Kingston is a well known company for their multitude of storage solutions, consisting of RAM, USB Drives, Solid State Drives, and Flash cards. Entering the market in 1987 with a surface mount memory chip, Kingston is a veteran when it comes to the storage market. Today, we have a quad channel kit of their HyperX Genesis desktop RAM line up ready for a thorough testing.
Here’s a picture of the label with the part number information, and here’s the Kingston HyperX Genesis page.
Looking at the SPD tab in CPUz we can see the rated timings and voltages of this kit and how I manually set the timings; 1067 MHz (2134 MHz DDR) clock speed, 11-12-11-30 timings, and 1.65 V vDIMM. The timings on this kit seems high compared to the typical 9-11-10-28 timings of 2133 MHz 4 GB 1.65 V sticks from other manufacturers. The combination of high voltage and high timings makes me wonder how this kit will do with overclocking, tightening timings, and/or reducing voltage.
However, if I use the XMP profiles to set the speed, timings, and voltage for me, I get the following results. Neither of the XMP profiles match the SPD. I was also told by Kingston that this set was rated at 11-12-11-30, so with all the conflicting info, I don’t know what to trust. However, for this review, I’ll be using the specs in the SPD and that Kingston told me, 2133 MHz 11-12-11-30 1.65 V.
HyperX Genesis Photo Op
I love the packaging that Kingston used with this set of RAM. The sticks just fit into slots in the box and stand up. There are ~10 slots for holding RAM, so this box is definitely a keeper for all the extra RAM I have laying around.
The Genesis HyperX sticks have low-profile heatspreaders that shouldn’t interfere with any heatsink installation. The heatspreaders on this particular set are bright blue with 12 pairs of holes on both sides. Kingston’s HyperX logo is on both sides as well, with one side denoting DDR3 and the other displaying Genesis. Some people gripe about green PCBs used with components, but they will be barely noticeable once installed.
Test Setup & Methodology
|CPU||Intel i7 3930K|
|RAM Kit 1||4×4 GB Kingston HyperX DDR3-2133 11-12-11-30 1.65v|
|RAM Kit 2||4×2 GB G.Skill Eco DDR3-1600 7-8-7-24 1.35v|
|Graphics Card||EVGA GTX 580 Classified 3 GB|
|Solid State Drive||40 GB Corsair Force|
|Power Supply||SeaSonic SS-1000XP|
|Operating System||Windows 7 x64 SP1|
The benchmarks used are as follows and they were run using their default settings.
- MemTest86+ – DOS-based RAM testing software
- LinX – CPU stressing program and the amount of RAM used can be specified
- SuperPi 1M – Single-threaded pi calculation to 1 million digits
- SuperPi 32M – Single-threaded pi calculation to 32 million digits
- wPrime 32M – Multi-threaded square root calculation of 32 million numbers
- wPrime 1024M – Multi-threaded square root calculation of ~ 1 billion numbers
- x264 – Video encoding at 480p
- x264 HD – Video encoding at 720p
- Cinebench R10 – Rendering test
- Cinebench R11.5 – Rendering test
- SisSoft Sandra – Memory bandwidth and latency test
- AIDA64 – Memory bandwidth and latency test
- 7zip – Compression test
- POV Ray – Rendering test
The results of benchmarks between RAM kits are usually very close without one kit running away with tests when everything else is constant. So, the following results are basically the differences between RAM speed and timings when running in quad channel mode on LGA2011.
MemTest86+ & LinX
I used MemTest86+ v4.20 to test the RAM at its rated specifications for over two hours without any errors. That seems to show that maybe the SPD and Kingston’s word was correct and the XMP profiles are just messed up.
However, LinX is a different story. I cannot make a single pass of LinX with the RAM at its rated specs of 2133 MHz 11-12-11-30 1.65 V. When the RAM is set to its rated specs, LinX fails at the exact same time of 23 seconds every time I try running the program at its default settings. At first, I thought it had to be the CPU somehow, so I tried locking the CPU to clocks well below the rated 3.2 GHz and running LinX. The results didn’t change, I would still fail to make a single pass and it would freeze at a certain time (depended on CPU clocks) every time I would run the stress test. Then I tried underclocking both the CPU and RAM by setting the CPU to 1.6 GHz and the RAM to 1333 MHz 9-9-9-24 1.5 V. Then, I actually passed all five runs of LinX with the only thing being changed was the RAM settings. So, I increased the CPU clocks to stock and even up to 4 GHz while keeping the RAM at 1333 MHz 9-9-9-24 and LinX would make it through all 5 passes every time. However, once I increased the RAM to its rated specs, LinX would start failing again.
So, even after these stability tests at stock settings, I’m still unsure if the RAM should actually run at 2133 MHz 11-12-11-30. I didn’t have the best feeling about the RAM after these results, and I thought I would probably see some instability during the other benchmarks as well. However, I stuck with the rated speed and timings according to the SPD and continued testing.
SuperPi 1M/32M & wPrime 32M/1024M
Both SuperPi and wPrime like the additional MHz the Kingston kit has over the G.Skill kit. Although, the difference was only 0.04-1.74%. These number crunching benchmarks show that there’s not a lot of difference between sets for this application, but speedy RAM is a little better for the math. No sign of instability in the HyperX in these tests, so that’s a good sign.
x264 & x264 HD
Interestingly, there seems to be a split decision as far as x264 encoding is concerned. The G.Skill kit beat the Kingston kit in Pass 1 during both the 480p and 720p encoding benchmarks, but Kingston won out on the Pass 2 of both benchmarks. The difference between Pass 1 and Pass 2 is that the first pass is just a scanning pass so that the encoding pass doesn’t have to guess what will be in the next frame based on the current frame. So, the actual encoding pass is a very, very small amount faster with the HyperX kit. Again, no HyperX instability when encoding either.
