Armed with a massive 32GB high-end Kingston HyperX Predator DDR4-3000 memory kit, we investigate how DDR4 frequencies and latencies affect real world performance on Intel’s latest mainstream desktop platform.
October 13, 2015 by Lawrence Lee
|Kingston HyperX Predator DDR4-3000 32GB (4x8GB)
Desktop DDR4 Memory
One interesting aspect of Intel’s new Skylake processor microarchitecture is its support for DDR4 memory, making it the first mainstream desktop platform to utilize the new standard. DDR4 was required by Haswell-E but the high cost of LGA2011-v3 hardware meant a limited adoption rate. Being considerably more affordable, LGA1151 should bring about a substantial increase in DDR4 sales, marking the beginning of the end for DDR3. AMD’s next generation of CPUs (slated for late 2016) also supports DDR4 and its release will undoubtedly accelerate DDR3’s descent to obsolescence.
Like previous iterations of DDR memory, the new standard brings significantly higher operating frequencies but this is coupled with high memory timings. For example, a typical stick of DDR3 1600 MHz will operate with 9-9-9-24 timings compared to a stick of DDR4 2666 MHz which may run at 15-15-15-35. These latency figures are the timing delays for specific memory operations measured in nanoseconds, so the higher the timing, the slower the performance. The voltage required to power the DIMMs has also decreased by about 0.3V (from a standard 1.5V to 1.2V). As a result, Skylake’s memory controller can also handle low voltage DDR3L as long as the motherboard manufacturer implements it.
If you’re in the market for a custom Skylake desktop, you may be tempted to buy the most expensive memory available to get the most of out of your new brand new hardware, but that may not be a wise investment. In years past, utilizing higher speed or lower latency RAM generally didn’t offer much of a performance gain. Today we’re going to investigate how much these factors affect a Core i7-6700K powered system in real world applications.
4 x 8GB Kingston HyperX Predator DDR4-3000.
The memory we’ll be using is a massive 32GB (4 x 8GB) kit from Kingston, which is packaged in pairs. HyperX Predator is Kingston’s highest grade of DDR4 memory, with this particular set specified to run at 3000 MHz with 15-16-16 timings at 1.35V. A top-shelf SKU like this is necessary for this kind of testing as it is qualified to operate at both high and low frequencies/timings. As this line of memory was originally marketed for X99 motherboards, Kingston only offers it in sets of 4/8 for quad channel operation but it works just as well in a dual channel Skylake setup. Like all the memory Kingston offers, it’s backed by a lifetime warranty.
Height comparison with a stick of HyperX Fury.
Installed on our test board.
As an enthusiast product, its outfitted with big heatspreaders making it significantly taller than most DIMMs. According to our measurements, they stand 55 mm (31 mm bare) tall and are 8 mm thick. We used a Scythe Kabuto top-down cooler on our test machine and the memory fit underneath with a couple of millimeters to spare.
CPU-Z, SPD information.
Kingston HyperX Predator DDR4-3000 32GB: Specifications
(from the product
|32GB (8GB 1G x 64-Bit x 4 pcs.)
|DDR4-3000 CL15 288-pin DIMM
|XMP Timing Parameters
|•JEDEC: DDR4-2133 CL15-15-15 @1.2V
•XMP Profile #1: DDR4-3000 CL15-16-16 @1.35V
•XMP Profile #2: DDR4-2666 CL14-14-14 @1.35V
|Row Cycle Time
|Refresh to Active/Refresh Command Time
|Row Active Time
|Maximum Operating Power
|94 V – 0
|0°C to +85°C
|55°C to +100°C
*Power will vary depending on the SDRAM used.
Test System Configuration:
- Intel Core i7-6700K
processor – four cores. 4.0~4.2 GHz, 14 nm, 91W, Hyper-threading, integrated Intel HD 530 graphics
- Scythe Kabuto
CPU cooler – stock fan at 800 RPM
- Asus Z170-A motherboard – Z170 chipset, ATX
- Kingston Predator memory– 2/4 x 8GB DDR4-3000
- Zotac GeForce GT 640 ZONE Edition graphics card – 900 MHz core, 2GB 1782 MHz memory, fanless
- ADATA XPG SX910 solid-state drive – 128GB, 2.5-inch, SATA 6 Gbps
SS-400ET power supply – 400W, ATX
Windows 7 operating system – Ultimate, 64-bit
IGP test platform device listing.
Tested Memory Configurations:
- 2 x 8GB at 2133 MHz, 12-12-12-35, dual channel
- 2 x 8GB at 2133 MHz, 15-15-15-35, dual channel
- 2 x 8GB at 3000 MHz, 15-15-15-35, dual channel
- 2 x 8GB at 3000 MHz, 15-15-15-35, single channel
- 4 x 8GB at 3000 MHz, 15-15-15-35, dual channel
CPU/General Benchmark Details:
- Adobe Photoshop CC 2015: Timed image manipulation using a variety of filters, a derivation
of Driver Heaven’s Photoshop
- Windows Defender: Timed virus scan of a folder containing files of varying sizes and types including RAR and ZIP archives.
