Retail DDR4 Memory vs. Engineering Samples
Extreme overclocker Michal Vobozil has a look at the retail DDR4 memory and compares it with engineering samples with some surprising results …
The Extreme OC Community [XOC] is a very specific group of PC hardware enthusiasts and power users. It maybe not be the largest one, but overall with so many influential people connected to the industry, hardware review websites and magazines, this a group of people whose opinions are respected and trusted by both normal users and everyday overclockers.
That’s why it’s not surprising that more and more vendors have decided to support our efforts and passion. HyperX, as an official partner of HWBOT PRO Championship, supplied me with a retail DDR4 memory kit which allowed me to prepare a very interesting comparison of retails DDR4 modules versus ES (Engineering Samples) DDR4 modules.
Where did get this idea? Do I really want to state the obvious or should I approach this more like Mythbusters? Well maybe. It’s actually not that big a secret. The OC community started to claim that for a few memory generations (ICs), it’s harder or even impossible to find a decent kit which would be able to match or outperform special binned ES samples that have been reserved for top overclockers or reviewers. After the launch of DDR4 with the Intel X99 platform the situation didn’t change at all, with more memory binners starting to feel unhappy and even somewhat de-motivated by the issue.
There were even opinions out there that a ‘weak’ ES kit would still be much better than the best available retail kit. The new Samsung B-Die made things even worse, with very big differences between retail kits – modules which can barely go above stock specifications, up to crazy clockers breaking Low Clock efficiency challenges.
After what I have seen so far and tested by myself I was really curious to compare and assess the differences. There could also be another factor involved. Could it be possible that some vendors were playing a more “dirty” game than others? So let the battle between AFR begin!
The retail product supplied by HyperX is: 2x HyperX FURY DDR4 (HX426C15FBK4/16 DDR4 – 8 modules).
This is not a typical product review so I am not going to test stock XMP, write down stock specifications, list of compatible motherboards and so on. If someone is interested in the specification details – everything essential (and even more) is available on the HyperX’s website here: http://www.kingston.com/dataSheets/HX426C15FBK4_16.pdf
We only have to keep in mind that stock specifications are 2666 MHz DDR4 CL15-17-17-35 @ 1.2v. We should also remember that HyperX Fury currently is the most affordable line of HyperX memory. Higher up the hierarchy we can find Savage and Predator memories.
HyperX Fury Product Gallery
The retail packaging, in the opinion of some users, resembles the shape of a clamp or shell design. For me it’s a bit more than this – in fact basically a small tray for memory DIMMs, which (after removal of the heatsinks) can hold up to 8 memory modules. For XOC guys like me, having so many sticks and stock trays – these boxes are a true blessing.
The packaging is of course factory sealed.
Full house: 2x HX426C15FBK4/16 & 4xKHX34C16PBK2/8 (ES) – 4 old ES and 4 new ES (from the HOT final contest)
From left: Retail, Old ES, new ES (HOT final)
All modules use a pretty similar 8 layer PCB. Note that the ‘Old’ ES modules are from week 32 of production and the ‘New’ ES are from week 42. Retail modules use the newest batch of AFR from week 43. There were some rumours, that Hynix was going to launch upgraded/fine-tuned AFR, and seems like this might be true. Apparently newer batches tolerate much higher VDIMM than old ones.
Disclaimer: The highest ES bin is 3466 C16, which is around 1% of all AFR. That’s why everyone should be aware that HyperX Fury 2666C15 are on paper at much worse (or even in losing) position– I am comparing the highest bin with one of the lowest so please don’t expect miracles. Also please note that ES HyperX samples are always rated higher than HyperX best retail models, with unique SPD/Serial numbers and protected EEPROM so re-flashing SPD in order to “cheat” in online contests is not an option.
- Intel Core i7 6700K ES with Intel BOX Cooler
- ASUS Maximus VIII Impact
- Galaxy Fish Bone
- Seasonic 1200W Platinum
- HyperX 3K 120GB SSD
Operating System: Chopped and tweaked Windows XP SP3 & Ramdisk. Double run of SuperPI 32 with Copy Waza for stability during the purposes of this comparison.
Ambient temperature: 26-28*C , memory modules cooled by Delta Fan.
Please note: In my experience – a fast and efficient XP SP3 with Ramdisk needs better stability than Windows 7/8/10 X64 for Skylake IMC.
All sub-timings (secondaries and tertiaries) were set manually (tight for air cooling and frequency) and I only changed CAS Latency value (12-13) and memory frequency (3866-4000) for the tests. All the settings used for the test are on the screenshots below:
HyperX Fury OC Results
I decided to present the results in the form of a table so it will be easier to understand and compare. There are 4 categories: 3866C13, 3866C12, 4000C13, 4000C12. Written voltage (e.g. 1.65V) in column means that stick passed the test, “boot” means the stick booted into OS but couldn’t run the test at any voltage (but might be good enough for some other benchmarks like 3D or XTU). N/A value means the stick couldn’t boot it to BIOS or OS at any voltage.
ES 3466C16 means all the sticks in one table. “NEW” means new PCB/revision – HOT FINAL stick. “OLD” is for older batches which appeared just after the Skylake launch.
Retail HyperX Fury 2666C15:
And because it’s a review/preview of HyperX Fury 2666C15… I matched two best sticks in dual channel and clocked them very easily at 3866 12-18-18-28 1T, just like decent ES sticks.
Summary: Retail DDR4 Memory vs. Engineering Samples
It’s not a surprise to see that the 2x best AFR sticks are ES samples from the HOT Final. As I wrote at the beginning – 3466C16 is a very hard bin for AFR and only 1% of ICs or less are capable of passing Kingston / HyperX validation. Also it was obvious to me from the beginning that among the ES there we would find the best stick(s) – that’s why I wanted to show outline the bigger picture.
8 retails vs 8 ES and overall from the “team” battle retails actually seem to be the winners. 7 out of 8 sticks passed 3866C13 settings, 3 out of 8 passed 3866C12. Only 5 ES sticks could run 3866C13 and only 2 (which are the best AFR in my hands) could do 3866C12.
What is interesting is that New ES HOT Final and retails seem to like/tolerate higher voltage which overall makes retail sticks (low bin) look better than ES. Maybe there was some truth in the rumors about better AFR stepping.
Looking at numbers: The 3 best retail sticks are better than the 3rd best ES stick. In such a case with full honesty I can say that in the case of HyperX products users can buy retails and bin them to get pretty decent sticks which can match or even outperform the majority of ES sticks.
There is a chance that in higher retail models, like Savage or Predators, there might be even better modules. In fact I really hope so, but nevertheless that wasn’t the point of this ‘not typical’ comparison. I wanted to test the thesis that with the new generation of RAM memory (DDR4) there are no more ‘good retails’. Apparently according to my testing that is not true. There are retails that are even better than most ES.