Test Setup and Benchmarks
|Processor||Intel Core i7-6700K (ES)|
|CPU Cooler||Noctua NH-U12S|
|Power Supply||Corsair HX850W 80 Plus Gold PSU|
|Memory||*see table below|
|Storage||Silicon Power S80 240GB SSD (OS), OCZ Vertex 4 256GB SSD (Test)|
|Graphics||Integrated CPU Graphics, Gigabyte GTx 960 Windforce 3x OC|
|Drivers||Intel 10.1.1.7 (Z170), Intel 10.0.2.1 (Z97), NVIDIA 353.62 WHQL (Discrete)|
|Motherboard(s)||Gigabyte Z170X-Designaire (BIOS F2)|
|Operating System||Windows 10 Pro|
DDR4 Memory Module Kits tested (Default):
|Manufacturer||Name||Memory IC||Capacity (per module)||Clock Speed (XMP)||Timings (XMP)|
|ADATA||XPG Dazzle||dual-sided SK Hynix H5AN4G8NMFR||8GB||2800||17-17-17-36 2T 1.25V|
|HyperX (Kingston)||Fury||Dual-sided SK Hynix H5AN4G8NMFR||8GB||2666||15-17-17-35 2T 1.2V|
|Patriot||Viper Elite||Single-sided Samsung K4A8G085WB||8GB||3200||16-16-16-36 2T 1.35V|
- The RAM modules were clocked to match for apples-to-apples direct comparison in 2800MHz as per the ADATA Dazzle module kit’s XMP profile.
- CPU multiplier clock locked at 40, BCLK locked at 100 and cache frequency locked at 4GHz. Turbo boost disabled. The goal here to is to observe behaviour and get consistent result, not to shoot for high benchmark numbers for the sake of shooting for higher benchmark numbers.
- Latest working BIOS and drivers were used at the time of the review. Each test was conducted at least three times for accuracy.
- Each kits were run on HCIMemtest and AIDA64 stability prior to testing to ensure that they are in working order.
- AIDA64 Engineer Edition 5.50.3600
- Passmark Performance Test 8.0 (build 1051)
- SiSoft Sandra Lite 2015 SP3 (21.47)
- RoG RealBench 2.41
- 7-Zip 9.20
Not surprisingly, there is not much room here for overclocking maxing out at 3000MHz CL18 with 1T command rate (HyperPI stable), although this Gigabyte motherboard needs some work on the BIOS as it is not able to tighten the timings as well as with my other kits compared to other motherboards with more mature firmware, then again, timings are not that important on DDR4. You are going to see higher copy speeds in AIDA64 but read/write is typically within less than 1% improvement with tighter CL timings from 18 to 14 for example. The frequency however, I am confident is pushed as far as it could with 1.3V, additional voltage did not help even with VCCSA and VCCIO tweaking. I’ll try it out with a different motherboard and will update this if my results change.
AIDA64 is a very versatile program that we use here at Modders-Inc for our CPU cooler, storage as well as CPU reviews but it also has a built-in synthetic memory benchmark that measures the Copy, Read, and Write speeds of the DRAM installed in Megabytes-per-second (higher is better). It also measures total latency in nano-second (lower is better).
Like AIDA64, Passmark’s Performance Test 8.0 suite is also a very versatile program with a different set of benchmarks per component and has a suite of memory tests of its own. The read, read uncached and write speeds are used (measured in synthetic scores, higher is better) as well as the built-in memory latency test in nano-seconds (lower is better).
Sandra LITE is a free version of SiSoft’s Sandra testing suite. Similar to AIDA64 and PassMark Performance Test, Sandra Lite is a synthetic benchmark program that has its own set of system memory specific benchmarks. Aggregated scores of the Memory bandwidth performance in Gigabyte-per-second and Memory Transactional Throughput in total throughput-per-second are measured (higher is better). Memory latency is also measured in nano-second (lower is better).
ASUS Republic of Gamers’ RealBench is a semi-synthetic testing suite that collects four different programs and can perform controlled runs of each related to image editing (GIMP), video encoding (Handbrake H.264), OpenCL (Luxmark) and Heavy Multitasking (Combination of all three at the same time). Only the image editing part is used as GIMP uses up to SSE4.2 CPU extensions and focuses on single threaded CPU and memory performance that is frequency and timing sensitive. The controlled aspect of this otherwise “real-world” test makes it semi-synthetic.
For compression, 7-zip is used which is an open source (GNU) compression program utilizing LZMA method as the default. This is a real-world program with a built-in benchmark measuring compression, decompression and total speed in million instruction per second (MIPS).
Futuremark’s 3DMark is a semi-synthetic gaming benchmark that calculates both graphics and CPU-bound physics in a controlled series of tests and provides scores that can be compared with other gaming platforms. The physics benchmark of the Skydiver suite for mainstream gaming PCs is used.
The ADATA XPG Dazzle demands attention with its fancy lighting and it certainly deserves it as it is executed well in terms of looks. There have been modules before with LEDs embedded but the ADATA XPG Dazzle’s LEDs have a diffused lighting which results in a much more evenly distributed light rather than the pointed LEDs dotting the PCB area. This diffuser adds more height to the modules but it is certainly worth it for the effect. It is a small attention to detail that is much more effective than other LED lighting implementations on computer memory.
This light also behaves in a sort of alternating breathing pattern so it is not static and is much more organic looking on a running system. A much welcome extra feature would be the ability to adjust this behaviour but that would obviously drive up the cost further. The current MSRP for this kit is only $109.99 so it is only $10 more than a typical DDR4 kit of similar speed; very much within reason and is still very competitively priced. In terms of performance, the XPG Dazzle is a little tight on the overclocking headroom but performance is more for its XPG Z1 sibling which is offered at a greater variety of options and higher speeds. The XPG Dazzle DDR4 is more for the creative types that want memory modules that match their personality.