HyperX HX321LS11IB2K2/8 Review

We now proceed to our favourite part of each article, which involves testing the full overclocking potential of each memory kit. To perform the testing, we faced a choice between a number of different platforms, but ended up going for the mini-ITX mainboard that comes with ASRock's Z97-M8 barebone.

HX321LS11IBK2-8_09

We first proceed with the HX321LS11IBK2/8 kit. The full details of our test setup are as follows:

Motherboard ASRock Z97-M8 (BIOS version P1.10)
CPU Intel Core i7-4770K (running at 4GHz)
Graphics card ASUS HD4350
Memory Kingston HyperX Impact HX321LS11IBK2/8
SSD Kingston SSDNow V300 60GB
Power supply Seasonic Platinum 1200W
Operating System Windows 7 x64 Service Pack 1

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 an 8 GB kit on a platform that can handle 8-threaded load, we use eight 850 MB 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 20 minutes to complete.

As very few people do base clock (BCLK) overclocking in their daily systems, we are going to fix it at 100MHz and adjust memory frequency only via available multipliers. We will then seek for optimal combinations of primary timings, for which we minimise stable memory voltage in 0.01V steps. Not to get things too complicated, we only set the primary timings, command rate (1T) and memory voltage by hand. The rest of memory settings are left at Auto values.

HX321LS11IBK2-8_results

Overclocking properties of Hynix H5TC4G83MFR-based memory modules can generally be described as follows. If CAS latency is kept without change, then there is a strong, almost linear dependency between stable frequency and minimal stable voltage. This is shown on the chart above.

What comes to second and third entries on the primary timings' list, tRCD and tRP, their minimal stable values depend primarily on the frequency with voltage playing a secondary role in borderline areas. The last of the primary timings, tRAS, does not seem to affect stability or performance by a measurable amount, so tRAS values that we used were sensible values induced by an educated guess. The overall lowest timing picture can be seen on tables below.

800MHz
(DDR3-1600)
900MHz
(DDR3-1800)
933MHz
(DDR3-1866)
1000MHz
(DDR3-2000)
CL7
7-8-7-20 @ 1.52V
7-8-7-20 @ 1.72V
not possible
not possible
CL8
8-8-7-20 @ 1.37V
8-9-8-20 @ 1.50V
8-9-8-20 @ 1.55V
8-9-8-20 @ 1.67V
CL9
9-8-7-20 @ 1.28V
9-9-8-20 @ 1.38V
9-9-8-20 @ 1.41V
9-9-8-20 @ 1.50V
CL10
not relevant
10-9-8-20 @ 1.29V
10-9-8-20 @ 1.32V
10-10-9-20 @ 1.35V
CL11
not relevant
not relevant
11-9-9-20 @ 1.25V
11-10-9-20 @ 1.30V

1066MHz (DDR3-2133)
1100MHz (DDR3-2200)
1200MHz (DDR3-2400)
CL9
9-10-9-20 @ 1.58V
9-10-9-20 @ 1.64V
not possible
CL10
10-10-9-20 @ 1.44V
10-11-9-20 @ 1.48V
10-11-10-20 @ 1.61V
CL11
11-10-9-20 @ 1.35V
11-11-10-20 @ 1.38V
11-11-10-20 @ 1.48V
CL12
12-10-10-20 @ 1.30V
12-11-10-20 @ 1.33V
12-11-10-20 @ 1.47V

As usual, we begin the result discussion by looking at the headroom above the specs. Taking 1.35V as the rated DDR3-2133 CL11 voltage for our sample, there was no voltage headroom as such, although timings could be lowered to 11-10-9 instead of 11-11-11.

Increasing the voltage to 1.60-1.65V we were able to achieve 1100MHz at 9-10-9 and 1200MHz at 10-11-10, neither of which is a shabby set of numbers even for a full-sized DIMM modules. Apparently, because SODIMM modules have to face much tougher operating conditions (in terms of temperature), the chip selection procedure has to be rather strict, which makes the overclockability comparable to desktop modules in ideal conditions.

Irrespective of voltage and timings (including manual secondary and tertiary values) we were not able to stabilise our setup at DDR3-2600 or beyond. At this point, we are not fully certain whether it is the platform or the SODIMM technology that has held us back.

Leave a comment

 

  1. Am I right in presuming the previous mode HX321LS11IBK2/8 had higher performance than the HX321LS11IB2K2/8 model? I was considering getting these chips, however I might as well go with say G.Skill because they achieve tighter timings at the same voltage as the older model here does, though no longer available.

    1. Yes, the older model with better timings is faster. Which one from G.Skill are you talking about? The F3-2133C11D-8GRSL?

      1. Yes that was. I am actually wanting to get a 16GB kit which is this one F3-2133C11D-16GRSL or this one HX321LS11IB2K2/16. I’ve read that however that even the older PnP CL12 out performs current G.Skill and vengeance memory from this forum.

        In the forum, the G.Skill was an 8GB dual channel kit, as opposed to the 8GB kit PnP CL12.

        Are you aware of any comparing reviews of F3-2133C11D-16GRSL vs HX321LS11IB2K2/16 and/or HX321LS11IBK2/16?

        For some reason the older and no longer made impact model goes for twice as much as the newer Kingston model. I want to say the Hynix dies are superior to the Samsung dies for 2133, but I am not sure.

        1. No idea to be honest. Maybe it has to do with secondary and tertiary timings. I can ask the F3-2133C11D-16GRSL and HX321LS11IB2K2/16 for review (and performance testing) but unfortunately I wont be able to get a sample of the HX321LS11IBK2/16 (EOL).

          1. That would be a nice review. I haven’t seen any comparative of the two. Do you perhaps have the 2x4GB kit of the G.Skill Ripjaws (F3-2133C11D-8GRSL)? That versus the HX321LS11IB2K2/8 would determine who is superior at a 16GB kit.

  2. Cpuz-z ignores Fine_tCK correction. For this reason the timing table of each from the modules is absolutely wrong!

    1. I got two HX321LS11IBK2/8 Kits and use them for my P570WM. Unfortunately the use is restricted to 1867Mhz CL10. Everytime I change the settings it reverts back to 1867Mhz CL10.
      I know that Ivy-E’s memory types is restricted to 1867Mhz. On the other Hand there are many P570WM owners who run their Kingston’s at 2133 and even 2400Mhz.

      My question is from what it is dependent to get a stable 24/7 on stock and overclocked?

      Clevo P570WM // GTX 970m SLI // i7 4960X // 16GB (4×4)RAM (HX321LS11IB2K2/8) // Intel 330 180GB SSD + 2 x SanDisk Ultra II 480GB SSD (RAID 0) // 120Hz Matte Screen // 2 x 330w PSU // Win 10 Pro