Cinebench R10 & R11.5
We’re on to rendering with Cinebench, and it looks like this benchmark may prefer the tighter timings of the G.Skill kit over the higher clock speed of the Kingston HyperX. G.Skill won on both counts in the Cinebench rendering tests, albeit not by much. The rendering tests were also free of troubles for the HyperX, it’s starting to look like there’s no issues with the RAM.
SiSoft Sandra & AIDA64
Finally, we see a significant difference between the two RAM kits. Bandwidth is significantly higher with the HyperX RAM, especially in the Memory Bandwidth test. The latencies are very close between the two, and who would can notice 0.2 ns anyways? The HyperX wins out in Sandra for sure!
As you can see, in AID64, the HyperX kit beats the G.Skill by more than a percentage point in all the tests except the writing test. The G.Skill RAM wins only by a hair in the write test, coming out in front by 0.34%. This test goes to the HyperX as well!
These tests were where I thought the HyperX would be most likely to show instability if there was any. The kit made it through these tests multiple times without a hiccup!
7zip & POV Ray 3.7 RC3
Now for a compression test and another rendering tests, and the last chance for the HyperX to to show instability at 2133 MHz 11-12-11-30. In the 7zip compression test, it looks like clock speed has the advantage over timings. For POV Ray, just as with the Cinebench results, the G.Skill comes out ahead; it’s just something about the rendering tests that the G.Skill really likes. Any instability with the HyperX? Nope, both test ran fine.
On the LGA2011 platform, overclocking RAM is a tough task because the RAM multipliers and gear ratios are incremented in rather large jumps. This makes RAM overclocking almost solely dependent on how much the bclk can be fine tuned outside the gear ratio. Just check the table below, and you’ll see what I mean…
|8.00||800 MHz||1000 MHz||1328 MHz||2000 MHz|
|10.66||1067 MHz||1333 MHz||1771 MHz||2667 MHz|
|13.33||1333 MHz||1667 MHz||2213 MHz||3333 MHz|
|16.00||1600 MHz||2000 MHz||2656 MHz||4000 MHz|
|18.66||1867 MHz||2333 MHz||3099 MHz||4667 MHz|
|21.33||2133 MHz||2667 MHz||3541 MHz||5333 MHz|
|24.00||2400 MHz||3000 MHz||3984 MHz||6000 MHz|
|26.66||2667 MHz||3333 MHz||4427 MHz||6667 MHz|
|29.33||2933 MHz||3667 MHz||4869 MHz||7333 MHz|
|32.00||3200 MHz||4000 MHz||5312 MHz||8000 MHz|
As far as increasing the clock speed, it turns out my board/CPU can do 102 MHz bclk, and with the RAM multiplier of 21.33, it turns out to be a 42.66 MHz overclock for a total of 2176 MHz.
I tried tightening the timings down at 2133 MHz and 1.65 V, but didn’t get very far before getting multiple BSODs. I started getting crashes at 10-12-10-29, and anything tighter than this caused a reboot loop after saving and exiting the BIOS. Next, I reduced the speed to 1600 MHz and 1866 MHz then tried tightening down timings at those speeds. The following two pictures show the lowest timings I could get while being able to load Windows and pass MaxxMEM. At 1600 MHz, I was able to tighten down to 7-9-8-24, which isn’t bad at all, but doesn’t quite match the G.Skill Eco at 1600 MHz 7-8-7-24. At 1866 MHz, the timing gladly went to 9-11-10-26, but no further.
The Kingston HyperX Genesis kit has some odd discrepancies between the what the sticks are claimed to be rated for and what the XMP profiles set the RAM at when used. Even the SPD table and XMP profiles do not match. There is also an odd issue with LinX and this kit of RAM when set to its rated specs of 2133 MHz 11-12-11-30. However, after the completion of these benchmarks, it seems to me that the HyperX runs fine at 2133 MHz 11-12-11-30 and LinX may be just an outlier.
The results of the benchmarks have the HyperX coming out ahead in four out of the five types of tests, as follows:
Kingston HyperX Genesis DDR3-2133 11-12-11-30
- Math Computation
- Bandwidth and Latency
G.Skill Eco DDR3-1600 7-8-7-24
As far as overclocking, it’s hard to come to a concrete conclusion because of the limitation of the platform, but from what I was able to test, I wasn’t impressed with the clock speed overclockability, but tightening the timings down wasn’t bad when lowering the clock speed. However, these are running at 2133 MHz already, which is darn fast itself, but the timings can’t be tightened much at 2133 MHz either. I was hoping to get to the 9-11-9-28 range like some other 2133 MHz kits on the market. However, lowering the speed of the RAM allowed the timings to be tightened quite a bit more. As always, overclocking is just a bonus and never a guarantee, so limited overclocking doesn’t go against the kit.
Overall, the Kingston HyperX Genesis kit did well throughout my 24/7 use and testing. The kit just needs a few kinks worked out with the XMP profiles and then there wouldn’t be anything wrong with this kit. I expect the MSRP on this kit to be around $140 since comparable 4×4 GB 2133 MHz 11-11-11-30 kits are around that price point at Newegg. With the only con of this kit being incorrect XMP profiles, which can be easily set manually in the BIOS/UEFI, the kit is still worthy of an approved stamp.
One last thing… Thanks goes out to Kingston for sending Overclockers.com a sample and making this review possible!
– Matt T. Green (MattNo5ss)