- 7-Zip: Timed ZIP archive creation using the folder containing our test suite installation files.
- Audacity: Timed audio conversion of an MP3 files to high quality FLAC.
- HandBrake: Timed video encoding of a XVID/MP3 video file to H.264/AAC.
- Cinebench: CPU Rendering benchmark.
- TrueCrypt: Encyption benchmark. 512MB, AES-TwoFish-Serpent algorithm.
Gaming Performance Benchmarks:
- Crysis demo standalone benchmark using Crysis Benchmark Tool.
- Lost Planet 2 standalone benchmark. DirectX 9 mode, test “A”, first stage only.
- Aliens vs. Predator standalone benchmark using Aliens vs. Predator DX11 Benchmark Tool.
- Just Cause 2 demo benchmark.
- Resident Evil 5 standalone benchmark. DirectX 10 mode, variable test, first stage only.
Integrated Graphics: Gaming Performance
For our real world gaming tests, the resolution and detail levels used are the same as those in our Skylake review. We used the highest resolution and image quality settings with which the i7-6700K’s Intel HD Graphics 530 is capable of delivering a reasonably good framerate (about 45 frames per second).
Gaming on integrated graphics is one area where the type of memory makes a remarkable difference in performance as the HD 530 graphics chip has to rely on up to 1GB of system RAM rather than its own onboard memory. 3000 MHz delivers a sizable improvement over 2133 MHz but the difference between C12 and C15 latency is minimal. Running the memory in dual channel is the most important factor of all so if you’re considering one high capacity stick over a pair of lower capacity DIMMs, think again.
Discrete Graphics: Gaming Performance
A fanless Zotac GeForce GT 640 2GB is utilized for our discrete graphics testing. The same games and resolution/quality settings are used as our integrated graphics testing.
As a discrete graphics card has its own dedicated RAM, system memory has almost no effect on the overall results. At most, the difference is a single frame per second and what little difference exists is not consistent between tests. What we’ve charted here is basically the slight variance that goes hand in hand with any type of performance testing.
Integrated Graphics: General Performance
For general performance, a collection of primarily CPU bound applications are used. Some of these benchmarks are new to SPCR but will be part of our new CPU performance test suite moving forward.
Our timed tests using real world applications shows little difference between memory configurations. Virus scanning using Windows Defender and video encoding using HandBrake show the greatest amount of variance with the single channel configuration lagging slightly behind the others in relative terms. The TrueCrypt encryption benchmark produces the exact same results across the board, while the Cinebench rendering scores are practically identical.
We arrive at our overall performance figures by giving each memory configuration a proportional
score with each test having an equal weighting. The scales have been adjusted so that the 2 x 8GB 2133 MHz C12 configuration is the reference point with a score of 100.
When gaming on integrated graphics, dual channel operation has a 32% advantage over single, so a matched pair is absolutely critical, but filling all four DIMMs instead of two is inconsequential. 3000 MHz delivers a ~10% performance bump over 2133 MHz while C12 offers only a 1% boost over C15.
General performance is so close, an extra decimal point on the chart is needed for a more accurate comparison. The general convention that higher frequency and lower latency is reinforced but the margins of difference are fractions of a percent.
Though the testing for this investigation was brutally tedious, the results were satisfyingly conclusive. On Skylake systems, the type of memory used is only important if you lean heavily on the integrated graphics chip. If this is the case, opting for a single stick rather than a pair of DIMMs in dual channel is a critical mistake that can cost more than 30% performance. Frequency is relevant as well with 3000 MHz offering around 10% improvement over 2133 MHz. The difference between C12 and C15 is negligible and should be ignored unless the price differential is minimal.
On the otherhand, if you only use the integrated graphics for pedestrian uses like video playback or if a discrete video card is to be utilized, the speed and latency of the system memory is inconsequential. Our general performance tests results depict the highest performing settings as having a one third of one percent advantage overall. High speed kits are marketed towards enthusiasts but even the most hardcore overclockers can achieve their desired goal without high frequency RAM. Skylake “K” variant chips have multiplier based overclocking and standard Skylake processors can be clocked up by increasing the BCLK frequency and using low memory dividers.
Our thanks to Kingston for the HyperX Predator DDR4-3000 for the DDR4 memory kit used in this article.
* * *
Articles of Related Interest
Asus Skylake Z170 Motherboards: Maximus VIII Gene vs. Z170-A
Intel Core i7-5775C: Broadwell for Desktops
Skylake: Intel Core i7-6700K
Gigabyte X99-UD4P Haswell-E Motherboard
Asus Maximus VII Gene vs. Gigabyte Z97MX-Gaming 5
Asus Z97-PRO LGA1150 Motherboard
* * *