.
For the first test case, only the memory timings were edited with no other settings adjusted. With a scant 5 mV from 1.35 V to 1.40 V increase in voltage, the primary timings were reduced from 14-14-14 to 14-13-13. This didn’t seem like enough of a change to make a huge difference, so the sub-timings were substantially reduced to provide a better test case for overclocking. As we will see later in the results, the benchmark results are heavily influenced by all of the memory timings, especially sub-timings.
For the second overclock test case, we used a small voltage increase of 1.35 V to 1.45 V. Contrary to the first test case, this time only the operating frequency was increased and no other settings were changed. The operating frequency was increased from 3200 MHz to 3600 MHz.
For the third, and final overclocking test case, the goal was to achieve the best overclock possible while still staying within the maximum allowable voltage of 1.50 V according to XMP 2.0 specifications. Finding the optimal overclock for given criteria like voltage is not an exact science and requires a high degree of knowledge and patience, as well as the right combination of equipment. However, the results can be quite impressive. The resulting overclock was a 400 MHz increase in frequency from XMP, with tighter timings (decreased) as well. Given the test equipment and the voltage constraint, this is an admirable overclock.
Benchmark programs allow us to examine the performance gain or loss from making changes to the various computer components such as the CPU, GPU, or memory. In this case, the goal is to remove all variables and examine only the performance change from memory overclocks. To accomplish this, all benchmarks were run with the CPU held at a constant 4 GHz for every benchmark. We will investigate the overclock potential not just because it’s fun, but because it has real-world implications and can help with the productivity of daily tasks.
First up, we used AIDA64 Cache and Memory Benchmark. Specifically speaking,
Memory bandwidth benchmarks (Memory Read, Memory Write, Memory Copy) measure the maximum achievable memory data transfer bandwidth. The code behind these benchmark methods are written in Assembly and they are extremely optimized for every popular AMD, Intel, and VIA processor core variants by utilizing the appropriate x86/x64, x87, MMX, MMX+, 3DNow!, SSE, SSE2, SSE4.1, AVX, AVX2, and AVX-512 instruction set extension. –AIDA64
In the chart below, we compare the four different memory speeds and how they affect the overall system performance. As the frequency goes up, or the latency becomes faster, the expected progression is linear. However, a few interesting results developed that fell outside of the expectation. For the read benchmark, altering the timings had an adverse effect on the overall performance. The adverse effect is obvious with test case 1 vs XMP, which shows that tightening the timings actually made the performance worse. In the case of write performance, the opposite was true. Altering the timings showed a substantial increase in performance.
We summarize that increasing the maximum frequency had the best overall effect on performance. Changing the primary and sub-timings, can dramatically increase the performance for some benchmarks, but also hinder the performance for others. Therefore, to achieve the best average overclock for daily performance, we recommend sticking to increased frequency only and leaving the timings on auto according to the XMP profile.
Next up is Geekbench 4, and it has proven itself to be an excellent tool for determining the real-world performance of the system being tested. This type of benchmark is purely 2D calculation-based, and there is no graphical processing element, so it’s a great analytical tool to evaluate memory performance. Specifically speaking, it:
Includes updated CPU workloads and new Compute workloads that model real-world tasks and applications. Geekbench is a benchmark that reflects what actual users face on their mobile devices and personal computers. –Geekbench
We chose to highlight only the sub-tests of Geekbench 4 Multi-Core and ignore the overall score. Looking only at the sub-test affords us a unique opportunity to observe the changes in overclocking on a microscopic level. Observing the overall score will show performance changes of just 1% to 3%, but if we examine the sub-tests then we get an obvious picture of how the overclocking affects system performance.
In the chart below, it can be observed that the overclock test cases show a noticeable change in performance for the various memory overclocking test cases. Each of the test cases showed an increase in performance over the XMP. It’s interesting to point out that for the Crypto score, altering the timings had the biggest effect compared to increasing the frequency. You might recall from the AIDA64 tests above that we observed the opposite trend for “read” performance. For all of the overclocking tests however, the highest score was achieved with the combined overclocks, test case 3.
Next, we will examine the performance using a few of the memory benchmark tests offered in the SiSoftware Sandra suite of benchmarks. The flagship product, known as Sandra, is a powerful suite of many different benchmarks to evaluate computer performance for all major components including CPU, Video/Graphics, Memory/Cache and Disk.
SiSoftware, founded in 1995, is one of the leading providers of computer analysis, diagnostic and benchmarking software….Since launch, SiSoftware has always been at the forefront of the technology arena, being among the first providers of benchmarks that show the power of emerging new technologies such as multi-core, GPGPU, OpenCL, OpenGL, DirectCompute, x64, ARM, NUMA, SMT (Hyper-Threading), SMP (multi-threading), AVX512, AVX2, AVX, FMA3, NEON, SSE4.2, SSE4, SSSE3, SSE2, SSE, Java and .NET. –SiSoftware
Today, we will be looking at only two of the memory tests within Sandra, Memory Bandwidth and Transactional Throughput. As you can see in the chart below, these tests offer a good representation of the performance increase or decrease that can be achieved from overclocking. For the Transactional Throughput test, we saw a mild increase of decreased timings, a noteworthy gain from frequency overclocking, but then an unexpected loss of performance when we combined the two together in test case 3.
In the Memory Bandwidth tests, we observed an astounding 29% gain from test case 3. The chart clearly shows that all of the overclocking test cases showed noteworthy improvements for the bandwidth.
The next benchmarks we will examine are those centered around 3D rendering and games. The UL Benchmark suite of benchmarks is a real-time graphical rendering benchmark which also contains an element of memory and CPU testing. For each of the benchmarks from Fire Strike to Time Spy, we will only examine the CPU/memory testing portion and disregard the graphical test elements.
In the chart below, we observe that the performance increase from memory overclocking is actually quite large. The Time Spy tests showed us marginal gains across the board with a total of 4% increase in performance for the combined frequency and timing overclock of test case 3.
The Fire Strike tests showed an unprecedented gain of performance from memory overclocking. We observed that decreased timings of test case 1 showed a bigger yield in performance compared to simply increased frequency of test case 2. Putting the two together in test case 3 gave us a 25% gain in performance beyond the XMP baseline.
Our overall impression of the Corsair Dominator Platinum RGB is positive. From the matte black anodized aluminum heat spreaders to the unique RGB LED top bar design, we can only conclude that this memory has an unparalleled premium feel and look. With a best-in-class speed rating of 3200 MHz at CL14-14-14 running in quad channel, this memory has the out-of-box performance capabilities to back up its superior style.
Being composed of Samsung B-DIE memory ICs, the memory was designed with overclocking in mind, and it certainly didn’t disappoint on that front either. Threadripper proved an excellent platform to show off the quad-channel overclocking prowess of the new Corsair Dominator Platinum RGB. Based on other Threadripper 2 findings, daily clocks of 3600 CL14-13-13 while being fully stable in quad channel, is proof that this memory deserves an overclocking moniker.
The Corsair Dominator Platinum RGB seemingly has it all. Branded as premium memory, it shouldn’t come as a surprise that price is among the highest in class. With a price tag of $519.99, Corsair is making a statement with the Corsair Dominator Platinum RGB being the most expensive in its class. Comparable RGB memory running at the same speed, timings, and voltage would include the G.Skill Trident Z RGB, which is on the market for $444.99. We give corsair credit for sticking their neck out and branding the memory premium. When it comes down to hard performance, though, several other kits of B-DIE based, 3200c14 memory, offer comparable overclocking performance for considerably less cost.
David Miller – mllrkllr88