HyperX HX428C14SBK2/16 Review

The time has now come for us to perform our usual overclock-testing of the test sample. Even if you are not planning to overclock your memory, knowing its spec headroom and overclocking capabilities can serve as a good assessment of manufacturer's quality control.

As has always been the case, our experiments are going to be done with a test platform based on Intel's latest generation of CPUs, that is commonly known as Skylake.

 CPU  Intel Core i7-6700K (running at 4.5 GHz)
 Motherboard  MSI Z170A Gaming M7 (BIOS version 1.83)
 Memory  Kingston HyperX Savage HX428C14SBK2/16
 Graphics card  ASUS HD4350
 Storage  Kingston SSDNow V300 60GB
 Power supply  Seasonic Platinum 1200W
 Operating System  Windows 7 SP1, 64-bit

There are many ways to test stability of the system but our method of choice is HCI Memtest as it is the toughest memory stresstest that we are aware of. As we are dealing with a 16GB kit on a CPU that can handle 8-threaded load, we use eight 1600MB instances and call things stable if we see all of them run past 150% without showing a single error. Each pass of such a test takes about one hour to complete.

Our test platform has a very wide range of memory clockspeeds to choose from, but to keep things simple we will only concentrate on DDR4-2800, -3000, -3200 and higher modes leaving the intermediate options out as potentially irrelevant. Once we select the memory frequency, we seek for optimal combinations of primary timings, for which we minimise the stable voltage in 0.01V increments.

When overclocking Hynix-based memory modules we usually observe the following behavior. If the CAS latency is kept without change, then the dependency between stable frequency and minimal stable voltage is close to linear. What comes to the rest of primary timings (tRCD, tRP and tRAS): the minimal stable values depend primarily on the memory frequency with voltage playing a secondary role in the borderline areas. However, on Skylake you also have to keep in mind that tRCD must be equal to tRP at all times and that the minimal tRAS value one can set in the BIOS is 28.

Below you can see the list of stable configurations that we were able to achieve with our sample.

1400MHz (DDR4-2800)
1500MHz (DDR4-3000)
1600MHz (DDR4-3200)
 CL11
not possible
not possible
not possible
 CL12
12-14-14-28 @ 1.41V
not possible
not possible
 CL13
13-14-14-28 @ 1.28V
13-15-15-28 @ 1.37V
not possible
 CL14
14-14-14-28 @ 1.27V
14-15-15-28 @ 1.28V
not possible
 CL15
15-14-14-28 @ 1.27V
15-15-15-28 @ 1.25V
not possible
 CL16
16-15-15-28 @ 1.27V
16-16-16-28 @ 1.24V
not possible

Speaking of the spec clearance, we have registered 0.08V of headroom on our sample. That is, running DDR4-2800 with CL14 has required our kit just 1.27V with a Kingston's rating of 1.35V. On top of that, the timings could be lowered from 14-15-15-39 to 14-14-14-28 without any additional voltage increase.

Typically for 8GB modules based on Hynix 4Gbit MFR, a fully stable operation at DDR4-3200 or higher clockspeeds was out of reach. At the end of the day, the best result we got out of our Savage was 1500MHz CL13 with 1.37V. This result is very impressive in its own right, however, there was a major issue that has spoiled our experience.

During our testing, the test setup would frequently fail to POST irrespective of the settings and voltages. At first, we thought that this is a result of compatibility issues between HyperX and MSI but a quick test on an ASUS Maximus VIII Gene platform has generated similar issues. We therefore believe the issues to lie at Kingston's side of things. Whether all Savage kits will have the same or we were just unlucky with our sample, remains an open question.

To see the practical gain of overclocking the memory, we made a quick performance comparison using AIDA64 Cache&Memory test, CineBench R15 and XTU between the XMP and our best overclock. Note that with the latter we have slightly optimised the secondary timings and you can also use our values as a starting reference.

HX428C14SBK2-16_performanceDEF HX428C14SBK2-16_performance1500c13

According to synthetic benchmarks like AIDA64 and XTU, the memory overclock has gained us nearly 10% higher bandwidth and a 5% higher score. However, coming back to benchmarks that are closer to reality, the gain in CineBench rending benchmark was not even one percent.

Leave a comment

 

  1. I like the fact that one can know the IC model/specs with thaiphoon burner on Kingston memory kits without the need to remove the heatsinks.